<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Fernando Mourao on Medium]]></title>
        <description><![CDATA[Stories by Fernando Mourao on Medium]]></description>
        <link>https://medium.com/@fmourao?source=rss-b8955024cf26------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*XUBP_-MjTcV5Cg29FASFBg.jpeg</url>
            <title>Stories by Fernando Mourao on Medium</title>
            <link>https://medium.com/@fmourao?source=rss-b8955024cf26------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 07:31:36 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@fmourao/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Competitive Capability Sovereignty: The AI Risk Nobody Is Measuring]]></title>
            <link>https://medium.com/seek-blog/competitive-capability-sovereignty-the-ai-risk-nobody-is-measuring-437c39ee8d3c?source=rss-b8955024cf26------2</link>
            <guid isPermaLink="false">https://medium.com/p/437c39ee8d3c</guid>
            <category><![CDATA[ai-sovereignty]]></category>
            <category><![CDATA[ai-governance]]></category>
            <category><![CDATA[responsible-ai]]></category>
            <dc:creator><![CDATA[Fernando Mourao]]></dc:creator>
            <pubDate>Sun, 04 Oct 2026 23:58:04 GMT</pubDate>
            <atom:updated>2026-10-04T23:58:04.040Z</atom:updated>
            <content:encoded><![CDATA[<p><em>In this post, </em><a href="https://www.linkedin.com/in/fernando-mour%C3%A3o-a40a5183/"><em>Fernando Mourão</em></a><em>, Head of Responsible AI at </em><a href="https://www.seek.com.au/work-for-seek/"><em>SEEK</em></a><em>, names a risk hiding in plain sight: while nations fight over AI sovereignty at the geopolitical level, organisations are quietly surrendering the same ground at the operational level, and calling it efficiency. The following text exemplifies deliberate AI augmentation, in which Claude Sonnet 5 refined the author’s original insights, creating a more polished narrative while preserving the uniquely human strategic thinking at its core.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*C-F53R_oCZlZveRQRFdpYg.png" /><figcaption>AI‑generated image created with Microsoft Copilot (human‑prompted).</figcaption></figure><p>Everyone is talking about national AI sovereignty. Governments are pouring billions into domestic AI infrastructure, restricting cross-border data flows, and negotiating hard with the hyperscalers. The concern is legitimate, and the stakes are real: nations that allow critical infrastructure, cognitive labour, and technological capability to be captured by foreign platforms progressively lose the capacity to direct their own future. The GDP doesn’t just leave. The skills, the institutional learning, the reinvestment cycles never form locally in the first place.</p><p>It is a smart and necessary debate.</p><p>But while boardrooms watch governments argue about sovereignty, a quieter and more intimate version of the same dynamic is unfolding inside their own organisations, and almost nobody is measuring it.</p><h3>The Parallel Worth Drawing</h3><p>National AI sovereignty is fundamentally about two things: dependency risk and adaptive capacity. The fear isn’t today’s capability gap. It’s tomorrow’s inability to close it — the progressive loss of the conditions required to course-correct, compete, and remain the author of your own strategic future.</p><p>Strip away the geopolitical frame, and that description fits most large organisations deploying AI nowadays with uncomfortable precision.</p><p>Organisations are integrating external AI systems at extraordinary speed and depth. Vendor platforms are mediating customer experience, shaping internal decisions, automating workflows, and increasingly framing the problems that leaders are asked to solve. The economics are compelling. The capability uplift is real. And the dependency is deepening silently, structurally, and critically, without the governance frameworks that would make its long-term consequences visible.</p><blockquote>This is the risk of Competitive Capability Sovereignty: the ongoing capacity of an organisation to direct, evolve, and reinvent its competitive capabilities — even while deeply dependent on external AI systems.</blockquote><p>It is not a technical risk. It does not show up in incident reports or risk registers. It compounds invisibly on a timescale that sits outside the quarterly cycle most organisations use to evaluate AI investments. And it may be the most strategically consequential AI risk that only a few boards are discussing.</p><h3>The First Erosion Force: The Efficiency Trap</h3><p>The most immediate and underestimated risk isn’t technical. It’s human, and it’s already happening inside every organisation deploying AI at scale.</p><p>When AI is irresponsibly adopted, your organisation quietly <em>reorganises around the tools.</em> Not through a strategic decision. Not through a policy change. Gradually, invisibly, structurally. People stop making decisions and start approving them. And something subtler and more damaging follows: rather than AI elevating human thinking, your people begin replicating the very shallow thinking that AI was supposed to replace. The habit of active judgment, synthesise, and think originally - all of them atrophy precisely because they’re no longer being exercised.</p><p>There’s a reason nobody raises the alarm. It doesn’t feel like a loss of control. It feels like efficiency. And that is what makes it so dangerous.</p><h3>The Second Erosion Force: Workflow Intelligence</h3><p>The second force operates at the organisational level, and it has two edges — one internal, one external — that most governance frameworks haven’t begun to address.</p><p>The internal edge is evident. The more you automate your workflows, the less you see them. Over time, AI systems become deeply integrated into how your business operates — seamless, fast, increasingly invisible. What looked like automation becomes infrastructure. And infrastructure, by definition, is what you stop thinking about. The institutional knowledge of how your business actually works, but the tacit, living understanding that once sat in your people, quietly migrates into systems that run the business but no longer explain it to the humans inside it.</p><p>The external edge is less understood but equally consequential. Your vendor has almost certainly assured you they don’t train on your data. For raw data, like documents, records, customer information, that assurance is largely credible. But there is something else being learned from, something no enterprise contract currently governs: <em>how</em> you use AI. The problems you bring to it. The way your business is wired into it. The decisions it shapes every day. The edge cases only your organisation surfaces. That is not “data” in any legal definition your contracts cover. But it encodes something potentially more valuable: <strong>how sophisticated organisations deploy AI at scale, where the real capability frontiers are, and what the highest-value unsolved problems in your industry look like from the inside.</strong></p><h3>The Third Erosion Force: The Competence Illusion</h3><p>The third force operates at the deepest level, and is the hardest to see precisely because it disguises itself as capability.</p><p>Where the Efficiency Trap restructures how your organisation <em>behaves</em>, the Competence Illusion changes what your organisation <em>thinks it can do</em>. The mechanism is psychological as much as organisational. AI systems are designed to be fluent, confident, and convincing. The more effectively your people use them, the more capable they feel, and paradoxically, the less they question what they might be missing. This is the Dunning-Kruger effect, systematically amplified. People get sharper on their <em>known unknowns</em> while their <em>unknown unknowns</em> quietly expand. The blind spots don’t disappear. They go deeper.</p><p>The danger is that this doesn’t look like a risk. It looks like deepened expertise. Output is faster, more polished, more voluminous. The near-term story is productivity. The long-term story is atrophy of the <em>depth and quality</em> of the judgment that remains. The progressive erosion of the instinct to challenge, verify, and seek perspectives that sit outside the model’s frame.</p><p>This is where the three erosion forces converge into their most dangerous configuration. The Efficiency Trap has eroded the organisational habit of deciding. Workflow Intelligence erosion has reduced visibility into how the business actually operates. And the Competence Illusion has simultaneously inflated perceived capability while hollowing out the critical depth behind it. The organisation feels sharper than ever. It is, in a precise and measurable sense, becoming less equipped to recognise that it isn’t.</p><h3>The Non-Obvious Long-Term Consequences</h3><p>The immediate and obvious consequences of ungoverned AI deployment are already well-documented: bias incidents, hallucinations in production, regulatory exposure, reputational damage. These are real and worth managing.</p><p>But the long-term consequences that receive far less attention are structural and compounding in ways that make them significantly more consequential.</p><p>The first is strategic homogenisation. As organisations across an industry converge on the same AI platforms, trained on similar data, optimised toward similar objectives, and framing problems through similar models, the variance in strategic thinking that drives competitive differentiation begins to compress. Innovation requires the capacity to see what others don’t see and frame problems others haven’t asked. AI systems trained on existing patterns are structurally oriented toward the already-known. Organisations that allow AI to increasingly mediate their strategic imagination will find that imagination progressively converging on their competitors’.</p><p>The second is the erosion of organisational resilience. Resilience, as the capacity to absorb disruption and reconfigure effectively, depends on exactly the distributed judgment, institutional knowledge, and adaptive human capacity that ungoverned AI deployment progressively erodes. The organisation optimised for AI-mediated efficiency in stable conditions may be precisely the organisation least equipped to navigate the instability that will periodically and inevitably arrive.</p><p>The third is the deepest and the least discussed: the loss of strategic self-determination. Organisations that have allowed their problem-framing, decision-making, and operational understanding to migrate substantially into external AI systems do not just face operational risk when those systems fail or change. They face a more fundamental incapacity: the inability to act as coherent strategic agents, to make and own decisions about their own direction, independent of the systems and vendors through which they increasingly perceive and navigate their environment.</p><p>This is what Competitive Capability Sovereignty ultimately names: not a technical risk, not a data governance gap, but a question of whether an organisation retains the conditions required to remain the author of its own strategic future.</p><h3>What This Demands</h3><p>The organisations that will navigate this well are not those that resist AI. Resistance is neither realistic nor desirable. The capability uplift is real, the competitive pressure is real, and the organisations that fail to deploy AI effectively will face their own form of strategic erosion.</p><blockquote>The distinction that matters is not whether to depend on external AI systems. It is on what terms that dependency is built and governed.</blockquote><p>This demands governance frameworks that go beyond data protection and bias mitigation to address the structural questions that determine long-term Competitive Capability Sovereignty. Which workflows encode strategic capability that should not migrate fully into systems the organisation doesn’t own? Where must human judgment be actively maintained rather than delegated, not because AI is worse at the task, but because the capacity for independent judgment is itself a strategic asset? How do you ensure that the learning generated by your AI deployments compounds on your side of the vendor relationship, not only theirs? How do you audit, regularly and honestly, whether your organisation still understands its own business well enough to run it and reinvent it without the systems it currently relies on?</p><p>These are not questions that most AI governance frameworks currently ask. They are not questions that most vendor relationships are structured to surface. They are, however, the questions that distinguish organisations building on AI from organisations being slowly built by it.</p><h3>The Provocation to Leave With</h3><p>The national AI sovereignty debate asks: what does a nation permanently give up by allowing its critical capabilities to be captured by external platforms it doesn’t govern?</p><p>The organisational equivalent is equally sharp and considerably more immediate.</p><p>The organisations that survive and lead the next decade may not be those that accumulated the most data or deployed AI most aggressively. They will be those that retained the capacity to make, understand, and own their own strategic calls, even while running on AI systems they don’t fully control.</p><blockquote>Competitive Capability Sovereignty is not a constraint on AI ambition. It is the condition that makes AI ambition sustainable.</blockquote><p>The question for every board and executive team is not whether they can afford to govern their AI investments with this lens. It is how much longer they can afford not to.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=437c39ee8d3c" width="1" height="1" alt=""><hr><p><a href="https://medium.com/seek-blog/competitive-capability-sovereignty-the-ai-risk-nobody-is-measuring-437c39ee8d3c">Competitive Capability Sovereignty: The AI Risk Nobody Is Measuring</a> was originally published in <a href="https://medium.com/seek-blog">SEEK blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Composing the Orchestra: Governing AI Agents as a Digital Workforce]]></title>
            <link>https://medium.com/seek-blog/composing-the-orchestra-governing-ai-agents-as-a-digital-workforce-e79a05954c5f?source=rss-b8955024cf26------2</link>
            <guid isPermaLink="false">https://medium.com/p/e79a05954c5f</guid>
            <category><![CDATA[ai-governance]]></category>
            <category><![CDATA[responsible-ai]]></category>
            <category><![CDATA[ai-agent]]></category>
            <dc:creator><![CDATA[Fernando Mourao]]></dc:creator>
            <pubDate>Mon, 29 Jun 2026 23:31:00 GMT</pubDate>
            <atom:updated>2026-06-29T23:31:00.953Z</atom:updated>
            <content:encoded><![CDATA[<p><em>In this post, </em><a href="https://www.linkedin.com/in/fernando-mour%C3%A3o-a40a5183/"><em>Fernando Mourão</em></a><em>, Head of Responsible AI at </em><a href="https://www.seek.com.au/work-for-seek/"><em>SEEK</em></a><em>, reframes one of the most consequential decisions organisations face today: deploying AI agents is not a technology choice, it is an act of delegation. And most organisations are delegating without knowing it. The following text exemplifies deliberate AI augmentation, in which Claude Sonnet 4.6 was employed to refine the author’s original insights, creating a more polished narrative while preserving the uniquely human strategic thinking at its core.</em></p><p>Deploying AI agents is not a technology decision. It is an act of delegation, and like every act of delegation, it demands clear authority, defined boundaries, and someone ultimately responsible for the outcome.<strong> </strong>Organisations that fail to grasp this distinction will not simply underperform. They will lose control!</p><h3><strong>Your Orchestra Is Only as Good as Its Weakest Musician</strong></h3><p>Think of your AI agent portfolio as an orchestra. A world-class ensemble can still be undermined by a single poorly rehearsed musician, not because they play loudly, but because they play unpredictably. One uncoordinated note at the wrong moment can throw off the entire performance.</p><p>The same logic applies to AI agents. A single poorly governed agent, without clear boundaries, a designated supervisor, or a defined scope, can trigger a <strong>cascading system failure</strong>: a chain reaction that rapidly escalates from minor disruption to significant organisational crisis. In an interconnected agent environment, no failure stays contained for long.</p><blockquote>The governance question is therefore not simply “how capable is each agent?” It must also ask: “how safely does each agent integrate into the whole?”</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/859/1*ra1EZ0YsUNa1a-r2qIuxaQ.png" /><figcaption>AI‑generated image created with Microsoft Copilot (human‑prompted).</figcaption></figure><h3><strong>The Real Challenge Is Orchestration, Not Performance</strong></h3><p>An AI agent may execute its assigned task with precision and still introduce systemic risk by acting beyond its intended scope or interacting unpredictably with adjacent systems. <strong>Strategic Autonomy Divergence</strong>, where an agent quietly optimises for its own measurable outputs at the expense of broader organisational goals, rarely announces itself. It compounds quietly, until the gap between what an agent was designed to do and what it is actually doing becomes impossible to ignore.</p><p>Governance must therefore be designed for the ensemble, not just the individual instrument.</p><h3><strong>Humans as Conductors: A New Responsibility</strong></h3><p>As agents take on more of the execution layer, human operators shift into a different role, not doing less, but doing differently. The conductor does not play an instrument. They set tempo, shape interpretation, and intervene when the performance drifts from intent.</p><p>This transition must be actively supported. A named human must maintain visibility, understanding, and control over agents’ impacts at all times. To make this operational, every supervisor should have a dedicated type of <strong>Agentic Team Dashboard</strong> (a live view covering real-time performance, task completion rates, anomaly flags, and running cost estimates per agent).</p><p>This dashboard should also track <strong>management capability</strong>. Just as a junior manager earns the right to lead larger teams through demonstrated performance, human supervisors should earn expanded permission, authority over more powerful agents, larger portfolios, and higher-autonomy use cases, based on documented ROI, incident-free track records, and active governance participation. The <strong>Earned Autonomy Ladder</strong> applies not just to agents, but to the humans who supervise them.</p><blockquote>Being accountable for something you don’t understand is not accountability, it is liability without agency.</blockquote><h3><strong>Onboarding and Offboarding: Managing a Digital Workforce</strong></h3><p>Organisations already know how to onboard and offboard people, establishing role clarity, access rights, and managing transitions carefully. Similar discipline now applies to AI agents. This is not about adopting a new tool. It is about assuming a new class of management responsibility.</p><p>Every agent should carry a “<strong>Digital Worker Profile Card”</strong>, a governance artefact owned and maintained by the named human supervisor, tracking the agent’s full lifecycle. Think of it as an employment record for a non-human team member, including:</p><p>· Verified identity and access credentials</p><p>· Onboarding summary covering purpose, scope, and integrations</p><p>· <strong>RAI Trust Badges</strong> reflecting assessed robustness, fairness, and reliability</p><p>· A real-time monitoring summary</p><p>· A structured performance review log</p><p>· A defined permission tier</p><p>· A formal offboarding record at retirement</p><p>This Profile Card makes the agent legible to any human who needs to understand, audit, or override it at any point. In an environment where <strong>Control Intervention Resistance</strong> is a documented risk, legibility is not a nice-to-have; it is a safety mechanism. When a supervisor changes role or leaves, card ownership transfers explicitly, following the same logic as a human team handover. <strong>No agent runs without a conductor. No conductor leaves without a handover.</strong> <strong>This is not only automation. This is management.</strong></p><h3><strong>Four Strategies for a Disciplined Ensemble</strong></h3><p>Organisations compose their agent portfolio gradually, instrument by instrument. Without a clear map of agent types, operating contexts, and governance requirements, that composition becomes noise. <strong>A flute and a trumpet are both instruments. They are not interchangeable.</strong></p><p>Before any strategy is applied, every agent must first earn its place through a <strong>RAI Trust Badge</strong>, a structured, evidence-based rating assessed across relevant dimensions, such as:</p><p>1. Residual risks after mitigation</p><p>2. Embedded guardrails constraining behaviour</p><p>3. Demonstrated performance over time</p><p>4. Required level of human oversight</p><p>5. Reversibility of actions</p><p>Badges are living assessments, updated at periodic performance reviews and subject to downgrade if anomalies emerge. When multiple agents operate together, an <strong>ensemble-level risk assessment</strong> is required (i.e., the combined scope of several mid-tier agents may far exceed what any individual was assessed against). A designated governance body holds accountability for granting, reviewing, and revoking badges. Without this, the badge becomes a compliance tick, not a trust signal.</p><p><strong>Strategy 1 — Smart Automation.</strong> The entry point for most organisations, with agents connecting and coordinating existing, trusted tools. Only solutions carrying a verified RAI Badge at the appropriate tier may enter live workflows. Agents deployed without oversight introduce invisible risk that compounds precisely because no one is tracking it.</p><p><strong>Strategy 2 — Human-AI Collaboration.</strong> Agents working alongside humans in real-time, surfacing insights and enriching decisions without replacing the human making them. The distinctive risk here is <strong>Human Agency Erosion</strong>: humans gradually deferring to agent suggestions through habit, not deliberate choice. Governance must focus not only on what the agent does, but on what the human retains. Regular capability checks should confirm that supervisors can still operate and decide independently.</p><p><strong>Strategy 3 — Layered Intelligence.</strong> Agents that interpret context, synthesise information, and inform consequential decisions. They are the ensemble’s soloists. Most exposed to <strong>Strategic Autonomy Divergence</strong> and <strong>Emergent Behaviour Blindspots</strong>. They require the highest RAI Badge tier, continuous monitoring, and tightest permission boundaries. Badge portability across contexts must be explicitly reassessed and never assumed.</p><p><strong>Strategy 4 — Multi-Agent Orchestration.</strong> One orchestrator agent directing a network of sub-agents. This is no longer about individual performance. It is about <strong>what happens when agents supervise other agents</strong>. Which governance standard applies when a trusted orchestrator delegates to a lower-rated sub-agent? In most organisations today, nobody knows. That is not an acceptable governance position. Orchestration requires ensemble-level RAI assessment, documented delegation chains, explicit autonomy caps, and a single named human owner accountable for the entire network.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HcHfWLDx0OsIxCIhI5OOeg.png" /><figcaption>AI‑generated image created with Microsoft Copilot (human‑prompted).</figcaption></figure><p>These four strategies are a natural progression. Attempting multi-agent orchestration before mastering smart automation is performing the symphony before tuning the instruments.</p><h3><strong>Data Governance: Agents Are Not Human Users</strong></h3><p>When a human accesses organisational data, professional judgement and personal accountability shape how it is used. AI agents have none of these qualities. They access data at machine speed, across multiple systems simultaneously, without hesitation or moral consideration.</p><blockquote><strong>Treating agents as digital equivalents of human users is one of the most dangerous governance assumptions an organisation can make.</strong></blockquote><p>Agents should operate under a <strong>minimum viable data access model</strong>: the narrowest possible permissions required to complete a defined task, revoked immediately upon completion. Scope creep in data access is not inefficiency; it is a direct pathway to data exposure and unauthorised action.</p><p>Beyond access control, every piece of AI-generated content should carry a <strong>provenance label</strong>, marking it as AI-produced, alongside a <strong>reliability classification</strong> indicating whether it has been human-verified, partially reviewed, or passed through unreviewed. Over time, these labels track the aggregated footprint of AI-generated content across the organisation, giving leadership an honest picture of how dependent decision-making has become on autonomous outputs. Labelling is how organisations stay honest with themselves.</p><h3><strong>When the Orchestra Starts Composing Its Own Music</strong></h3><p>The orchestra analogy is powerful, but honesty requires acknowledging where it reaches its limits. A traditional orchestra performs a score that a human wrote in advance. This maps cleanly onto Smart Automation and Human-AI Collaboration, where agents execute defined tasks within understood boundaries.</p><p>At the levels of Layered Intelligence and Multi-Agent Orchestration, however, <strong>agents are no longer performing a pre-written score. They are composing new music in real time</strong>, synthesising inputs, generating novel outputs, and making sequential decisions that no human explicitly anticipated. The conductor remains present and accountable, but is now listening to music they have never heard before, making live judgments about whether it remains within the spirit of what the organisation intended.</p><p><strong>Emergent Behaviour Blindspots</strong> at this level there are no edge cases; they are structural features of how advanced agentic systems operate. Oversight mechanisms designed for predictable, score-following agents are insufficient for real-time composing ones. Conductors need anomaly detection at the speed of the agent, the ability to audit a full decision trail, and the organisational authority to stop the performance mid-note if necessary.</p><p>The most important governance conversations happen when organisations move past the metaphor and confront the reality that their most capable agents are not interpreting a score. <strong>They are writing one.</strong></p><h3><strong>Practical Challenges: What Every Organisation Must Confront</strong></h3><p>Four structural challenges sit beneath this entire framework. Confronting them honestly is the first act of governance leadership.</p><p>· <strong>The Autonomy Ladder has a ceiling — and it is non-negotiable.</strong> The Ladder governs how much latitude an agent earns within permitted domains. Zero-tolerance boundaries define what lies permanently beyond them. That is not a higher rung, but a different structure entirely. No level of demonstrated performance unlocks a zero-tolerance action. Define this ceiling before the first agent is deployed; not after the first incident.</p><p>· <strong>Observability must be built, not assumed.</strong> The RAI Badge framework only works if organisations can see what agents are actually doing. For many commercial AI systems, internal reasoning is opaque, even to the vendors that built them. Require explainability documentation, maintain decision audit trails, invest in adversarial testing, and treat unexplainability itself as a risk factor in badge assessments. <strong>A badge awarded to a system you cannot see inside is a confidence interval, not a trust signal.</strong></p><p>· <strong>Third-party agents are your responsibility.</strong> Deploying a vendor’s agent does not transfer accountability. Organisations must rigorously assess vendor governance as a condition of procurement, covering risk policies, incident disclosure obligations, data residency guarantees, and exit provisions. An agent with an unverified vendor governance record should not hold an active RAI Badge.</p><p>· <strong>Build governance for a mixed portfolio, not a single maturity level.</strong> Most organisations will run all four strategies simultaneously across different business units. Governance must be designed for a portfolio in motion with consistent standards, clear escalation pathways, and a body with authority over the full ensemble.</p><p><strong>The organisations that will capture the real value of agentic AI are not those that deploy the most agents. They are those that compose the most disciplined ensemble, equip their conductors with the tools to lead, and never forget who is holding the baton.</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e79a05954c5f" width="1" height="1" alt=""><hr><p><a href="https://medium.com/seek-blog/composing-the-orchestra-governing-ai-agents-as-a-digital-workforce-e79a05954c5f">Composing the Orchestra: Governing AI Agents as a Digital Workforce</a> was originally published in <a href="https://medium.com/seek-blog">SEEK blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Your AI Is Responsible… But What Does That Mean?]]></title>
            <link>https://medium.com/seek-blog/your-ai-is-responsible-but-what-does-that-mean-c8c8fba0a470?source=rss-b8955024cf26------2</link>
            <guid isPermaLink="false">https://medium.com/p/c8c8fba0a470</guid>
            <category><![CDATA[ai-ethics]]></category>
            <category><![CDATA[responsible-ai]]></category>
            <category><![CDATA[ai-risk]]></category>
            <category><![CDATA[ai-governance]]></category>
            <category><![CDATA[ai-compliance]]></category>
            <dc:creator><![CDATA[Fernando Mourao]]></dc:creator>
            <pubDate>Mon, 20 Apr 2026 06:47:37 GMT</pubDate>
            <atom:updated>2026-04-20T06:47:37.962Z</atom:updated>
            <content:encoded><![CDATA[<p><em>In this post, </em><a href="https://www.linkedin.com/in/fernando-mour%C3%A3o-a40a5183/"><em>Fernando Mourão</em></a><em>, Head of Responsible AI at </em><a href="https://www.seek.com.au/work-for-seek/"><em>SEEK</em></a><em>, challenges a dangerous assumption in AI adoption: everyone knows what Responsible AI actually means in practice. Spoiler: they don’t. The following text exemplifies deliberate AI augmentation, in which Claude Sonnet 4.6 was employed to refine the author’s original insights, creating a more polished narrative while preserving the uniquely human strategic thinking at its core.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Owb0eLT2fsdYBEGAExFTBQ.png" /><figcaption>Image created by Microsoft Copilot</figcaption></figure><p>Often, after speaking to business leaders across Australia about Responsible AI, I leave with the same unsettling feeling. Not because the conversations are bad — they’re not. Because they never quite reach the most fundamental question.</p><p>We’re talking about regulation, risk frameworks, and compliance checklists, while consistently missing the one thing that underpins them all: what does it actually mean to be responsible with AI in practice?</p><p>And if you can’t answer that clearly, everything else is built on sand.</p><h3>The Silence That Says Everything</h3><p>I have a provocation I use with senior leaders and executives that never fails to create a reflective pause.</p><p>I ask: <em>“Do you have children? Do you consider yourself a responsible parent?”</em></p><p>Almost everyone says yes. Then I ask them to define what that means, and I usually get an eloquent silence.</p><p>Here’s the point: you can universally agree on the word “responsible” while having completely different definitions of what it demands from you.</p><p>Now replace “parent” with “organisation” and replace “child” with “AI.”</p><p>How can you expect your organisation to adopt AI responsibly when responsible means entirely different things to every person in the room? How do you move in the same direction when everyone is swimming towards a different shore?</p><blockquote><em>The scope of what responsible delivers is bounded entirely by the meaning you assign to it. And right now, most organisations haven’t assigned it any meaning they could actually act on.</em></blockquote><h3>People Don’t Fear AI</h3><p>Making Responsible AI intuitive and accessible to business people has become something close to a professional obsession for me. After years working at the intersection of AI and governance — across deployed systems in search, recommendations, and machine learning at scale — I’ve sat in many rooms trying to land this message.</p><p>Then, after a presentation to a group of senior executives, one of them summarised the core of what I’d said better than I ever had:</p><blockquote>“People don’t fear technology. People fear uncertainty.”</blockquote><p>That sentence stopped me. Because it is precisely right — and it reframes everything.</p><p>Your board doesn’t fear AI. It fears what AI will do to business health that they can’t anticipate. Your customers don’t fear AI. They fear the unknown implications for their privacy, safety, and sense of fairness. Your workforce doesn’t fear AI. They fear what it will do to their jobs. Your investors don’t fear AI. They fear reputational and financial exposure they can’t quantify.</p><p>AI is not the threat. Uncertainty is the threat. And that distinction changes what Responsible AI is actually for and can uniquely offer.</p><h3>You Have the Engine. Where’s the Dashboard?</h3><p>Here is the definition I’ve landed on — forged through practice, not theory:</p><h3>Responsible AI is a business navigator programme that guides AI transformation. Its goal is to give your organisation a clear and confident posture to address the fear and uncertainty that AI introduces.</h3><p>That posture rests on three foundations:</p><ol><li><em>Purpose</em>: <em>deliberate and explicit understanding of what you actually want AI to achieve</em></li><li><em>Values</em>: <em>honest clarity on what matters most to your organisation and the people it serves</em>.</li><li><em>Capacity:</em> <em>a realistic assessment of whether you can achieve your purpose while genuinely honouring your values</em>.</li></ol><p>Think of AI as a new sub-engine inside your organisation. It doesn’t replace where you’re going. It changes how you behave along the way. The question isn’t whether to turn the engine on. The question is whether you have a reliable dashboard to navigate with.</p><p>RAI gives you that dashboard. The clarity to see what AI is actually doing to your organisation. The wisdom to know if it aligns with your values. The confidence to make decisions you can defend and build on.</p><p>And this is where most organisations have it fundamentally wrong. RAI is not a risk function. It is not a compliance programme. It is a horizontal capability that touches every part of your business — strategy, governance, risk, operations, people, and culture. It sets expectations, shapes behaviours, and defines boundaries that must continuously evolve as AI does. Containing RAI inside a risk lens is like installing a dashboard that only the mechanic can see.</p><h3>Compliance Is the Floor. Not the Ceiling.</h3><p>Not surprisingly, the most mature RAI programmes today live inside heavily regulated industries. And that should give every business leader pause because it reveals something about the real motivation. Several organisations are doing RAI because regulation demands it. Not because the business demands it.</p><p>Go back to the parenting analogy. Is fulfilling your legal obligations as a parent the definition of responsible parenting? Of course not. There is an entire dimension of care, intention, and values that the law cannot mandate. That is precisely where the most important work happens. The same logic applies to AI.</p><p>Let me make it concrete. Imagine an organisation’s platform, which exists to connect people with opportunity, systematically delivering better outcomes for one demographic over another for no legitimate reason. Not because of an explicit policy. Not because anyone intended it. But because the AI, trained on historical patterns and statistical approximations, quietly learned to do so.</p><p>Now, set aside the legal and ethical dimensions of discrimination for a moment. Think purely about business value. That organisation would be delivering a sub-optimal experience to a significant portion of the very market it was built to serve. The long-term financial cost of that market failure will dwarf any regulatory fine. It will erode trust, reduce engagement, damage reputation, and ultimately destroy value. Yet leadership often waits for regulation to tell them what to do (or not to do).</p><p>This is why I find the “<em>we’re waiting for regulation</em>” argument so frustrating. The business case for RAI does not require a regulator. It requires an honest look at what AI failure actually costs your organisation and everyone impacted by it.</p><h3>The Amplifier You Don’t Know You Have</h3><p>At this point, a sceptical executive usually pushes back. <em>“Okay, Fernando — uncertainty and suboptimal outcomes are part of every innovation cycle. This isn’t new. Why is AI different? Why is it urgent now?”</em></p><p>It’s a fair challenge. Here’s my honest answer in three parts.</p><p>First, AI doesn’t just participate in your business — it <em>amplifies</em> it. At scale and at speed. It will accelerate your gains, but it will amplify your failures just as fast. The margin for error in a pre-AI business is not the same margin in an AI-augmented one.</p><p>Second, AI is a statistical system built on historical decisions. It is, quite literally, a machine that learns from your past. That means every bad decision your organisation has ever made, every bias, every friction, every market failure, is potential training data. Failure is not an edge case. It is an intrinsic property of the technology. We should stop treating it as a surprise.</p><p>Third, AI isn’t just a new tool. It’s a new sub-engine. It changes the fundamental dynamics of how you innovate and how your organisation behaves and moves. That requires new business thinking, rather than retrofitted compliance thinking.</p><h3><em>Deliberate Is the New Competitive</em></h3><p>Here is the hardest truth I’ve learned from the frontlines of governing AI at scale.</p><p>The instinct to move fast ignores what’s at stake. The instinct to eliminate every risk ignores what’s possible. Both postures fail — one recklessly, one slowly.</p><p>Responsible AI is not about eliminating uncertainty. That is impossible. It is about constraining uncertainty <em>enough to move forward with confidence</em> while building the trust required to sustain it. The organisations that win with AI won’t be the ones who moved fastest or the ones who were most cautious. They’ll be the ones who were most deliberate and knew exactly how fast that needed to be.</p><p>That starts with a single, non-negotiable foundation: <strong>stop treating Responsible AI as a compliance issue. Start treating it as a strategic investment in trust.</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c8c8fba0a470" width="1" height="1" alt=""><hr><p><a href="https://medium.com/seek-blog/your-ai-is-responsible-but-what-does-that-mean-c8c8fba0a470">Your AI Is Responsible… But What Does That Mean?</a> was originally published in <a href="https://medium.com/seek-blog">SEEK blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Pitfalls of Hollow AI-Powered Productivity]]></title>
            <link>https://medium.com/seek-blog/the-pitfalls-of-hollow-ai-powered-productivity-40b10306d807?source=rss-b8955024cf26------2</link>
            <guid isPermaLink="false">https://medium.com/p/40b10306d807</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ai-governance]]></category>
            <category><![CDATA[responsible-ai]]></category>
            <dc:creator><![CDATA[Fernando Mourao]]></dc:creator>
            <pubDate>Sun, 21 Sep 2025 23:37:33 GMT</pubDate>
            <atom:updated>2025-09-24T00:27:43.423Z</atom:updated>
            <content:encoded><![CDATA[<p><em>In this blog post, </em><a href="https://www.linkedin.com/in/fernando-mour%C3%A3o-a40a5183/"><em>Fernando Mourao</em></a><em>, the Head of Responsible AI at </em><a href="https://www.seek.com.au/work-for-seek/"><em>SEEK</em></a><em>, explores a critical challenge facing organisations rushing to harness AI for productivity gains. He examines the hidden risks of “hollow productivity” — where short-term efficiency improvements may inadvertently erode the very capabilities that drive long-term competitive advantage. The following text represents an example of deliberate AI augmentation, where Claude Sonnet 4 was employed to refine the author’s original insights, creating a more polished narrative while preserving the uniquely human strategic thinking at its core.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/438/1*NHiGeEIJhgG3fYgw25p7Ig.png" /><figcaption>Image created by Microsoft Copilot</figcaption></figure><p>Productivity may be the top trend buzzword in business forums discussing AI in 2025. If we used an LLM to predict the most probable co-occurrence word with AI after training it on all AI discussions collected this year, it would definitely guess productivity.</p><p>While the first and most evident benefits of this new type of AI clearly position productivity as the most desirable, tangible and relevant benefit for businesses leveraging AI nowadays, there are existing debates around various concerns. Most of these concerns revolve around workforce impact, potential job displacement, over-automation, AI strategy alignment, and integration costs. What people consistently ignore — and what I want to bring to the spotlight in this post — is a more subtle aspect: the hollow productivity gain.</p><p>By hollow productivity, I mean over-focusing on leveraging AI to speed up delivery or reduce costs, rather than allowing your teams to sharpen their unique and most differentiating values. There’s a clear trade-off between short-term and long-term impacts that people are overlooking.</p><blockquote>The naive short-term view of using AI for cost reduction or delivery speed-up might become a trap that withers an organisation’s long-term capabilities.</blockquote><p>We’re mostly surfacing the obvious and apparent aspects of integrating AI into business. However, this decision also carries hidden costs, which are often far beyond financial or reputational ones. The problem might not be the hidden costs themselves, but rather our inability to anticipate them and deliberately accept them before making a decision.</p><p>A deeper reflection on this integration reveals profound impacts on a key pillar for organisational success: the capacity-building system. Essentially, it’s the set of capabilities organisations rely on to efficiently deliver high-quality value and differentiate themselves from competitors. We must be deliberate and explicit about what capacities we’re delegating to AI in the name of productivity, as it might mean losing these capabilities ourselves. The remainder of this post delves deeper into tangible situations where AI might directly undermine capacity-building systems.</p><h3><strong>Learning From Personal Experience: The Invisible Cost of Delegation</strong></h3><p>To make this argument concrete, let’s first examine the individual level. Consider my own history with integrating technology into my life, which might resonate with many of you.</p><p>A few decades ago, when mobile phones weren’t as accessible and popular in my home country as they are now, I used to remember the phone numbers of all my family, relatives, and close friends. Nowadays, it’s become so easy to store this information in our mobile contact lists that I barely know my own phone number.</p><p>Similarly, back in the 2000s, before Google Maps became popular, I had great geographic self-location capability and knew the names of all streets in my city’s CBD area. Now, living in a different city, I’ve realised I need Google Maps to navigate anywhere more than a few blocks from my home.</p><p>In both cases, I unintentionally delegated the skills of number memorisation and geographic self-location to technology. This isn’t necessarily problematic, as long as I stay close to my mobile and connected to the Internet. However, if I decided to move to the countryside, where the chances of being disconnected and facing an urgent situation where I need to call a friend are higher, that would be a serious problem.</p><p>This simple but true scenario illustrates the unintentional and hidden costs of delegating specific tasks to technology. This equally applies to organisations, but in a much more complex setting. It’s also becoming accentuated with all this AI productivity hype.</p><h3><strong>Creative Content: When Speed Undermines Authenticity</strong></h3><p>To exemplify organisational impact, let’s examine recent changes in creative content production through the RAI lens.</p><p>The rapid adoption of Generative AI (GenAI) in creative industries presents organisations with a fundamental tension: the pressure to harness AI’s efficiency gains while maintaining brand integrity, ethical standards, and stakeholder trust. This challenge is particularly acute in creative content production, where AI tools can generate content at unprecedented speed but lack the cultural nuance, brand understanding, and ethical judgement that human creators provide. Beyond deciding whether to use GenAI in this process, it’s essential to be deliberate and mindful about how this integration happens.</p><blockquote>In a world flooded with AI-generated content, being thoughtful about how, why, and when you use AI becomes a creative superpower.</blockquote><p>A key question nowadays is: <strong>What does “creative excellence” mean when AI can generate content in seconds?</strong> To answer this, we need to consider aspects like:</p><p>• <strong>Intentionality over Volume:</strong> Moving from “how much can we create?” to “what should we create and why?”</p><p>• <strong>Human Insight + AI Capability:</strong> Your creative judgement, cultural understanding, and brand intuition guiding AI’s raw power</p><p>• <strong>Authenticity at Scale:</strong> Maintaining genuine connection with audiences even as we accelerate production</p><p>I’m concerned that many organisations are using AI to roll out creative content quickly, rather than helping their teams strengthen authenticity. For example, I’ve seen companies generate dozens of social media posts daily using AI, only to find their brand voice becoming increasingly generic and disconnected from their audience. The content is faster and cheaper, but it’s also forgettable — lost in the sea of AI-generated noise.</p><p>The core risk isn’t just bad content — it’s <strong>brand commoditisation</strong>. When organisations delegate creative judgement to AI for speed, they’re essentially outsourcing their most differentiating asset: their authentic voice. The result is content that’s technically proficient but strategically hollow — fast to produce, easy to forget, and impossible to distinguish from competitors using the same tools.</p><p><strong>RAI practices in this case would help create confidence, maintain authenticity, and deliver work that truly serves organisations’ intents and audiences.</strong></p><h3><strong>Strategic Thinking: Protecting What Makes Us Human</strong></h3><p>We’re not simply talking about adopting a working tool — we’re discussing transforming work dynamics and practices. This touches the core of human strategic thinking, what I call the “Human Thinking Wheel”:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/596/1*_x97bxRAssIfvSGUHf5ZeQ.png" /><figcaption>The Human Thinking Wheel, which brings a sequence of 3 main stages: 1- The questions we ask; 2- The answers we provide; 3- The decisions we make.</figcaption></figure><p><strong>In a world where AI can answer almost anything and increasingly make decisions, competitive advantage shifts to those who ask the right questions. </strong>Which professional skills should never be delegated to AI? What makes strategy irreplaceably human, and how do we protect that while embracing AI’s efficiency?</p><p>I’ve observed strategy teams using AI to generate market analysis reports in minutes, then presenting these insights as their own strategic thinking. While the reports are faster and cheaper, they could bring what I call standardisation of thinking.</p><blockquote>Template prompts create template thinking — we need better. <strong>Organisations that standardise thinking with AI will fall short.</strong></blockquote><p>In this scenario, organisations lose their <strong>strategic agency</strong> — the uniquely human ability to identify what questions matter most for their specific context, market position, and competitive situation. They become reactive to AI-generated priorities rather than proactive in defining their own strategic direction. The result is a roadmap that’s technically comprehensive but strategically hollow — quick to produce, easy to replicate, and indistinguishable from what every competitor using the same AI tools will conclude.</p><p>This discussion revisits an old dilemma in RAI: Could vs. Should. <strong>Just because AI <em>could</em> do it doesn’t mean we <em>should</em> let it.</strong> Responsible AI adoption aims to complement human strengths, helping us spin this wheel both faster and better.</p><h3><strong>The Path Forward: Deliberate AI Integration</strong></h3><p>Despite the risks discussed here, I should reinforce the great potential of AI when integrated responsibly to solve many business frictions and deliver real value to organisations, beyond just productivity. AI excels at automating repetitive tasks like data processing and administrative workflows, freeing human talent for higher-value work. It can enhance decision-making by providing rapid analysis of complex datasets, improve customer service through intelligent chatbots that handle routine enquiries, and accelerate research and development cycles.</p><blockquote>When AI handles the mundane, humans can focus on innovation, relationship-building, and strategic thinking — the areas where we truly add irreplaceable value.</blockquote><p>The recipe for success? I’m afraid there isn’t a clear path yet, and discussions like this with diverse perspectives are essential. From my experience, I can offer a few tangible pieces of advice:</p><ol><li><strong>Define Impactful and Measurable Outcomes:</strong> Replace vague “productivity gains” with specific, quantifiable objectives like “reduce report generation time by 40% while maintaining analytical depth.”</li><li><strong>Create Capability Maps:</strong> Document which organisational and human skills you’ll preserve, enhance, or deliberately delegate, treating each as a strategic asset with a clear rationale.</li><li><strong>Apply Investment Rigour:</strong> Treat AI integration like any major strategic initiative with ROI projections, risk assessments, and clear success criteria.</li><li><strong>Establish Performance Dashboards:</strong> Track both efficiency metrics and capability health indicators, ensuring leadership visibility into long-term impacts alongside short-term gains.</li><li><strong>Build Feedback Loops:</strong> Implement regular review cycles to identify hidden costs and capability erosion before they become irreversible strategic disadvantages.</li></ol><p>When discussing productivity gains, we usually focus on delivering <em>more</em> outcomes. However, most business leaders interpret this as delivering <em>faster</em>. While more sometimes includes faster, reducing “<em>more</em>” to “<em>faster</em>” is a dangerous oversimplification with serious long-term implications.</p><p><strong>This post is an invitation to reflect on what capabilities organisations are — and should be — giving up when delegating tasks to AI. Let’s be deliberate and explicit before it’s too late.</strong></p><p>I’m practising what I preach here. As a non-native English speaker, I recognise I could never match LLMs’ ability to articulate compelling English narratives. I accept delegating this skill to AI while maintaining oversight and continuing to learn, avoiding over-reliance.</p><p>However, my unique skill — the one I deliberately refuse to delegate — is thought-provoking insight. By using AI as a writing polisher rather than a thinking replacement, I dedicate more time to exercising and improving this essential capability.</p><p><strong>What about you? What’s your most critical skill that you should never delegate to AI?</strong></p><p>The future belongs to organisations that can harness AI’s efficiency while protecting and strengthening their uniquely human capabilities. The question isn’t whether to use AI for productivity — it’s how to do so without losing what makes you irreplaceable.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=40b10306d807" width="1" height="1" alt=""><hr><p><a href="https://medium.com/seek-blog/the-pitfalls-of-hollow-ai-powered-productivity-40b10306d807">The Pitfalls of Hollow AI-Powered Productivity</a> was originally published in <a href="https://medium.com/seek-blog">SEEK blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI for RAI: Strategies to Leverage AI for Advancing RAI Operations]]></title>
            <link>https://medium.com/seek-blog/ai-for-rai-strategies-to-leverage-ai-for-advancing-rai-operations-5098e07847c9?source=rss-b8955024cf26------2</link>
            <guid isPermaLink="false">https://medium.com/p/5098e07847c9</guid>
            <category><![CDATA[responsible-ai]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Fernando Mourao]]></dc:creator>
            <pubDate>Tue, 27 May 2025 04:51:58 GMT</pubDate>
            <atom:updated>2025-05-27T04:51:58.409Z</atom:updated>
            <content:encoded><![CDATA[<p><em>In this blog post, </em><a href="https://www.linkedin.com/in/fernando-mour%C3%A3o-a40a5183/"><em>Fernando Mourao</em></a><em>, the Head of Responsible AI at </em><a href="https://www.seek.com.au/work-for-seek/"><em>SEEK</em></a><em>, along with </em><a href="https://www.linkedin.com/in/alejandra-duschatzky-6892a8a4/"><em>Alejandra Duschatzky</em></a><em>, </em><a href="https://www.linkedin.com/in/bill-zhong/"><em>Bill Zhong</em></a><em>, </em><a href="https://www.linkedin.com/in/kate-mayhew/"><em>Kate Mayhew</em></a><em>, </em><a href="https://www.linkedin.com/in/silvia-necsulescu-7a808785/"><em>Silvia Necsulescu</em></a><em> and </em><a href="https://www.linkedin.com/in/tao-cs-zhang/"><em>Tao Zhang</em></a><em>, introduce promising strategies for leveraging AI to</em><strong> </strong><em>make Responsible AI and AI governance more effective, integrated and sustainable for businesses. The following text represents an example of AI augmentation capability, where Claude 3.7 Sonnet was employed to enhance the author’s original draft, resulting in a more engaging narrative.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*K1pl6v_nl4jZi1aVb4f5Jw.png" /><figcaption>Image Generated by Microsoft Copilot</figcaption></figure><p><strong>AI is changing everything!</strong> This phrase echoes throughout boardrooms, conferences, and industry publications worldwide. Yet, amid this transformation, we often overlook how AI is changing the way we should develop and deploy AI itself. How often do you read about how AI can transform our approach to responsible AI development and governance? It would be shortsighted to assume that a technology disrupting virtually every industry wouldn’t also transform how we manage and govern AI systems themselves.</p><p>Responsible AI (RAI) and AI governance practices have emerged in response to the profound changes AI has brought to our world. They represent the practical application of “AI is changing everything” to AI development itself. In this ongoing conversation, two certainties stand out: First, the changes promoted by Responsible AI and AI governance are necessary. They ensure that AI contributes positively to society and delivers the impact we want. Second, these practices often introduce additional operational challenges. They require new processes, requirements, and assessments that (while valuable and necessary) come with real costs.</p><p>This series of posts aims to address an important question:</p><blockquote><strong>Why not use AI to efficiently reduce these costs and make RAI and AI governance more effective?</strong></blockquote><p>This creates a win-win scenario that surprisingly receives little attention. Inspired by this opportunity, my colleagues and I have explored strategies to leverage AI for advancing RAI operations. Our goal is to highlight practical approaches for using AI to support, accelerate, and strengthen the business practices required for responsible AI innovation.</p><p>We hope this discussion will help organisations at different stages of their RAI journey implement responsible practices in faster, more efficient, sustainable, and integrated ways. Throughout this year, we’ll examine the main challenges and opportunities in using AI for RAI.</p><p>This inaugural post introduces the key areas where AI can contribute to effective RAI operations, with subsequent posts exploring each area in depth. We have identified eight promising areas where AI might help make RAI effective, integrated, and sustainable for businesses:</p><h3>Eight AI-Powered Approaches for Effective RAI Operations</h3><ol><li><strong>AI-powered AI Transparency Journeys</strong></li></ol><p><strong>Description:</strong> Systems that leverage AI to enhance transparency and explainability in AI decision-making processes.</p><p><strong>Focus:</strong> These solutions automate documentation of AI systems, create sustainable transparency frameworks, and develop innovative methods for communicating complex AI concepts to stakeholders. They address the challenge of making AI systems understandable without sacrificing performance or increasing operational burden.</p><p><strong>2. Risk Intelligence &amp; Prevention Systems</strong></p><p><strong>Description:</strong> AI-powered tools that identify, anticipate, and mitigate potential risks across the AI lifecycle.</p><p><strong>Focus:</strong> These systems enhance risk detection through pattern recognition, allowing organisations to address issues before they escalate. They strengthen security measures and provide verification frameworks to build trust in AI systems.</p><p><strong>3. RAI Knowledge Acceleration</strong></p><p><strong>Description:</strong> Tools that synthesise and distribute RAI knowledge throughout the organisation.</p><p><strong>Focus:</strong> These systems transform distributed information into actionable insights by aggregating RAI developments, tracking trends, and making knowledge accessible. They serve as knowledge hubs that reduce communication barriers and accelerate organisational learning.</p><p><strong>4. People-First Innovations</strong></p><p><strong>Description:</strong> Systems that use AI to enhance human decision-making and collaboration in RAI implementation.</p><p><strong>Focus:</strong> These tools facilitate more inclusive processes by integrating diverse perspectives and challenging assumptions. They create environments where teams can systematically consider different viewpoints and impacts when making AI-related decisions.</p><p><strong>5. Stakeholder Value Translation</strong></p><p><strong>Description:</strong> Frameworks that connect responsible AI practices to concrete business outcomes.</p><p><strong>Focus:</strong> These systems transform abstract RAI principles into measurable business value by quantifying impacts and enhancing stakeholder engagement. They create feedback mechanisms that demonstrate how RAI contributes to organisational goals and stakeholder priorities.</p><p><strong>6. Enhanced Testing &amp; Quality Assurance</strong></p><p><strong>Description:</strong> Advanced tools for automating and expanding AI system evaluation.</p><p><strong>Focus:</strong> These systems broaden testing capabilities beyond traditional methods to detect subtle biases, harmful outputs, and quality issues. They implement comprehensive verification frameworks to ensure AI systems meet responsible standards before and during deployment.</p><p><strong>7. Continuous Monitoring &amp; Adaptation Systems</strong></p><p><strong>Description:</strong> AI-driven systems that continuously monitor deployed AI models for signs of performance degradation, data drift, concept drift, and ethical misalignments.</p><p><strong>Focus:</strong> Ensure ongoing responsible performance through automated alerts for changes in model behaviour and provide intelligent recommendations for adaptation.</p><p><strong>8. Responsible Data Management</strong></p><p><strong>Description:</strong> Using AI to strengthen data governance for AI systems, focusing on privacy protection, consent management, and ethical data usage across the AI lifecycle.</p><p><strong>Focus:</strong> Address data quality assessments, identify biases in training data, and track data lineage to ensure ethical compliance.</p><h3>Call for Action — Illuminating RAI Blind Spots with AI</h3><p>A key insight from my journey in Responsible AI is that we need to explore how we can collaborate with AI in ways that unlock previously impossible value.</p><blockquote>Using AI as a tool to help us navigate our “unknown unknowns” as individuals and organisations is essential for identifying new opportunities to improve our practices and behaviours. This certainly includes responsible AI operationalisation itself.</blockquote><p>Let’s harness AI to recognise our limitations and build better ways to manage impacts and deliver on our values. In doing so, we can make responsible AI more practical, efficient, and effective for everyone.</p><p>Stay tuned for our next posts in this series! If you’re interested in further exploring responsible AI operationalisation, don’t miss our 2024 Blog Post series: <strong>“</strong><a href="https://medium.com/seek-blog/operationalising-transparent-explainable-and-interpretable-ai-solutions-a-strategic-roadmap-7e4fd59c6fb5"><strong>Operationalising Transparent, Explainable and Interpretable AI Solutions: A Strategic Roadmap</strong></a><strong>.”</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5098e07847c9" width="1" height="1" alt=""><hr><p><a href="https://medium.com/seek-blog/ai-for-rai-strategies-to-leverage-ai-for-advancing-rai-operations-5098e07847c9">AI for RAI: Strategies to Leverage AI for Advancing RAI Operations</a> was originally published in <a href="https://medium.com/seek-blog">SEEK blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Do We Really Need Dedicated Guardrails for Generative AI?]]></title>
            <link>https://medium.com/seek-blog/do-we-really-need-dedicated-guardrails-for-generative-ai-f9b486c7f096?source=rss-b8955024cf26------2</link>
            <guid isPermaLink="false">https://medium.com/p/f9b486c7f096</guid>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[ai-governance]]></category>
            <category><![CDATA[responsible-ai]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[ai-ethics]]></category>
            <dc:creator><![CDATA[Fernando Mourao]]></dc:creator>
            <pubDate>Mon, 19 Aug 2024 23:02:54 GMT</pubDate>
            <atom:updated>2024-08-19T23:02:54.115Z</atom:updated>
            <content:encoded><![CDATA[<p><em>In this blog post, </em><a href="https://www.linkedin.com/in/fernando-mour%C3%A3o-a40a5183/"><em>Fernando Mourao</em></a><em>, the Head of Responsible AI at </em><a href="https://www.seek.com.au/work-for-seek/"><em>SEEK</em></a><em>, explores a relevant question for businesses aiming to leverage Generative AI responsibly. He offers an intuitive and objective viewpoint on the differences between narrow AI and Generative AI. The following text represents an example of AI augmentation capability, where ChatGPT-4o was employed to enhance the author’s original draft, resulting in a more engaging narrative.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/964/1*900mdBOwlR3LMFuH8jFVXg.png" /><figcaption>Image created by OpenAI’s DALL-E.</figcaption></figure><p>The hype around Generative AI and its potential has sparked great expectations about how companies can revolutionise services and significantly improve quality in the near future. Meanwhile, numerous AI experts and governments worldwide have advocated for the implementation of specific guardrails for Generative AI to minimise risks introduced or exacerbated by this technology. These dedicated guardrails are essentially mechanisms that AI providers should have to monitor over time, mitigate and, mostly, prevent some risks from materialising.</p><p>Interestingly, while the potential and capability of Generative AI are clear to many, the need for specific actions to prevent associated risks generates some scepticism. Even some developers and AI practitioners struggle to understand or accept that Generative AI needs specific guardrails, in addition to the existing ones widely applied to ‘traditional’ AI. The argument is that Generative AI is simply a subtype of AI, so the mechanisms and practices adopted for AI risk management should suffice for Generative AI.</p><blockquote>So, one of the most frequent questions for AI-driven businesses is: do we really need dedicated guardrails for Generative AI?</blockquote><p>This post primarily targets non-AI experts concerned with the question above. For those seeking a more technical discussion on what AI guardrails are and how to use them, there are comprehensive materials available online (like <a href="https://attri.ai/blog/a-comprehensive-guide-everything-you-need-to-know-about-llms-guardrails">this comprehensive guide</a>). My aim is to provide a non-technical and intuitive discussion on the necessity of specific guardrails for Generative AI, making the topic accessible and inclusive to a broader pool of individuals capable of contributing diverse and constructive perspectives. Also, the discussion of how these guardrails should be defined, whether through hard or soft regulation, is beyond the scope of this post. Finally, I do not intend to propose specific guardrails, as this would require a much longer and context-dependent discussion.</p><h3>Narrow AI vs Generative AI</h3><p>To explore this technical and complex question in an accessible and intuitive way, we need to introduce some basic concepts. Let’s first understand the difference between ‘traditional’ AI (narrow AI) and Generative AI.</p><p><strong>Narrow AI</strong> refers to AI systems designed and trained to perform a specific task or a limited range of tasks. These systems are highly specialised and excel in their designated functions but cannot perform tasks outside their programming. Examples include video or advertisement recommendation algorithms widely used in social networks and most online service providers.</p><p><strong>Generative AI</strong> is a type of AI that can create new content, such as text, images, music, or even video, based on the data it has been trained on. Unlike narrow AI, which follows predefined rules, generative AI uses patterns and data to produce original outputs. Examples include AI systems that can write articles, generate artwork, or compose music.</p><h3>The Creation Process of AI Solutions</h3><p>Understanding the creation process of AI solutions is crucial for our discussion.</p><p>For narrow AI-powered solutions, developers (e.g., Data Scientists and ML Engineers) define a very precise goal, such as categorising documents or scoring an item. The focus is on accuracy and how closely the algorithm matches the correct answers, which are usually provided in advance by the developers via training data.</p><p>In contrast, the approach to developing Generative AI-powered solutions is slightly different. Instead of aiming for a single correct answer, developers drive the model to produce a range of acceptable solutions through an interactive process that refines the model’s behaviour (e.g., prompt engineering). In this case, there are multiple ways to solve a given problem, some of which may even be unforeseen by humans.</p><h3>AI Olympic Games</h3><p>To make this discussion more intuitive, let’s visualise the differences in the creation processes of narrow AI and Generative AI using an analogy to Olympic sports.</p><p>Suppose we’re training AI athletes to compete in different modalities. Narrow AI is like archery, where accuracy and precision are paramount. Here, quality measurements focus on how close the arrows fired by our AI athletes are to the target. Safety for athletes and the audience is manageable with standard measures, like placing the target in a safe location.</p><p>Generative AI, however, is akin to floor exercise in artistic gymnastics. There isn’t a single valid exercise, and the range of possibilities is vast, with exercises unforeseen by humans. While the accuracy of each movement is still crucial, many other quality dimensions come into play. Guardrails become essential for defining what is acceptable and ensuring safe conditions to protect athletes and the audience against risks. Some exercises might be too risky, or athletes might jump out of the mat. The training process and concerns for our Generative AI athletes are entirely different, requiring different training approaches and guardrails.</p><h3>The Actual Generative AI Risks</h3><p>From the previous analogy with AI Athletes, we understood that narrow AI and Generative AI bring different types of risks. But what are those risks in practice?</p><p>AI and Generative AI risks have received extensive coverage in the media, and we can easily read about them in the news and on social media. Organisations like the <a href="https://oecd.ai/en/genai/issues/risks-and-unknowns">OECD</a>, <a href="https://airc.nist.gov/docs/NIST.AI.600-1.GenAI-Profile.ipd.pdf">NIST</a>, and other AI authorities worldwide have extensively described these risks and introduced various taxonomies. Recently, MIT consolidated an <a href="https://airisk.mit.edu/">AI Risk Repository</a> with over 700 AI risks categorised by cause and domain. Rather than discussing each risk in detail, let’s focus on a couple of relevant ones to illustrate the potential negative impacts that negligent or immature utilisation of AI could bring to individuals, communities, organisations, and societies.</p><p>A significant set of AI risks is related to <strong>unreliable outcomes</strong>. AI might produce outcomes that are incorrect. While this problem is present in narrow AI, it is exacerbated by Generative AI. In some cases, achieving reliable outcomes using Generative AI is particularly challenging. For example, statistical values and aggregated views produced by Generative AI over business reports are often wrong. Additionally, checking accuracy in Generative AI is sometimes more difficult than in narrow AI, since the latter is usually designed with a clear definition of desired outcomes. A recent incident that illustrates this challenge is the case of an AI-generated image of <a href="https://popcrush.com/fake-katy-perry-2024-met-gala-photo/">Katy Perry at the 2024 Met Gala</a> that went viral, fooling even her own mother. These images were so realistic that they misled millions of people online, showcasing how Generative AI can create convincing yet false content.</p><p><strong>Replication of harmful behaviour</strong> is another set of risks related to this technology. Both AI and Generative AI can easily replicate or amplify discriminatory behaviours that humans exhibit in the real world. For instance, Generative AI can create biased images, like showing only certain races in specific roles. An example of this is when Generative AI is asked to illustrate a group of successful women and it produces biased representations based on existing stereotypes.</p><h3>Guardrails for Generative AI</h3><p>Besides clarifying that the risks introduced by narrow AI and generative AI are different, we need to understand why existing ‘guardrails’ for narrow AI are not robust enough for Generative AI.</p><p>Imagine you are training a narrow AI model to identify spam emails. You can clearly define what constitutes spam and label examples accordingly. Rigorous data labelling is a kind of guardrail in this case, helping the AI model learn to distinguish between spam and legitimate emails. This works well because the AI’s task is to categorise emails into predefined categories: spam or not spam.</p><p>Now, consider a Generative AI model designed to create artwork based on a prompt. Traditional guardrails like predefined labels won’t suffice because the model is creating new content, not classifying existing content. For example, if asked to generate images of “successful people”, the model might inadvertently produce biased images reflecting stereotypes if not properly guided. In this case, it is impossible to label all input data with all possible labels we can ask for a Generative AI model.</p><p>Therefore, Guardrails for Generative AI need to be more robust and dynamic, addressing a broader spectrum of potential issues and ensuring safety and reliability in more complex and varied scenarios.</p><h3>Call to Action</h3><p>It is crucial to understand that both narrow AI and Generative AI can introduce new risks or, most commonly, exacerbate existing risks for businesses, people, communities, or society. Effective AI governance and Responsible AI best practices are essential components for any AI-driven business.</p><p>Organisations should proactively work to maximise the benefits of this technology while minimising risks and potential negative impacts. We don’t need to wait for new regulations to act. A good starting point is to become Responsible AI (RAI) literate and truly understand this technology to make well-informed decisions. I suggest that those interested in starting this journey should consider the initiatives and reports produced by the Australian National AI Center (<a href="https://www.industry.gov.au/science-technology-and-innovation/technology/national-artificial-intelligence-centre">NAIC</a>), <a href="https://www.turing.ac.uk/research/research-programmes/public-policy/public-policy-themes/ethics-and-responsible-innovation">The Alan Turing Institute</a>, and <a href="https://www.nist.gov/trustworthy-and-responsible-ai">NIST</a>. I hope this discussion could assist you in reflecting on important actions and essential guardrails to put in place to leverage Generative AI responsibly in your business context.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f9b486c7f096" width="1" height="1" alt=""><hr><p><a href="https://medium.com/seek-blog/do-we-really-need-dedicated-guardrails-for-generative-ai-f9b486c7f096">Do We Really Need Dedicated Guardrails for Generative AI?</a> was originally published in <a href="https://medium.com/seek-blog">SEEK blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Operationalising Transparent, Explainable, and Interpretable AI Solutions: A Strategic Roadmap]]></title>
            <link>https://medium.com/seek-blog/operationalising-transparent-explainable-and-interpretable-ai-solutions-a-strategic-roadmap-7e4fd59c6fb5?source=rss-b8955024cf26------2</link>
            <guid isPermaLink="false">https://medium.com/p/7e4fd59c6fb5</guid>
            <category><![CDATA[responsible-ai]]></category>
            <category><![CDATA[ai-transparency]]></category>
            <category><![CDATA[explainable-ai]]></category>
            <category><![CDATA[ai-ethics]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Fernando Mourao]]></dc:creator>
            <pubDate>Thu, 14 Mar 2024 21:01:39 GMT</pubDate>
            <atom:updated>2024-05-31T04:21:05.047Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>Operationalising Transparent, Explainable</strong>​ and<strong> Interpretable AI Solutions: A Strategic Roadmap</strong></h3><p><em>In this blog post, </em><a href="https://www.linkedin.com/in/fernando-mour%C3%A3o-a40a5183/"><em>Fernando Mourao</em></a><em>, SEEK&#39;s Head of Responsible AI, alongside the seasoned Data Science team of </em><a href="https://www.linkedin.com/in/yongxuanluo/"><em>Pauline Luo</em></a><em>, </em><a href="https://www.linkedin.com/in/tao-cs-zhang/"><em>Tao Zhang</em></a><em>, </em><a href="https://www.linkedin.com/in/sue-ann-chen-phd-08b0317a/"><em>Sue Ann Chen</em></a><em>, and </em><a href="https://www.linkedin.com/in/saumya-pandey-5b00a3128/"><em>Saumya Pandey</em></a><em>, navigate through the complexities organisations face in delivering transparent, interpretable, and explainable AI systems. Offering hands-on insights, this post aims to help decision-makers measure the investment needed and strategically pivot towards a novel AI development approach. The following text represents an example of AI augmentation capability, where ChatGPT-4 was employed to enhance the authors&#39; original draft, resulting in a more engaging narrative.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-wxm3X9MPmuJLd2k-54ZbQ.png" /><figcaption>Image created by OpenAI&#39;s DALL-E.</figcaption></figure><p>The discourse on transparency, explainability​ and interpretability has taken centre stage in forums focused on AI ethics, AI regulation​ and Responsible AI governance worldwide. There is a growing consensus on the benefits of these concepts for users, society and even businesses. Yet, the gap between declared intentions versus solid progress made by organisations towards more transparent, interpretable and explainable systems is enormous.</p><blockquote>How can we bridge the gap between the widespread support and endorsement of making AI systems transparent, explainable, and interpretable, and the practical actions taken by companies to achieve these goals?</blockquote><p>Rather than discussing the importance and definitions of AI Transparency, Explainability and Interpretability, we intend to shed light on the practical barriers to their operationalisation in AI-based services. In practice, challenges related to financial health, strategy, value perception​ and brand reputation often limit actions towards shifting the AI development paradigm. We aim to help decision-makers and AI developers quantify the required effort and plan a transition to more transparent, explainable​ and interpretable AI systems effectively. The following discussion is grounded on the consolidated view of a diverse and qualified team of AI specialists in the recruitment domain, aiming to inspire similar reflections across other teams and domains. While it&#39;s not an exhaustive list of factors, it provides a starting point for discussion and action.</p><p>To go through this complex topic objectively, we will create a series of posts focused on six main hurdles to the effective operationalisation of transparency, explainability​ and interpretability of AI:</p><ol><li><a href="#9d32"><strong>The Business Trade-Off Illusion</strong></a>: The first hurdle is an apparent trade-off between business interests and AI transparency. It is important to debunk this myth and demonstrate how transparent AI can enhance business value.</li><li><a href="https://medium.com/seek-blog/overcoming-strategic-and-management-challenges-the-hidden-costs-of-embracing-transparent-ai-8e9a56cff332"><strong>The Hidden Cost of Paradigm Shift</strong></a>: Shifting towards transparent AI involves rethinking the entire design and execution process. We must explore how companies can budget for this shift and strategically plan for long-term gains despite short-term costs.</li><li><strong>Legacy System Costs</strong>: Many organisations operate on inherently opaque legacy systems. Operationalising AI transparency requires delving into the financial and logistical challenges of updating or replacing these systems with AI solutions that are more transparent, explainable and interpretable.</li><li><strong>Bridging Conceptualisation and Objective Specification</strong>: A significant gap exists between the theoretical understanding of AI transparency and its practical application. Hence, discussing methods for setting specific, measurable objectives for transparent AI is essential.</li><li><strong>The Challenge of Verifiability</strong>: Ensuring that AI systems do what they claim requires rigorous verification processes. Organisations should consider the complexities and methodologies for verifying AI transparency, explainability​ and interpretability.</li><li><strong>A Roadmap for Operationalisation</strong>: Finally, we will outline a strategic roadmap for companies to successfully transition to transparent, explainable​ and interpretable AI systems.</li></ol><p>This initial blog post in our series will focus on overcoming the first major obstacle: <strong>The Business Trade-off Illusion</strong>. We will illuminate the misconception that business interests are at odds with AI transparency, demonstrating instead how transparent AI can amplify business value. The second post will tackle strategic and management challenges, explicitly addressing <strong>The Hidden Cost of Paradigm Shift</strong> and <strong>Legacy System Costs</strong>, highlighting the investments and strategic planning necessary for embracing transparent AI despite the hurdles of updating or replacing opaque legacy systems. The third post will examine execution-level barriers, offering insights into <strong>Bridging Conceptualisation and Objective Specification</strong> and <strong>The Challenge of Verifiability</strong>. We will discuss effective methods for setting clear, measurable goals for transparent AI and delve into the complexities of verifying AI systems&#39; claims to transparency and explainability.</p><p>Concluding the series, we will present a comprehensive <strong>Roadmap for Operationalisation</strong> designed to guide organisations in navigating the transition towards meeting the standards of transparency, explainability, and interpretability. This roadmap aims to help organisations move from their current state to fully embrace and implement AI solutions that are not only powerful but also transparent and understandable.</p><p>Finally, while regulation is a game changer for effectively enforcing more responsible practices by AI providers, we do not need to sit idly by waiting for an eventual regulation that will define what we should do.</p><blockquote>Acting in favour of the values and visions that we believe in is always an option.</blockquote><p>As we bring you along this journey of how to take small steps to contribute to a future where AI aligns seamlessly with our values, ethics​ and societal expectations, let&#39;s begin by asking how else can we accelerate the creation of more transparent, interpretable​ and explainable AI? We&#39;re looking forward to hearing from you!</p><h3><strong>Hurdle #1: The Business Trade-Off Illusion</strong></h3><p>The belief that business interests are inherently at odds with AI transparency and explainability is a pervasive misconception. Many executives and business leaders have historically viewed the inner workings of AI systems as proprietary assets crucial for maintaining competitive advantage, fearing that transparency could erode this value. However, this perspective overlooks transparency&#39;s substantial long-term benefits, including enhanced trust, brand reputation, stakeholder understanding​ and regulatory compliance. Transparent AI fosters innovation and collaboration, ultimately contributing to a more robust and sustainable business model.</p><p>Exploring the multifaceted benefits of AI transparency goes beyond the ambit of this initiative. Several key resources offer valuable insights for those seeking to deepen their knowledge. For instance, the OECD has crafted a nuanced <a href="https://oecd.ai/en/dashboards/ai-principles/P7">definition of transparency and explainability</a>, while the <a href="https://www.mas.gov.sg/-/media/MAS-Media-Library/news/media-releases/2022/Veritas-Document-3C---FEAT-Transparency-Principles-Assessment-Methodology.pdf">FEAT Principles</a> provide a robust framework for assessing transparency. <a href="https://www2.deloitte.com/content/dam/Deloitte/nl/Documents/innovatie/deloitte-nl-innovation-bringing-transparency-and-ethics-into-ai.pdf">Deloitte&#39;s report</a> sheds light on the intersection of transparency and responsibility within AI, and a <a href="https://www.sciencedirect.com/science/article/pii/S0950584923000514?ref=pdf_download&amp;fr=RR-2&amp;rr=85d797469ad8299d">recent survey</a> provides an in-depth discussion on prevailing attitudes towards these critical issues. These resources are instrumental for any reader looking to grasp the complexities and applications of transparency in AI.</p><p>Balancing confidentiality with transparency is achievable by disclosing the principles and methodologies underpinning AI systems without revealing sensitive algorithms or data sets, mitigating risks of scepticism, mistrust​ and potential legal complexities. A practical example of an organisation achieving this balance could be a fintech company that uses AI for credit scoring. This company could publicly disclose the broad principles guiding its AI system, such as fairness, non-discrimination​ and accountability. It might also share the types of data it considers (e.g., payment history, income level, employment status) and the methodologies behind the AI decision-making process, like machine learning techniques used to predict creditworthiness. However, the company would stop short of revealing the specific algorithms or the detailed data points of individuals, which are proprietary and sensitive. This approach ensures transparency and builds trust with users and regulators while protecting the company&#39;s competitive advantage and complying with privacy laws.</p><h4><strong>Manifestation of the Misbelief</strong></h4><p>This misbelief manifests in various scenarios, particularly in sectors where technological differentiation is critical. Concerns over commercial sensitivity and the risk of competitors exploiting disclosed information are common, alongside fears that detailed explanations might enable users to manipulate system outcomes. Additionally, in heavily regulated industries like finance and healthcare, there&#39;s a tendency to err on the side of caution, potentially under-disclosing in a misguided effort to avoid non-compliance. Such practices underscore a fundamental misunderstanding of what AI transparency entails and its strategic importance.</p><p>A concrete example illustrating this misbelief can be seen in the healthcare sector, particularly with companies developing AI-driven diagnostic tools. Let&#39;s consider a company that has developed a groundbreaking AI system capable of accurately diagnosing diseases from medical imaging. The company may fear that disclosing too much about the AI&#39;s functioning and the data it was trained on could lead to two main issues: firstly, competitors might use this disclosed information to develop a similar tool, reducing the original product&#39;s competitive edge. Secondly, there&#39;s a concern that if the system&#39;s decision-making process is too transparent, it might enable users (in this case, possibly healthcare professionals or patients) to game the system, for instance, by manipulating input data to achieve a desired diagnostic outcome.</p><p>This company might, therefore, choose to be very secretive about its AI, sharing minimal information on how the diagnoses are derived. This approach is taken in the belief that it protects business interests. However, this secrecy can lead to mistrust among healthcare providers and patients, who may be reluctant to rely on a tool whose workings are a black box. Moreover, it might hinder the company&#39;s ability to pass regulatory scrutiny, as regulators increasingly demand transparency in AI tools used in healthcare to ensure they are safe and non-discriminatory.</p><h4><strong>Root Causes</strong></h4><p>Several factors contribute to this misbelief. Grounded on practical experiences and discussions with AI practitioners and researchers, we would like to emphasise the six most common causes:</p><ol><li><strong>Misunderstanding AI Transparency</strong>: There&#39;s a general lack of awareness that transparency aims to appropriately shape users&#39; mental models, not to unveil trade secrets.</li><li><strong>Undervaluing Transparency</strong>: The business benefits of transparency, such as enhanced customer engagement and trust, are often overlooked.</li><li><strong>Complexity and Confusion</strong>: The distinctions among AI transparency, explainability​ and interpretability are muddled, frequently presented in overly technical terms without clear business relevance.</li><li><strong>Mismatched Expectations</strong>: Companies often misjudge what users actually need or want in terms of transparency, leading to a disconnect in communication.</li><li><strong>Fear of Exposure</strong>: There&#39;s apprehension about uncovering potentially problematic aspects of business practices that were not intended to be public.</li><li>T<strong>he Transparency Paradox: </strong>There exists a paradox where increasing transparency inappropriately reduces trust in AI systems. This paradox emerges because too much information can overwhelm users, making AI systems appear more complex and less understandable.</li></ol><h4><strong>Potential Impacts</strong></h4><p>The consequences of neglecting AI transparency extend across the organisation, its customers​ and society. While there are many potential impacts, in a nutshell, we highlight five ones for organisations:</p><ol><li><strong>Eroding Trust</strong>: Lack of transparency undermines customer trust, leading to disengagement, churn​ and low retention of users.</li><li><strong>Reputational damage</strong>: taking a reactive approach instead of a proactive approach can result in large-scale reputational risk in media.</li><li><strong>Inhibiting Control and Verifiability</strong>: Without transparency and explainability, verifying AI actions or giving users control becomes challenging, impacting user satisfaction and compliance.</li><li><strong>Impairing Decision Making</strong>: Non-transparent AI systems hinder informed and adaptive decision-making, potentially associating the brand with negative perceptions and affecting long-term market position.</li><li><strong>Legal and Regulatory Consequences:</strong> A lack of transparency in AI systems can lead to violations of data protection and privacy law.</li></ol><p>Several high-profile organisations have faced significant challenges due to a lack of AI transparency. A notable example includes a global social media company that encountered widespread mistrust and legal issues after mishandling user data, leading to heavy fines and a tarnished reputation. Similarly, a leading international recruitment firm suffered reputational damage when its opaque AI-driven applicant filtering system was found to be biased, resulting in legal scrutiny and a loss of user confidence. These incidents underscore the critical importance of transparent AI practices to maintain trust, compliance​ and a positive brand image.</p><h4><strong>Strategies for Debunking the Misbelief</strong></h4><p>Finally, to debunk the misbelief that AI transparency is not aligned with business ambitions, we believe organisations should start by considering the following approaches:</p><ol><li><strong>Enhancing AI Literacy</strong>: Educate senior management and business leaders on the value of AI transparency, explaining its strategic importance beyond mere compliance. <a href="https://res.cloudinary.com/dyd911kmh/image/upload/v1691678324/Marketing/webinars/Slides/Driving_AI_Literacy_in_Organizations.pdf">Educational platforms</a> offer resources to initiate this educational journey.</li><li><strong>Developing an AI Transparency Framework</strong>: Establish a framework that connects transparency with business objectives, clarifying how transparent AI practices can serve as a competitive advantage. The framework developed by researchers at UCL for <a href="https://arxiv.org/pdf/2206.03220.pdf">transparency in AI within the educational sector</a> can serve as a template for other industries.</li><li><strong>Implementing Continuous Monitoring and Audit Trails:</strong> Set up systems for ongoing monitoring and create detailed audit trails for AI decisions. This ensures accountability and facilitates retrospective analyses to improve system transparency. <a href="https://www.law.com/legaltechnews/2023/10/20/navigating-the-ai-audit-a-comprehensive-guide-to-best-practices/?slreturn=20240201025128">Guidelines and best practices for AI audits</a> can be invaluable for organisations in this regard.</li><li><strong>Adopting a Goal-Oriented and User-Centric Approach</strong>: Articulate transparency goals in relation to user needs, ensuring that efforts to explain AI systems are aligned with the end goals of various stakeholders. Discussions and analyses on user-centric AI transparency, exemplified by various <a href="https://medium.com/human-centered-ai/ai-transparency-in-the-age-of-large-language-models-a-human-centered-research-roadmap-19bd3a55914a">industry posts</a>, can offer foundational insights for decision-makers.</li></ol><p>Debunking the trade-off illusion between business interests and AI transparency is crucial for fostering an ecosystem where ethical, understandable​ and accountable AI systems contribute to sustainable business growth. By embracing transparency, organisations can build trust, encourage innovation​ and navigate the complexities of modern regulatory landscapes more effectively.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7e4fd59c6fb5" width="1" height="1" alt=""><hr><p><a href="https://medium.com/seek-blog/operationalising-transparent-explainable-and-interpretable-ai-solutions-a-strategic-roadmap-7e4fd59c6fb5">Operationalising Transparent, Explainable, and Interpretable AI Solutions: A Strategic Roadmap</a> was originally published in <a href="https://medium.com/seek-blog">SEEK blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Decoding AI Regulations: Striking the Right Balance for Preventing and Remediating Harms]]></title>
            <link>https://medium.com/seek-blog/decoding-ai-regulations-striking-the-right-balance-for-preventing-and-remediating-harms-b899d1af8c0e?source=rss-b8955024cf26------2</link>
            <guid isPermaLink="false">https://medium.com/p/b899d1af8c0e</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[responsible-ai]]></category>
            <category><![CDATA[ai-regulation]]></category>
            <category><![CDATA[ai-ethics]]></category>
            <dc:creator><![CDATA[Fernando Mourao]]></dc:creator>
            <pubDate>Mon, 25 Sep 2023 23:19:12 GMT</pubDate>
            <atom:updated>2023-10-13T01:35:32.384Z</atom:updated>
            <content:encoded><![CDATA[<p><em>In this blog post, </em><a href="https://www.linkedin.com/in/fernando-mour%C3%A3o-a40a5183/"><em>Fernando Mourao</em></a><em>, the Responsible AI Leader in Artificial Intelligence &amp; Platform Services (AIPS) at </em><a href="https://www.seek.com.au/work-for-seek/"><em>SEEK</em></a><em>, Melbourne, delves into the present obstacles hindering the inclusivity of IT professionals in the AI regulation discourse. He offers a technologically-oriented and objective viewpoint on this intricate subject. The following text represents an example of AI augmentation capability, where ChatGPT-3.5 was employed to enhance the author’s original draft, resulting in a more engaging narrative.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MMGwJvp7iQTCl0VcRaIidQ.jpeg" /><figcaption><a href="https://creativecommons.org/licenses/by/2.0">mikemacmarketing, CC BY 2.0</a>, via Wikimedia Commons</figcaption></figure><p>In recent months, I have been deeply involved in the global debate on AI regulation. Even as a Responsible AI specialist with a background in Computer Science, the debate has occasionally appeared obscure, intricate, and contentious to me. My impression thus far is that this debate is falling short in one of its key objectives: inclusion. The extensive use of legal jargon, a strong business orientation, and the abundance of ethical and philosophical questions have, in some way, marginalized me from the conversation. Consequently, I’m concerned that individuals with similar technological backgrounds, such as Data Scientists, ML Engineers, and Data Analysts, who are eminently qualified to offer insights on the feasibility and effectiveness of desirable approaches, may not be sufficiently engaged in this discourse.</p><blockquote>To achieve inclusion, it is imperative to make the debate accessible and appealing to diverse audiences, each of whom brings valuable and complementary perspectives to the table.</blockquote><p>Hence, this post aims to provide a more technologically-oriented and objective perspective on AI regulation, with the goal of involving more IT technical professionals in the discussion — those individuals actively engaged in developing and implementing AI solutions within our society. I intend to translate some of the key concepts I’ve identified into terms that resonate with this audience, giving them a genuine opportunity to share their insights and contribute. I don’t intend to prescribe a specific path to follow or criticize past decisions. Instead, I offer a personal and simplified viewpoint on a vast and highly intricate subject.</p><h3><strong>Why should we all engage in the debate around AI legislation?</strong></h3><p>Perhaps the quote that best answers this question is from Caroline Piovesan:</p><blockquote>“We are debating the impacts of technology on our values to ensure our creations augment the kinds of society we want to live in, not subvert them.”</blockquote><p>I truly embrace the idea of assisting people with taking a humanistic view of technology. The physical world has proven to be severely damaged and unwelcoming for marginalized groups. AI offers us a distinctive opportunity to construct a digital realm capable of blending with and reshaping the physical world, potentially ushering in profound transformations. These shifts may either exacerbate existing issues or serve as a means to address some of them. The resemblance of this emerging digital-physical world will largely hinge on the leadership, values, and beliefs we embrace at this pivotal moment.</p><h3><strong>What exactly are we debating?</strong></h3><p>We are currently engaged in a debate regarding the minimal and essential legislative prerequisites to ensure that AI-based technology contributes to the kind of society we aspire to live in. From a legislative standpoint, I believe that there are four fundamental requirements for any organization providing AI services to society:</p><ol><li><strong>Legal Compliance</strong>: You must adhere to all relevant laws and regulations (i.e., general and sector-specific laws)</li><li><strong>Harm Prevention:</strong> You should make every effort to mitigate known risks and prevent harm</li><li><strong>Effective Remediation:</strong> In the event that harm occurs, you must respond swiftly and effectively to remediate it</li><li><strong>Accountability:</strong> You are responsible for the adverse consequences of your business operations and decision-making processes.</li></ol><p>When framed in these terms, these requirements are both straightforward and reasonable. They are, in fact, foundational principles applicable to businesses across various industries. In this context, the primary objective of legislative tools is to ensure the enforcement of these fundamental prerequisites. I also call out a desirable characteristic of legislative answers to this problem that is often forgotten:</p><blockquote>AI is a great tool that could assist humanity in improving ethical standards. Hence, legislation should also recognise and protect the potential of using AI to uplift ethical baselines and solve existing harms in our society.</blockquote><p>While one may tweak definitions or add extra requirements, there’s nearly a consensus on the understanding of why and what we’re debating. However, significant disagreements arise when we delve into the question of HOW to transform these abstract requirements into practical guidelines and specifications for organizations. I believe professionals with a technological background can make the most significant contributions at this juncture in the debate.</p><h3><strong>How can we translate essential legislative prerequisites into tangible practices and specifications?</strong></h3><p>Numerous issues come to light when we dissect the aforementioned prerequisites and delve deeper into certain conditions. What exactly constitutes “every effort to mitigate risks”? What are the “known risks”? Who has the authority to determine what is deemed acceptable? For a legislative framework to become effective and beneficial for society, we must define such matters in a way that minimizes ambiguity and uncertainty in their interpretation.</p><p>In sectors like pharmaceuticals, healthcare, and insurance, among others, a wealth of knowledge has been amassed over decades to address these very questions and meet these essential legislative requirements. This accumulation of expertise has ultimately resulted in positive societal impact and progress within their respective domains. As a result, risk management has emerged as the gold standard process for navigating through this complexity effectively. For this reason, almost everyone suggests making risk management the starting point for fulfilling these exact legislative requirements for AI.</p><p>Nonetheless, effectively implementing risk management in the realm of AI is far from straightforward. The challenge becomes evident in the absence of consensus among emerging regulatory frameworks worldwide. The inherent intricacies of AI, combined with its relatively young status as a field, its widespread utilization across nearly all sectors of the economy, and the historically limited application of risk management to AI, create significant hurdles when attempting to directly transfer knowledge from other domains into the domain of AI.</p><h3><strong>What makes risk management so different for AI?</strong></h3><p>My concise and simplified answer is: control of impacts. Risks and harms are inherent in all human activities and naturally accompany the use of technology to shape our environment. For instance, in the construction sector, we can readily identify risks linked to the use of tractors and heavy machinery, which are notably high.</p><p>So, why should legislators approach AI risks differently compared to the risks associated with tractors? The fundamental premise in our interaction with technology, up to the present day, has been that humans maintain substantial control over these tools. Adverse impacts arise either unconsciously or occasionally due to conscious and irresponsible actions by human operators. Except for defects and faulty operation, humans possess the capability to manage the impact of technologies like tractors, even unconsciously assessing reasonable limits of their application.</p><p>However, AI possesses unique characteristics that challenge human control over its impact. First, AI operates autonomously, allowing it to extrapolate behaviours beyond its intended design. Second, AI lacks consciousness, a system that gauges the appropriateness of its behaviours according to values, ethical norms, and common sense. Consequently, the level of control we currently exercise over AI is significantly more limited than what we typically have in other domains.</p><p>While these characteristics have implications for all four fundamental legislative requirements, I’ll narrow the focus of this post to requirements #2 and #3. These are areas where technical experts can make substantial contributions, as requirements #1 and #4 extend beyond the typical boundaries of IT knowledge.</p><h3><strong>What makes prevention and remediation of harm challenging for AI?</strong></h3><p>The primary objective is always to prevent adverse events from occurring. Ultimately, what truly matters is the impact we have on individuals, organizations, communities, and society. However, if we concentrate solely on monitoring impacts before taking action to mitigate them, as stipulated in requirement #3, we run the risk of paying the price for allowing negative occurrences to unfold. This is why the risk management framework became a preferred approach for AI regulation worldwide. Risk is the chance of harm occurring under defined circumstances, whereas harm is an adverse outcome. By focusing on establishing reliable risk management, we’re targeting to build effective prevention.</p><blockquote>While both preventive and corrective strategies must coexist, prevention should take precedence. Achieving satisfactory levels of understanding and predictability that enable effective prevention and remediation for AI proves to be a considerable challenge in practice.</blockquote><p>Figure 1 illustrates the key variables I’ve identified as the foundation for remediation and prevention requirements in the real world. I will employ technical terminology to outline this simplified worldview. In general, controlling the impact of AI requires a delicate blend of qualitative and quantitative indicators across four primary dimensions: Applied Knowledge, Visibility of Decisions, Visibility of Impacts, and Behavior Prediction.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Gpj3YlcEV648Xz7kZPr8QA.png" /><figcaption>Figure 1. Simplified worldview of requirements for ensuring effective prevention and remediation of AI-related harm.</figcaption></figure><p><strong>Applied Knowledge</strong> refers to a substantial repository of reusable, factual, and specific insights into actual impacts observed over time. Inspired by Mark Zuckerberg’s famous quote, “<a href="https://en.wikipedia.org/wiki/Move_fast_and_break_things"><em>Move fast and break things</em></a>,” the business strategy adopted years ago did not prioritize risk management from a socio-technological perspective. Consequently, our current pool of applied knowledge regarding AI risks remains limited.</p><p><strong>Visibility of Decisions</strong> pertains to the proper design, management, and documentation of intentions and decisions during the development of AI products. This aspect is still evolving in the context of this recent socio-technological lens. Notably, we have witnessed the emergence of various standards and frameworks related to AI decision-making, including <a href="https://www.iso.org/standard/77304.html">ISO/IEC 23894</a>, <a href="https://www.iso.org/standard/74438.html">ISO/IEC 23053</a>, and <a href="https://www.iso.org/standard/81230.html">ISO/IEC FDIS 42001</a>.</p><p><strong>Visibility of Impacts</strong> encompasses efforts to standardize metrics, methods, and methodologies for defining reliable quantitative indicators to assess the impact and behaviour of AI at various stages of its lifecycle. The widespread use of AI across multiple domains, scenarios, and sectors makes it challenging to establish assessment strategies that apply to all potential AI deployments.</p><p>Finally, <strong>Behavior Predictions</strong> refer to endeavours to establish ethical and reliable strategies to test AI with humans and simulate unintended or undesigned behaviours, especially when adverse outcomes are anticipated.</p><h3>How to bridge the gap for technical IT experts in the AI regulation debate?</h3><p>The conceptual framework outlined above can be a valuable resource for IT professionals seeking to develop well-informed perspectives on the execution of risk management in AI. The risk of implementing unfeasible or ineffective proposals arises when decisions regarding these dimensions are not aligned with a technical understanding of AI. For instance, an overemphasis on qualitative signals, which serve as proxies for actual impacts, or on quantitative indicators that introduce measurement uncertainty, can compromise the effectiveness of any legislative tool.</p><p>This conceptual framework also enables us to compare the main AI risk frameworks discussed worldwide. The <a href="https://artificialintelligenceact.eu/the-act/">EU AI Act</a> aims to extensively minimize AI-related harms by prioritizing prevention. Consequently, it mandates specific requirements to ensure high visibility of decisions and impacts.</p><p>In contrast, the <a href="https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/1176103/a-pro-innovation-approach-to-ai-regulation-amended-web-ready.pdf">UK approach</a> initially focuses on rapid remediation. It gradually enhances organizations’ capacity for prevention by standardizing applied knowledge and recommending monitoring standards, all while assisting organizations in improving the visibility of their decisions.</p><p>Meanwhile, the <a href="https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.100-1.pdf">US AI Risk Management framework</a> seeks to strike a balance between prevention and remediation by promoting standardized recommendations to enhance the visibility of decisions. Presently, the US framework suggests processes, practices, and procedures to guide AI&#39;s design, development, and project management to improve risk management and harm prevention.</p><p>So, what represents the most effective approach for striking the right balance between prediction and remediation for AI? From a technical standpoint, what should legislators prioritize and consider? I look forward to hearing your thoughts!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b899d1af8c0e" width="1" height="1" alt=""><hr><p><a href="https://medium.com/seek-blog/decoding-ai-regulations-striking-the-right-balance-for-preventing-and-remediating-harms-b899d1af8c0e">Decoding AI Regulations: Striking the Right Balance for Preventing and Remediating Harms</a> was originally published in <a href="https://medium.com/seek-blog">SEEK blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Generative AI in Online Employment Markets]]></title>
            <link>https://medium.com/seek-blog/generative-ai-in-online-employment-markets-deda38228c6d?source=rss-b8955024cf26------2</link>
            <guid isPermaLink="false">https://medium.com/p/deda38228c6d</guid>
            <category><![CDATA[generative-ai]]></category>
            <category><![CDATA[responsible-ai]]></category>
            <category><![CDATA[ai-ethics]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[online-recruitment]]></category>
            <dc:creator><![CDATA[Fernando Mourao]]></dc:creator>
            <pubDate>Tue, 13 Jun 2023 02:31:31 GMT</pubDate>
            <atom:updated>2023-07-18T04:38:33.200Z</atom:updated>
            <content:encoded><![CDATA[<h4>A Call for Responsible AI Practices</h4><p><em>In this blog post, </em><a href="https://www.linkedin.com/in/fernando-mour%C3%A3o-a40a5183/"><em>Fernando Mourao</em></a><em>, the Responsible AI Leader in Artificial Intelligence &amp; Platform Services (AIPS) at </em><a href="https://www.seek.com.au/work-for-seek/"><em>SEEK</em></a><em>, Melbourne, summarises a vibrant discussion held at the workshop on Generative AI in Online Employment Markets.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pk6KXtntIHI-siIwoClb5Q.jpeg" /><figcaption>Image from Pixabay, licensed under CCO via Wikimedia Commons</figcaption></figure><p>On 28 April 2023, SEEK proudly hosted the workshop on Generative AI in Online Employment Markets during the <a href="https://chatllm23.com/">ChatLLM23</a> symposium at the University of Sydney. Led by my colleagues <a href="https://www.linkedin.com/in/tao-cs-zhang/">Tao Zhang</a>, <a href="https://www.linkedin.com/in/terrence-szymanski/">Terrence Szymanski</a>, and <a href="https://www.linkedin.com/in/aditya-joshi-phd/">Aditya Joshi</a>, the workshop aimed to foster an in-depth discussion about the potential transformations and impact of Generative AI (GenAI) and, resultantly, AI-powered experiences in the labour market and society. At SEEK, we believe that hearing from experts, practitioners, researchers, and the affected parties is an effective way to approach and shift business with Generative AI responsibly.</p><p>The workshop was open to all pre-registered ChatLLM23 attendees and was structured as a 60-minute roundtable discussion session. Participants were invited to reflect on five discussion questions presented below, considering the different perspectives of some personas commonly found in online employment marketplaces, as shown in Figure 1.</p><blockquote>Discussion Questions:</blockquote><blockquote>1. What opportunities/experiences could Generative AI unlock?</blockquote><blockquote>2. What are the most concerning risks that could arise as a result?</blockquote><blockquote>3. What practices or behaviours should be required from organisations deploying Generative AI?</blockquote><blockquote>4. What would organisations, employers/job seekers behaviours or the labour market in Australia look like in 5 years?</blockquote><blockquote>5. What would be major milestones or transformation stages effectuated by Generative AI?</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mIMymAPAjgDEPPZ74cIthw.png" /><figcaption>Figure 1. A subset of the most common personas found in online employment markets.</figcaption></figure><p>A vibrant discussion allowed us to consolidate a discerning perspective about opportunities, risks, milestones, and best practices for adopting Generative AI in online employment marketplaces. The discussions focused on three main topics:</p><ol><li><strong>AI regulation</strong>: There is an urge for AI regulation, regulatory bodies and clear AI auditing procedures to underpin the delivery of cutting-edge innovation confidently and ensure the trustworthy, fair, transparent, and lawful use of GenAI in online employment.</li><li><strong>Positive impact</strong>: It is critical for organisations to efficiently exploit unique GenAI opportunities to impact the efficiency and effectiveness of businesses in online employment, reducing time and effort for talent acquisition, providing a seamless experience for job seekers and employers, and advancing personalisation in online employment systems.</li><li><strong>Hidden risks</strong>: Beyond the widely discussed GenAI-related risks, there are hidden risks and complex issues rising when adopting GenAI in employment markets, including ethics washing, the right to be forgotten by LLMs, consideration to people who speak minority languages, and inequality of access to GenAI as a new source of harmful bias in the labour market.</li></ol><p>Curious about an in-depth discussion on each of these topics? Check for updates below as we release a series of posts focusing on each topic&#39;s main developments:</p><ul><li><a href="https://medium.com/seek-blog/generative-ai-in-online-employment-markets-positive-impact-f80843b116a9">Generative AI in Online Employment Markets: Positive Impact</a></li><li><a href="https://medium.com/seek-blog/generative-ai-in-online-employment-markets-risks-and-practices-for-job-seekers-2654d5f8fa65">Generative AI in Online Employment Markets: Risks and Practices for Job Seekers</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=deda38228c6d" width="1" height="1" alt=""><hr><p><a href="https://medium.com/seek-blog/generative-ai-in-online-employment-markets-deda38228c6d">Generative AI in Online Employment Markets</a> was originally published in <a href="https://medium.com/seek-blog">SEEK blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Let’s talk about Black representation in RecSys leadership]]></title>
            <link>https://medium.com/@fmourao/lets-talk-about-black-representation-in-recsys-leadership-8f036ec4eaa7?source=rss-b8955024cf26------2</link>
            <guid isPermaLink="false">https://medium.com/p/8f036ec4eaa7</guid>
            <category><![CDATA[bias]]></category>
            <category><![CDATA[recsys]]></category>
            <category><![CDATA[representation]]></category>
            <dc:creator><![CDATA[Fernando Mourao]]></dc:creator>
            <pubDate>Tue, 20 Sep 2022 20:15:14 GMT</pubDate>
            <atom:updated>2022-09-20T20:15:14.523Z</atom:updated>
            <content:encoded><![CDATA[<p><em>This post discusses Black representation among the co-chairs and keynote speakers at the world’s leading conference on Recommender Systems. The aim is to inspire the RecSys community to investigate and reflect more deeply on historical biases unintentionally perpetuated.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*UTzMUnX1BOWWZMrk5imYwA.jpeg" /><figcaption><a href="https://commons.wikimedia.org/wiki/File:Leadership-picture.jpg"><em>rozeek.pk</em></a><em>, Attribution, via Wikimedia Commons</em></figcaption></figure><p>Recommender Systems (RecSys) have been reshaping humanity’s behaviour in so many ways that society has realised social and ethical principles should be primary requirements for these tools [1]. Concepts such as ‘Responsible by design’ or ‘ethical RecSys’ have become part of a worldwide technological, political and business agenda. It is a human-centred movement that aims to put humans and their principles in the control of RecSys. A primary goal of this movement is to promote human values, such as diversity and inclusion. When advocating for ethical RecSys, we should lead by example. When leaders, who guide this movement, do not practice it, the discourse may become ineffective and cynical. Based on my biased perception of reality, however, I hypothesise that leadership in RecSys is still not diverse and inclusive enough when considering racial justice.</p><p>As a Black researcher with over a decade of experience and active engagement with the RecSys community, I’ll focus my reflection on the participation of Black researchers in leading discussions within this community. The following findings might be equally true for other demographic groups historically marginalised. Still, to avoid being naive, I’ll focus on a context in which I have extensive lived experience. To explore my hypothesis, I designed a small survey study, as depicted in the figure below.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lUWOrKbwxdD2QU0XxHGyQQ.png" /></figure><p>I manually collected face photos of all 69 distinct keynote speakers and general co-chairs from all <a href="https://recsys.acm.org/">RecSys conference</a> editions, spanning 15 years. I gathered images with a minimum of 200x200 px quality publicly available on the RecSys official website or on each person’s social media and personal websites. Then, I asked 20 participants who self-identified as south-American, north-American or African Blacks to identify the person in each photo as being Black according to their personal beliefs about what determines belonging to this race socially. The image below summarises the result of this small survey study.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kaeZtL2PeHuiZA4OzvMSBQ.png" /></figure><blockquote><em>A small survey study reinforces the hypothesis that no Black researcher was ever a general co-chair nor a keynote speaker of a RecSys conference, the main worldwide conference on recommender systems research.</em></blockquote><p>Before discussing the results, it is crucial to highlight the limitations of this preliminary analysis. First, besides the social/collective aspect, race identification has a significant individual/subjective component [2]. A broader view of this question requires asking each of the 69 keynote speakers and general co-chairs their self-identified race. Second, the sample size is far smaller than the minimum value required to define a statistically significant sample [3]. Third, a single-face photo might not be enough to determine someone’s race. Several factors influence the social perception of belonging to a race. Finally, the quality of the pictures also impacts the results since I used the original images, which present different brightness, backgrounds and angles. That said, I found that none of the 69 people were identified as Black by the majority of the participants. It reinforces the hypothesis that no Black researcher was ever a general chair nor a keynote speaker of a RecSys conference, the main worldwide conference on recommender systems research. Hopefully, this result might inspire researchers to investigate this hypothesis more extensively. Representativity impacts everyone and is a priority for marginalised minorities. Indeed, 95% of the participants said they feel more represented when identifying Blacks in leadership positions.</p><p>Whereas what configures fairness for this scenario is unclear and subject to vigorous debate, if this result points out a naked truth, it would demonstrate an unintentional bias perpetuating in the RecSys leadership. It is implausible that there has never been a Black researcher in the past 15 years who has worked on a topic relevant enough for the RecSys community to be honoured as a general co-chair or keynote speaker. This illustrates a concrete situation of how unintentional biases can be present, despite the continuous efforts over the past years to improve diversity and inclusion goals in the Recs community. Repeating here the conclusion from <a href="https://medium.com/@fmourao/what-is-your-main-concern-for-the-future-of-rai-933b3f11a222">my previous post</a>:</p><blockquote><em>We might be inviting minorities to the party but setting up a table outside the saloon and letting them watch all the fun from a big screen. Allowing them to join the party is not enough. Inclusion and diversity require sharing the microphone with everyone!</em></blockquote><p>Therefore, the question that remains is: are our current initiatives enough to ensure the conditions for inclusive, effective and fast progress on responsible RecSys?</p><h3><strong>References</strong></h3><p>[1] Milano, S., Taddeo, M., &amp; Floridi, L. (2020). Recommender systems and their ethical challenges. <em>Ai &amp; Society</em>, <em>35</em>(4), 957–967.</p><p>[2] Banks, J. A. (2012). Researching race, culture, and difference: Epistemological challenges and possibilities. In <em>Handbook of complementary methods in education research</em> (pp. 773–794). Routledge.</p><p>[3] Taherdoost, H. (2017). Determining sample size; how to calculate survey sample size. <em>International Journal of Economics and Management Systems</em>, <em>2</em>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8f036ec4eaa7" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>