<?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 Jenna Niven on Medium]]></title>
        <description><![CDATA[Stories by Jenna Niven on Medium]]></description>
        <link>https://medium.com/@jennaniven?source=rss-77f92ddbb1e0------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*i3Bj0kZRO7EaGYdUbgLJ3w@2x.jpeg</url>
            <title>Stories by Jenna Niven on Medium</title>
            <link>https://medium.com/@jennaniven?source=rss-77f92ddbb1e0------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 15:04:17 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@jennaniven/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[Designing The Next Wave of Products: A UX Perspective on Agentic Systems]]></title>
            <link>https://medium.com/@jennaniven/designing-the-next-wave-of-products-a-ux-perspective-on-agentic-systems-3ad0238c50cd?source=rss-77f92ddbb1e0------2</link>
            <guid isPermaLink="false">https://medium.com/p/3ad0238c50cd</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[design]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[ux-design]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Jenna Niven]]></dc:creator>
            <pubDate>Mon, 10 Nov 2025 15:07:00 GMT</pubDate>
            <atom:updated>2025-11-10T15:07:00.156Z</atom:updated>
            <content:encoded><![CDATA[<p>The origins of this article actually start by trying to solve a different problem. So let’s start there to give you a bit more context behind what ultimately become a set of technical recommendations I believe will improve the UX of an agentic system.</p><p>I work on products that are used in building Autonomous Vehicles — think data mining, labelling, test management etc. Recently, I have come to realize that software for the development of AV (autonomous vehicles) needs to be designed and handled by a new, novel design paradigm because traditional approaches result in significant challenges that are particularly pronounced in this complex and rapidly evolving domain.</p><p>The current set of software tooling frequently fails to align with the real-world workflows and expectations of AV developers. Traditional methods attempt to pre-determine all user needs and journeys, which is flawed because the ever-evolving nature of AV development (which includes computer vision model development) introduces a significant amount of unknown unknowns — far exceeding those in more stable industries. This means static, predefined designs cannot adequately support ever changing, multi-dimensional workflows.</p><p>The diversity of users and their evolving needs also demand adaptability, which traditional, static groupings and rigid processes cannot provide. Tools need to accommodate a broad spectrum of needs and adapt as workflows evolve.</p><p>To address these issues, I proposed a new approach. This involves:</p><ul><li>Decoupling Capabilities from Presentation: Creating infrastructure that separates functionality from user interfaces to allow for flexible recomposition.</li><li>Dynamic Workflow Adaptability: Designing products UI with modularity and configurability to adapt to evolving workflows.</li><li>User-Centric Ecosystem: Moving away from static lists of user journeys and embracing user diversity and change.</li><li>Scalable Infrastructure for Custom Solutions: Building systems that enable tailored solutions without redundancy or fragmentation.</li></ul><p>I call the proposed approach a “Dynamic Multi-dimensional UX Framework” which uses an agentic platform that can construct products at runtime.</p><p>This system dynamically generates user interfaces and experiences based on the user’s Intent, Context, and Desired Action. This approach departs from traditional static applications with fixed layouts and workflows, and leverages language models to understand user needs and dynamically compose task-specific UI components on demand.</p><p>To prove out my theory that a “Dynamic Multi-dimensional UX framework” would actually solve our users problems, I built out a proof of concept. Here is a quick break down of my POC:</p><h3>The User Experience</h3><p>The UX blends a conversational interaction with an interactive, dynamic workspace.</p><p><strong>Conversational Prompt Interface: </strong>Users engage with the system through a persistent chat interface, submitting natural language prompts that describe their tasks or questions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fbfO5zOoZ6gSBsR0tAmWVw.png" /><figcaption>Conversational interface wireframe with empty workspace.</figcaption></figure><p><strong>Dynamic Canvas Workspace: </strong>The system responds by generating and updating a live Workspace, which displays interactive UI elements tailored to the current task. Each interactive workspace may include elements like dashboards, charts, filters, tables, and data viewers.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oFI6Tr9FlqhRFOC-QqV8cQ.png" /><figcaption>Workspace display wireframe with persistent conversation history.</figcaption></figure><p><strong>Persistent Session Context: </strong>Conversations and Workspaces are tied together through persistent session context, allowing users to ask follow up questions and continue their workflows seamlessly.</p><h3>Technical Info</h3><p>I used the following to help me build quickly:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9x-6TNQbXidGebhe2kK82w.png" /></figure><ul><li><a href="https://flask.palletsprojects.com/en/stable/">Flask</a>: A web framework for handling web requests.</li><li><a href="https://uvicorn.dev/">Uvicorn</a>: An ASGI server to run the Flask application.</li><li><a href="https://openai.com/api/">OpenAI API</a>: Provides language model services.</li><li><a href="https://github.com/facebookresearch/faiss">FAISS</a>: A library for efficient similarity search and clustering of dense vectors.</li><li><a href="https://pypi.org/project/langchain-core/">Langchain Core</a>: Provides core tooling for language models.</li><li><a href="https://pypi.org/project/langchain-community/">Langchain Community</a>: Offers additional community-contributed tools.</li><li><a href="https://alpinejs.dev/">Alpine.js</a>: A lightweight JavaScript framework for UI interactivity.</li><li><a href="https://www.webcomponents.org/">Web Components</a>: A native browser technology for creating reusable, encapsulated UI elements using custom HTML tags.</li></ul><p>Complex reasoning tasks demand a multifaceted approach, requiring retrieval of domain-specific knowledge, data fetching, and multi-step inference. While LLMs offer robust conversational capabilities, they need structured outputs to reliably power UI generation and tool execution.</p><p>In the POC, I implemented a hierarchical agent understanding and tool selection approach that outputs structured JSON data for use in subsequent reasoning and tool execution phases.</p><p>The system operates in two distinct phases, each consisting of several critical steps that work together to deliver a dynamic, context-aware user experience.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*MFv36RS5I_z0inYj" /></figure><h3>Phase 1: Understanding the User Query</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*noF0J1yPZHCXrao8pfp8ug.png" /></figure><h4>Intent and Context Extraction</h4><p>The system begins by analyzing the user’s prompt to determine their intent, context, and desired action. This is crucial due to the system’s broad functional scope. Understanding the user’s working environment ensures that the response is accurate and relevant.</p><p>There are 2 tasks involved in this step:</p><p><strong>1. Query Classification and Routing</strong></p><p>Query classification is broken up into an additional two components, each of which is critical downstream.</p><p>Context Container Classification:<br>Context refers to the working environment or nature of the data being interacted with. In AV development, users can engage with sensor data of a car drive, or derivatives of this (e.g., a small segment of that drive, or a frame from that drive etc). Context containers provide an additional layer of information regarding that data — such as test results for assessing the current performance of an event in a drive.</p><p>This classification informs display, data filtering, data fetching, and UI panel construction that are all subsequent steps.</p><p>Query Type Classification: <strong><br></strong>Queries are classified as Task or Question. Tasks request actions (displaying data, running processes) and updates the workspace UI. Questions seek information/explanations and answers return via chat, with no Workspace UI changes.</p><p>These classifications use pre-defined label classes to avoid non-deterministic LLM output and minimize failure risk. Responses are structured JSON objects, validated before proceeding.</p><p><strong>2. Tool Selection and Invocation</strong></p><p>Once classifications are complete, the system selects the appropriate Tool to fulfill the user’s request. Tools perform specific actions within the system, such as executing commands or file operations. The tools available for selection are constrained by the classification of the query type.</p><p>Task Tool Examples:</p><ul><li>Page Navigator: Navigates the user to a different data hierarchy display.</li><li>Perform Action: Executes a UX action on behalf of the user.</li><li>Data Visualization: Displays the data in a different way, like a chart or table.</li></ul><p>Question Tool Examples:</p><ul><li>Documentation Retrieval: Generates an answer to the user’s question from relevant technical documentation.</li><li>Data Analysis: Performs different types of analysis on the data. Finds insights or answers questions specific to the data.</li></ul><p>Each tool has required inputs and structured outputs and is made up of one or more sub-tasks. An LLM is used to choose the most appropriate tool and responses are structured JSON objects for the next step.</p><h4>Page Type and Data Hierarchy Selection</h4><p>Once intent and context are established, the system identifies the appropriate page type and data hierarchy. Because our data schemas vary depending on the user’s working context, this step ensures that the generated UI reflects the correct structure and provides the right data for the task at hand.</p><p>To compose the display needed for the user to complete their task, there is again information that needs to be extracted from the user’s query, combined with prior classifications, to produce structure output.</p><p>Each Context Container Classification contains multiple Page Types, defined by a manifest that describes the data hierarchy. To be able to construct the correct UI, we use an LLM to extract this information (Context Container Type and Relevant Data Hierarchy) from the user query.</p><p>User Query Example: <br>“Show me all the clips for the edge case ‘obstructed left turn in the snow’.”</p><p>Context Container Type: <strong>Individual </strong>Edge Case (since the user is requesting details about a specific edge case).</p><p>Relevant Data Hierarchy: <strong>List</strong> of Clips (multiple clips of a drive that contain an obstructed left turn in the snow)</p><h3>Phase 2: Generating the User Interface</h3><h4>Data Retrieval and Proto Introspection</h4><p>After understanding the query, the system needs to retrieve the required data to fulfill the request. Depending on the nature of the query, this may involve fetching data from and end point, querying a database, or retrieving documentation from a separate repository. If needed, the system dynamically inspects available gRPC services and proto messages to build precise queries.</p><h4>Rendering the Dynamic UI in the Workspace</h4><p>Finally, the system uses the extracted parameters to compose the UI. The UI components (using Web Components) are dynamically instantiated and rendered in the user’s browser, producing an interactive workspace tailored to the task. Users can immediately interact with dashboards, charts, tables, or data explorers seamlessly integrated into their workflow.</p><h3>Key Architectural Considerations for Desired UX</h3><p>During the development of the POC, several architectural decisions had a significant impact on the user experience. This section documents those decisions as recommendations for future production implementation.</p><h4>Accuracy and Trust</h4><p>Getting things wrong is the quickest way to erode user trust. Errors in interpreting user queries, retrieving incorrect data, or generating misleading responses diminish confidence and discourage adoption. Accuracy must be prioritized to build and maintain trust.</p><p><strong>Recommendations:</strong></p><ul><li>Enforce structured outputs using formats like JSON to ensure LLM responses are consistent and machine-validated.</li><li>Implement output validation and correction layers to catch and address errors before tool execution or UI generation.</li><li>Introduce fallback mechanisms to gracefully handle uncertainties or failures, providing clear and actionable feedback to users.</li></ul><h4>Responsiveness and Latency</h4><p>A seamless experience is not just about reducing the number of steps, it also requires speed. Even the most efficient workflows lose value if users are waiting too long for results. The system must feel responsive and predictable.</p><p><strong>Recommendations:</strong></p><ul><li>Optimize query processing and tool invocation by segmenting responsibilities between LLM-driven reasoning and lightweight algorithmic execution, reducing unnecessary processing delays.</li><li>Leverage session context and memory management to avoid redundant operations and minimize repetitive calls, particularly for multi-step workflows.</li><li>Implement progressive rendering and non-blocking UI patterns, using loading indicators and partial results to keep users informed and engaged while background processes complete.</li></ul><h4>Capability and Coverage</h4><p>If the system lacks the capability to handle a user’s core tasks or answer key questions, adoption will stall. Breadth of functionality is essential for positioning the platform as an indispensable tool in daily workflows.</p><p><strong>Recommendations:</strong></p><ul><li>Design a scalable and extensible tool framework, allowing for the reliable execution of a broad set of common tasks and queries through modular ‘Tools’.</li><li>Integrate with essential data sources to cover the highest-priority known use cases and data needs.</li><li>Prioritize currently known tasks and workflows.</li></ul><h4>Systematic Design</h4><p>Agentic systems don’t remain static — they adapt as new data, workflows, and tools are introduced. Designing for this level of dynamism means shifting from fixed interfaces to components and patterns that can reconfigure themselves while still feeling cohesive and intentional.</p><p><strong>Recommendations:</strong></p><ul><li>Focus on flexible building blocks that can be assembled dynamically while maintaining the product’s visual and interaction language.</li><li>Instead of hard-coding screens, define rules, structures, and context-aware layouts that the system can generate at runtime from metadata.</li></ul><p>In the evolving landscape of AI driven software products, we are moving past <em>software that waits</em> to <em>software that understands</em>. This new paradigm means that the way in which products are designed and built (ie the technical choices made throughout development) have even more of a monumental effect on the desired UX outcome. The future of UX includes agentic systems, and the designer’s role is not just to design interfaces, but to design dynamic, multi-dimensional UX experiences.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3ad0238c50cd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Design Is The New Oil]]></title>
            <link>https://medium.com/@jennaniven/design-is-the-new-oil-61cb173e37ee?source=rss-77f92ddbb1e0------2</link>
            <guid isPermaLink="false">https://medium.com/p/61cb173e37ee</guid>
            <category><![CDATA[engineering]]></category>
            <category><![CDATA[ai-applications]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[ai-safety]]></category>
            <dc:creator><![CDATA[Jenna Niven]]></dc:creator>
            <pubDate>Sun, 23 Jul 2023 18:03:22 GMT</pubDate>
            <atom:updated>2023-07-23T18:03:22.350Z</atom:updated>
            <content:encoded><![CDATA[<p>I recently presented at the <a href="https://summit.supermomos.com/">SUPER AI ML SUMMIT</a> in NYC, and wanted to share the content I spoke through more broadly. Enjoy!</p><p>In 2006, Clive Humby, a British mathematician, coined the term ‘Data Is The New Oil’.</p><p><em>“Data is the new oil. Like oil, data is valuable, but if unrefined, it cannot really be used. It has to be changed into gas, plastic, chemicals, etc. to create a valuable entity that drives profitable activity. So, must data be broken down, analysed for it to have value.”</em></p><figure><img alt="Duarte, F, April 3, 2023, Amount of Data Created Daily (2023)" src="https://cdn-images-1.medium.com/max/1024/1*1vNRbHl19XtaVAXxpb3TKw.png" /><figcaption><strong>Duarte, F, April 3, 2023, Amount of Data Created Daily (2023)</strong></figcaption></figure><p>Data is the lifeblood of a machine learning model. It was the explosion of data (and compute) that took machine learning out of the research labs, and enabled companies with easy access to this data, to create the game changing foundation models we see all over the media today.</p><p>But today I am going to talk about something that has now become more important to machine learning than data….and that’s design.</p><p>Firstly, let me clarify what I mean by ‘design’. Design, in the context of this talk, is the process of creating a <strong>useful</strong>, <strong>usable</strong> and <strong>safe product or service</strong>, by making thoughtful decisions that have the <strong>end-user experience at the center</strong>.</p><p>There is no shortage of information out there educating designers on the fundamentals of machine learning. Most typically they focus on the different methods of training a model — supervised, un-supervised, semi-supervised etc. This is good information that feeds into the ideation phase of the design process.</p><p>However, for the next generation of products and services leveraging AI, to truly make a meaningful impact on our lives, design needs to understand and contribute to the entire ML development cycle.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8TVyjNSjzYEhiZOfdYRdmw.png" /></figure><p>The typical design and engineering engagement model involves design coming up with an idea or concept, and iterating on that until we reach a solution that we can hand off to the engineering team to build.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tWLxdv1Q31ElCViKizqDsQ.png" /></figure><p>Design then enters back into the conversation towards the end of the development process to make any adjustments based on technical feasibility, user test, visual QA etc.</p><p>In this world, design doesn’t typically get into the weeds of each and every technical decision. Angular versus Lit? Server vs client? These are primarily engineering decisions that have little impact on the end user.</p><p>This is very different in the development process of products and services that are powered by AI. We call this the Software 2.0 development cycle.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vfYqjFAT8j_K0F_l5vLE1w.png" /></figure><p>The term Software 2.0 was coined by Andrej Karpathy in 2017 and it explains the difference between the development process of classical software versus ML software.</p><p><em>“The “classical stack” of Software 1.0 is what we’re all familiar with — it is written in languages such as Python, C++, etc. It consists of explicit instructions to the computer written by a programmer. By writing each line of code, the programmer identifies a specific point in program space with some desirable behavior.” </em><a href="https://karpathy.medium.com/software-2-0-a64152b37c35"><em>Source</em></a></p><p>In software 1.0, code drives the development of the application — if needed, one can inspect the code to understand the logic behind decisions that have been made and make change retrospectively pretty easily. In contrast in software 2.0 its data that drives the development of the application. And the models produced from this data, will not give you any insights into the logic behind decisions that have been made along the way.</p><p><em>“In contrast, Software 2.0 is written in much more abstract, human unfriendly language, such as the weights of a neural network. No human is involved in writing this code because there are a lot of weights (typical networks might have millions), and coding directly in weights is kind of hard” </em><a href="https://karpathy.medium.com/software-2-0-a64152b37c35"><em>Source</em></a></p><p>In the software 2.0 development cycle, each stage is a juncture where important decisions are being made that have a dramatic impact on the end user. It’s imperative, if we are to successfully build the next generation of useful, usable and safe applications of AI, that these decisions are made with the end user at the center. That these decisions are designed.</p><p>If we take a more in-depth look at the process involved in developing a machine learning model, we can see it’s very complex — and while there are many points within this process that deserve designer engagement, today I will highlight just 3.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*T6_rf99QD9mZ7-hFr6BNEw.png" /></figure><h3>Data Processing</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dD5luDZQNDj-e7ZobbK-4w.png" /></figure><p>Data defines a large part of the outcome of an ML model. You’ve probably heard of ‘garbage in, garbage out’. But there is a particular stage within the data processing pipeline that has a large impact on the end user experience — it’s called ‘feature selection’.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HDE224Goa_QctdMjw63ebw.png" /></figure><p>To train a model, we collect enormous amounts of data to help the machine learn better. Usually, a large portion of the data collected is noise, and some might not contribute to the training of our model. The model may also learn from irrelevant data in the dataset and become inaccurate. Feature selection is the process of defining what attributes of your data you will use to train your model.</p><p>So why is this important for design?</p><p>Consider this example:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zTrNElOwPQR4U4b_BqykIg.png" /></figure><p>A few years ago in 2019, Apple and Goldman Sachs were accused of creating an ML powered credit card application algorithm for the Apple Card that was sexist. Users, including Steve Wosniak, noticed that it was offering smaller lines of credit to women than to men, even though from a credit worthiness perspective they were relatively the same.</p><p>According to <a href="https://medium.com/@vla6">Valerie Carey</a>, the speculation of reasons tended to fall into 2 major categories:</p><blockquote>Women have lower income because of workplace discrimination, and credit limits are linked to income; therefore, women are assigned lower limits.</blockquote><blockquote>The Apple Card algorithm’s training data contained information about shopping patterns, which were a proxy for gender, and led to stereotyping of female applicants. <a href="https://medium.com/towards-data-science/fairness-metrics-wont-save-you-from-stereotyping-27127e220cac">Source</a></blockquote><p>Let’s look at an example…</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*awbq9VECaaedwNshaDfijA.png" /></figure><p>If we look at this data, we can assume that the ground truth data was primarily dependent on the annual income.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JWgMAEZ4OPjcmRaaZrWedw.png" /></figure><p>But the training data used to train this model had additional features added — they added spending patterns.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nE8hxY8SgNKJ7cswQMEFoQ.png" /></figure><p>So we know the ground truth had already been added given the annual income — so annual income = lower limit.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MrcN1P-UfJH0TKizzNbR4A.png" /></figure><p>But unfortunately the model learned that instead of annual income being the major factor, spending patterns were. And we can see from this data that, even though the total spending per month was the same, there were particular patterns in the categories of spending that inferred whether a data point was either male or female.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*f6eNikd66nyiGegItQoM7g.gif" /></figure><p>So the spending patterns features became a proxy for gender, so from the models perspective we can just assume that the spending pattern were a gender feature.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*S0UR7X9qchqsjtE2On8Ivg.gif" /></figure><p>So now when we look at unseen data, we can see that annual income is high, however the spending patterns are representative of someone who is female. And because of this, the predicted credit limit is low.</p><p>There are a few statistical ways, called ‘fairness metrics’, of choosing features to reduce the likelihood of issues like this, but as stated by Valerie Carey in her article <a href="https://medium.com/towards-data-science/feature-choice-and-fairness-less-may-be-more-7809ec11772e">Feature Choice and Fairness: Less May Be More</a>, these are not always effective.</p><p>Feature selection needs human judgment and more specifically, needs to involve a <strong>designer who can extrapolate the impact of certain feature choices, to the impact on the end user</strong> (and society!)</p><p>Now going back to designs focus — creating useful, usable and safe products with the user at the center.</p><h4>Useful</h4><p><strong>If we ask ourselves about usefulness — will including or excluding certain features, effect the intended purpose of the product or service?</strong></p><p>If we ask ourselves about usefulness — Will including or excluding certain features, effect the intended purpose of the product or service?</p><h4>Usable</h4><p><strong>Will including or excluding certain features, impact whether a user is willing or able to use the product?</strong></p><p>You can assume that if we were to continue to use the spending habits feature, that a large part of our user base — females — may not want to use our product.</p><h4>Safe</h4><p><strong>Do any of the features contain information about a protected group — explicitly or by proxy? </strong>How does including these features impact the end user?</p><p>Well we know in the Apple Credit Card case, that it actually ended up discriminating against an entire gender and perpetuating an existing societal bias.</p><h3>Model Development and Experimentation</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tVMnMXFrD8AhPwPMZm1P9g.png" /></figure><p>Now moving along in the development process, we get to ‘Model Development and Experimentation’. This is where we make choices on things like the model architecture, hyperparameters, training datasets and evaluation metrics. ML engineers experiment with different combinations of these things until they find the combination that produces the best model that then can be promoted to be ‘productionized’ and scaled up.</p><p>Now if you are up-to-date with the latest design/AI blogs I mentioned before, you would be familiar with reinforcement learning.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QY4qFFDRv2jJFrXgK1S65A.png" /></figure><p>In reinforcement learning (RL), training a model consists of two stages: First, the programmer defines the <strong>objective</strong>; then an optimisation algorithm tries to find the best possible solution based on the reward function.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Lzwn1xlKhQR4l4j049wjog.png" /><figcaption><strong>Ron Armstrong from Helena, MT, USA — HMKC Spring 2007 Agility Trial</strong></figcaption></figure><p>Imagine if you were training your dog to compete in an agility competition with the following course:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eSgBdVgaM1uby0atZKXLAw.png" /></figure><p>If I only set the objective to ‘complete the circuit in the least amount of time’ and only reward the dog at the end of the circuit, it may learn that the best way to achieve that goal is to ignore all the agility obstacles and run right to the finish line.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WevKl6NsvYwyeph-jVFQ4Q.png" /></figure><p>This is an example of where the objective’s definition may not accurately capture the human’s intention. This is called a “misaligned” objective/reward function. <a href="https://towardsdatascience.com/how-learning-reward-functions-can-go-wrong-6e794e42f4fc">Source</a></p><p>Now if I update the objective to “complete each obstacle, in order, without faults, in the least amount of time” we are much more likely to get the result we are looking for.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zBYCLkghZsWMSyFyIVxntA.png" /></figure><p><strong>Designers need to understand, and actively contribute</strong> to, the way a machine learning model learns <strong>to make sure the objective/reward function is aligned with the intended user outcome</strong>.</p><h4>Useful</h4><p><strong>Looking at design again — is it useful? By choosing a specific objective and reward function, will the product still provide value to the user? Are we solving the right problem?</strong></p><p>In the case of our show dog — we can see that by setting the objective function to just speed, the dog would not be very useful for me to collect the competition prize.</p><h4>Usable</h4><p><strong>Can a user interact with the product or service given the objective function chosen?</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/645/1*D5gRKgm6vqlJqj-4zO2IHg.gif" /></figure><p>Hypothetically, if a robotic arm that was built to assist in remote surgeries, was trained on the objective to perform E2E surgery at the highest precision — but when deployed the movement of the robotic arm was too fast or completed the steps in a nonsensical order, making it impossible for the surgeon to take control — this would not be usable.</p><h4>Safe</h4><p><strong>Will the learned behavior from the objective function, put anyone in harms way?</strong></p><p>The classic example of this is the paperclip maximizer theory, where philosophers have speculated that an AI tasked with creating paperclips might cause an apocalypse by learning to divert ever-increasing resources to the task, and then learning how to resist our attempts to turn it off.</p><h3>System Development and Debugging</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YU8TQlRJ7JoWiUcT5IykDg.png" /></figure><p>Let’s now skip to ‘System Development and Debugging’. In theory, we now have a robust model that doesn’t introduce unintended bias, and performs as designed. The next step in to now embed that model into some sort of application that delivers the predictions to the end user.</p><p>This is probably the most obvious place where design is integral — but again, there are some technical details that are needed to be ‘designed’.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pixZRLsOoRZfcVQcREbKEg.png" /></figure><p>In a classification task, when a model makes a prediction — the accuracy is determined by assessing whether the prediction is either a ‘True Positive’, ‘True Negative’, ‘False Positive’ or a ‘False Negative’. There are many metrics you can use to asses the performance of a model — but today I am going to focus on the concepts of precision and recall.</p><p>Precision is ‘what proportion of positive identifications was actually correct?’ and recall is ‘what proportion of actual positives was identified correctly?’.</p><p>Here is an example….Imagine we’ve just trained a model to identify whether a patient is having a heart attack.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RRD0sgv9tCP_GvH6y8t-EA.png" /></figure><p>Our model made 15 total predictions of whether a patient is having a heart attack or not.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QTB_BEGAcGzDd9peJXi6Ig.gif" /></figure><p>The number of times the model PREDICTED that the patient WAS having a heart attack is 10. But was only right in 6 cases, so our precision is 6 out of 10: 60%.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yIIxp8wtvYYq4w20nndRcQ.gif" /></figure><p>Now the number of times the patient was ACTUALLY WAS having a heart attack is 9. And the amount of times the model was able to get it right was 6, so our recall is 6 out of 9 = 67%.</p><p>The important thing to know is <strong>there is an inverse relationship between precision and recall</strong>. You can only optimize for one.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yVhKVWX7Eu3Y6cFftixJzg.png" /><figcaption><a href="https://pair.withgoogle.com/guidebook/patterns#precision-and-recall-tradeoffs">https://pair.withgoogle.com/guidebook/patterns#precision-and-recall-tradeoffs</a></figcaption></figure><p>By optimizing for precision you are serving less predictions, but highly likely to be correct. And by optimizing for recall, you are serving more predictions, but less likely to all be correct.</p><p>Each time we create a product or service that uses classification, <strong>design must assess the impact of the precision/recall tradeoff on the end user experience so the right optimization can be made</strong>.</p><h4>Useful</h4><p><strong>If you are optimizing for precision, and only display a reduced amount of results — will the product still be useful?</strong></p><p>If you were developing an AI stylist, but the product only ever returned 1 option — would this be useful?</p><h4>Usable</h4><p><strong>On the other hand, if you were to optimize for recall, would the large quantities of results at varying degrees of confidence make the product non-usable?</strong></p><p>If it is best to focus on high recall — can we consider design techniques such as progressive disclosure to make sure the product does not cause cognitive overload?</p><h4>Safe</h4><p><strong>What happens to the user if you display a False Positive or a False Negative?</strong></p><p>We need to understand the impact of showing false results and decide whether the harm done by showing a False Negative outweighs the harm done by providing a False Positive.</p><p>Is it better to treat a patient who is not having a heart attack, versus not treating a patient who is?</p><p>These are just a few examples of decisions within the ML development cycle that clearly make a huge impact on end-users, and there are plenty more to look at.</p><p>Data may have gotten us this far — but it’s <strong>design that will make sure we are solving a true user need</strong>.</p><p>It’s <strong>design that will make sure the solutions are easily understood and delivered in a way that provides value</strong>.</p><p>It’s <strong>design that will make sure we’re developing these solutions to be a benefit to society, not a detriment</strong>.</p><p>Design is the new oil.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=61cb173e37ee" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Limitless Creativity: How to Design in an AI World]]></title>
            <link>https://becominghuman.ai/limitless-creativity-how-to-design-in-an-ai-world-30cc7e26f28d?source=rss-77f92ddbb1e0------2</link>
            <guid isPermaLink="false">https://medium.com/p/30cc7e26f28d</guid>
            <category><![CDATA[creativity]]></category>
            <category><![CDATA[ai-creativity]]></category>
            <category><![CDATA[computational-design]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Jenna Niven]]></dc:creator>
            <pubDate>Sun, 23 Jun 2019 16:39:17 GMT</pubDate>
            <atom:updated>2019-07-03T13:30:45.635Z</atom:updated>
            <content:encoded><![CDATA[<p>Lately I have been asked a lot about my opinion on AI and Creativity, so I have finally translated my 2018 SXSW talk into a post.</p><p>— — — — — —</p><h4>I presented this talk at SXSW 2018. If you would like to listen rather than reading — <a href="https://youtu.be/qvaoPCPrI08">here it is on YouTube!</a></h4><p>Before I get into the nitty gritty, I want to give you a bit of background about how I ended up so obsessed with AI.</p><p>Long story short, it started when I went to university to do a bachelors degree in design, and there was an option so do an engineering unit.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/595/0*oekLKPdVeazaoM8T" /></figure><h3>Trending AI Articles:</h3><blockquote><a href="https://becominghuman.ai/cheat-sheets-for-ai-neural-networks-machine-learning-deep-learning-big-data-678c51b4b463">1. Cheat Sheets for AI, Neural Networks, Machine Learning, Deep Learning &amp; Big Data</a></blockquote><blockquote><a href="https://becominghuman.ai/data-science-simplified-principles-and-process-b06304d63308">2. Data Science Simplified Part 1: Principles and Process</a></blockquote><blockquote><a href="https://becominghuman.ai/getting-started-with-building-realtime-api-infrastructure-a19601fc794e">3. Getting Started with Building Realtime API Infrastructure</a></blockquote><blockquote><a href="https://becominghuman.ai/bursting-the-jargon-bubble-deep-learning-22da43caa525">4. Bursting the Jargon bubbles — Deep Learning</a></blockquote><p>So last year of uni, I scored a job as a web developer and my engineering days began! But it turns out I am a really shit developer….so I moved back to design and well, the rest in history. But the 4 years I spent as a web developer were, retrospectively, actually incredibly important.</p><p>I learnt HTML, CSS, (AS2?!) and PHP and they haven’t really changed much since. So I am fascinated with figuring out how things works and I can backwards engineer pretty much anything.</p><p>This has been an enormous advantage for my creativity. Being able to understand how and why technology works the way it does, allows me to really push the boundaries of my creativity.</p><p>In comes AI and it’s the same. Understanding the fundamentals of AI tech, has allowed me to create products and services that have literally never been able to be done before.</p><p>But that got me thinking…every time we have invented a new technology — it has enabled us to do things that were not humanly possible before. It happens all the time. Us humans, are constantly trying to find new ways to extend our limitations!</p><p>From horsepower extending the limitations of the human muscle. To the steam engine extending the limitations of animal power. And the internet extending our ability to exchange information….</p><p><strong>AI is no different. It allows us to extend the limitations of our mind and hence our creativity.</strong></p><h4>Impact of Technology Over Time</h4><p>Throughout history, when we invent these new technologies that allow us to overcome limitations, we as a society change.</p><p>Look at the chart below — there was something happened here that catapulted human social development….what do you think it is? Us standing up right? Or maybe fire?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*D1RA4TkjujCzE7lGl-T_4Q.png" /><figcaption>Brynjolfsson, E, McAfee, A (2016) The Second Machine Age: Work, Progress, and Prosperity in a Time of Brilliant Technologies, Norton, W. W. &amp; Company, Inc.</figcaption></figure><p>It was the steam engine. As time goes on, the things we create have the potential to have a massive impact on society.</p><p>At a more granular level — below is chart that shows the distribution of jobs by sector.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mGcLq6CN3pRYG0b4sqGdkg.png" /><figcaption><a href="https://www.minnpost.com/macro-micro-minnesota/2012/02/history-lessons-understanding-decline-manufacturing">https://www.minnpost.com/macro-micro-minnesota/2012/02/history-lessons-understanding-decline-manufacturing</a></figcaption></figure><p>We can see when we implemented steam power, we saw a shift in jobs from agriculture to industry. When the technological revolution hit, we saw a shift of jobs from industry to services - essentially moving from physically taxing jobs to cognitively lead jobs.</p><p>AI is tipped to be the next technological revolution, and that’s where it gets interesting. AI allows us to overcome the limitations of our mind and will ultimately disrupt these cognitively lead jobs.</p><p>In fact, we have already seen this begin to happen. In a recent study a computer algorithm correctly diagnosed 90% of lung cancer cases presented to it, while human doctors only had a success rate of about 50%.</p><p>AI’s ability to be able to process so much information, identify hidden patterns or perform repetitive tasks faster and more accurately — make it a far better choice than a human alone.</p><h4><strong>Now let’s take a look at creativity….</strong></h4><p>Creativity is illogical and abstract. This has been ingrained in us with myths and stories like the story of ‘Eureka’.</p><p>Archimedes, the 16th century scientist, was tasked with finding out if the goldsmith had stolen any gold from the king when asked to make his crown.</p><p>He racked his brain day and night trying to figure out how he could tell if there were any other metals mixed into the gold crown. He needed to measure volume or density, not weight.</p><p>He finally gave up and ran himself a bath. But when he stepped in he noticed that the water spilled out of the sides. ‘Eureka eureka’ he scream (apparently then ran down the street naked…) In that moment, he realized that by measuring the amount of water displaced — he could measure the objects volume.</p><p>The moment he turned off his mind and sat in the bath, was the moment of great illumination.</p><p>And the same thing goes when we look at more relevant creative works — like product design, music and even fine arts.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*gZ0GrLBAWgP-revI" /><figcaption>Brown, P (2016) Under Armour Changes the Footwear Game With Generative Design and 3D-Printed Shoes, Autodesk.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UtlGN7qsgWt97eB5hlb0KA.png" /><figcaption>Image: <a href="https://www.nextrembrandt.com/">https://www.nextrembrandt.com/</a></figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IoePPvz0koK8Sh6nlo7cXA.png" /><figcaption>Listen —&gt; <a href="https://www.youtube.com/watch?v=BR6bin0NZSE">https://www.youtube.com/watch?v=BR6bin0NZSE</a></figcaption></figure><p>If you were the creator, it would be pretty difficult for you to describe exactly how theses works were created. We think of creativity is an illusive thought process that sits in the subconscious. So it would be incredible difficult to recreate this with a computer right?</p><p>Well, maybe not….</p><p>All three of the creative works above were created WITH an AI system. And all three meet or exceed the capabilities of a human.</p><p>The idea that humans will always have a unique ability beyond the reach of machines, does not seem to be the case anymore. The creative industry is not immune to AI disruption!</p><h3>So how can AI systems produce these creative artifacts?</h3><p>Well you probably have heard about deep learning. It’s the main concept behind many of the AI capabilities like image recognition.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*V8zvrK85q-YACyaAcyTFmw.png" /></figure><p>A simple way of looking at it is - the computer looks at a huge amount of images. When it’s, show a new one, it breaks down into chunks and 1st tries to see if it recognizes any of the edges, based on all the other images it has seen before.</p><p>It then passes that knowledge along to the next layer which can calculate a probability, that those edges belong to - say a car. The information then gets passed to each layer, repeating the same process, and by the time it gets to the end it has a pretty good idea (high probability) that this image is a white car, driving down a highway at night.</p><p>The interesting thing about this, is that this process actually mimics the human brain. We feed an image into our brains via the retina, and information gets passed through layers of the neocortex until it has a pretty good idea what it is.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UFb75iQok-XTKWAfSIvZ2g.png" /></figure><p>So both of these methods — have the same set of rules it applies each time. The output is just different dependent on the input. So you could classify both of these as algorithms.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/608/0*V-Fkc2kmlGSfzMGg" /></figure><p>And this is where things get a little crazy. Consider this:</p><ol><li>If the brain is just a set of algorithms like a machine…</li><li>And algorithms are not affected by the materials from which the calculator is built. Whether an abacus is made out of wood or plastic — 1 + 1 still equals 2.</li><li>Does that mean an AI algorithm should be able to do everything a human algorithm can??</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/424/0*pXlOqAColGrtgUSB" /></figure><blockquote>“According to the life sciences, art is not the product of some enchanted spirit or metaphysical soul, but rather of organic algorithms recognising mathematical patterns.” Homo Deus, Yuval Noah Havari</blockquote><p>Who agrees with that statement?</p><p>When it comes to believing that non-organic algorithms can do whatever an organic algorithm can, I agree! It’s true that at the moment there are a lot of things that organic algorithms can do better than non-organic algorithms — but it’s just a matter of time.</p><p>Until a short time ago, facial recognition was a favorite example of something a baby could do but a machine would struggle. Today facial recognition algorithms are able to identify people far more effectively and efficiently than people.</p><p>But, when it comes to believing that creativity is just a set of algorithms — that’s where I’d have to disagree.</p><p>There are some things within the creative process that machines are better at, and others that are so completely illogical and abstract that they could never be replicated by an algorithm.</p><h3>So where does that leave us in our quest to overcome the limitations of our mind?</h3><p>The answer is to re-frame how you think about AI. As creatives, we should think of AI as just another tool in our creativity tool box. Just when we invented computer software to help us overcome the limitations of having to design by hand, we should be thinking of AI as a way of helping us overcome the limitations we think are holding us back creatively.</p><h3>How exactly do we ‘partner’ with AI?</h3><p>Let’s break down the creative process from an algorithmic perspective, and see when, where and how we should be leveraging non-organic vs organic algorithms.</p><p>We can say broadly speaking that the creative process consists of 3 distinct stages.</p><ol><li>Preparation</li><li>Incubation</li><li>Illumination</li></ol><h3>Preparation</h3><p>This is the stage of information gathering. Of mastering the knowledge base of the subject.</p><p>However, our knowledge base is limited to the world in which we live. We have seen the impact of a wide breadth of knowledge on creativity when we look back on how homo sapiens exponentially evolved as our interaction and communication with each other increased.</p><p>Hunters and gatherers lived in disperse little bands in a wide variety of circumstances. This aided in the development of a diverse range of cultures and knowledge of survival, but at the expense of connectivity. Their world views were very limited and creativity was limited to developing simple knives and stone tools. And we didn’t innovate on these for 1.6 million years!!</p><p>Then fast forward to the agricultural revolution, which allowed many more people to live in close proximity. Thereby generating dense local networks of knowledge and the invention of profound things like the wheel or the plow.</p><p>And soon curiosity leads to modern explorers beginning to connect diverse continents and countries, and the flow of knowledge increases, and other amazing innovations appear. Cue the industrial revolution, then the information age….and well you get the point.</p><p><strong>Connectedness = The wide range of knowledge need for creativity.</strong></p><p>But as much as humans have made significant progress over the last 70,000 years — connectedness is at the very core of non-organic algorithms.</p><p>The internet is millions of machines in a network, exchanging knowledge. And thanks to connected devices, the amount of data is going to reach 44 zettabytes by 2020.</p><p>This is a non-organic algorithm’s playing ground. An AI system can ingest and ‘understand’ millions and millions of documents, much more than a human could in a lifetime.</p><p>So let’s look at an example of a company using AI to understand a huge amount of data to amplify their creativity.</p><h3>Stitch Fix</h3><p>Stitch Fix have a number of stylists that work with you one on one, to understand your style and preferences, then curate a box of clothes that will meet your needs at that time.</p><p>They have around 2,000 stylists, but manage to do millions of ‘fixes’! How does this work? How does this hands-on creative model scale?</p><p>Their whole business model is based on AI — from warehouse distribution to clothing design. But here we are going to talk specifically about the AI knowledge base they use to support the creativity of the stylists.</p><p>Stitch Fix uses AI to ‘understand’ millions of comments from customers everyday to help stylists do their job more efficiently.</p><p>But how? The AI has a knowledge base it can draw from so it can read and understand user’s comments and direct the stylists in the most appropriate direction.</p><p>So the way this works is the AI agent suggests the stylist recommend the maternity line when a customer says she’s in her ‘third trimester’ or distill ‘taking a trip’ into a Fix for vacation clothing. So the AI system is essentially doing a lot heavy lifting for they stylists.</p><p>So how does the AI know what these words ‘mean’?</p><p>Now this is going to get a little technical — but stay with me even if you don’t get it all. We’ll see this principle multiple times throughout the rest of this post.</p><p>Firstly we need train our model to understand the language Stitch Fix customers use.</p><p>In Stitch Fix’s case they probably took a huge amount of comments, industry related content and more, and stored each word in a corpus with its surrounding context using an algorithm called word2vec.</p><p>To explain this further, let’s pretend for a second the following sentence is our corpus we want to feed into word2vec.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Fjp3Uj5jdE8QoHh322dxOw.png" /></figure><p>So we take every word and store these pairs. In our case we are looking at one word each side, but usually it’s about 5 to 10 words either side.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MQHOee1qPSLv6Z-ey90Prw.png" /></figure><p>Then we can see how many times the words appear together and train the model to predict the chances of seeing one word given another word.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/704/1*Ueq6H51iI5jq9dUlZ0vm9Q.png" /></figure><p>Eventually the model see’s so many examples we can give it a new word, and it can give us the words that have a high probability of appearing around it. Therefore giving us the words that are contextually — or semantically related.</p><p>This can be sliced and diced many ways, for example by gender: king − man + woman = queen</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/891/0*nt42rOAinfoUuiGq" /><figcaption>Image: <a href="https://www.tensorflow.org/tutorials/word2vec">https://www.tensorflow.org/tutorials/word2vec</a></figcaption></figure><p>Verb, capital cities or anyway you want to cluster your model.</p><p>Word2Vec can hold up to 100 billion words which about 1 million books — crazy!</p><p>Now, we can begin to look for meaning in larger bodies of text, like sentences, and understand the relationship between words.</p><p>So the AI has built this large knowledge base, that collects and understands more data than 1 stylist ever could. Then passes this information on to the stylist where and when relevant.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/920/1*5vEON56_FKea8HmBTCLgsg.png" /></figure><p>So if the stylist sent the customer who said “I’m going to Belize for a work trip” just a bikini and a straw hat…they’d be in trouble.</p><p>This service just could not have been done at this scale before — it is only using AI that we can overcome the limitations of limited creative capacity.</p><p>We should be leveraging AI to help us understand, and identify insights in the huge amounts of data we have access to.</p><h3>Ok now moving on to Incubation</h3><p>This is most important part of the creative process, and its only now we have more sophisticated technology such as functional MRI’s, can we understand it a little more.</p><p>In this stage, ideas incubate without the individual directly working on the problem. Much restructuring and free association happens outside the conscious awareness — often referred to as divergent thinking.</p><p>Divergent thinking uses a combination of the knowledge base we described before, and a region of the brain called ‘primary process thinking’.</p><p>This is really important…creativity comes from not just the large logical knowledge base. We need the addition of the content stored in the primary process thinking.</p><p>What is primary process thinking?</p><p>Primary process thinking is a network of memories stored in your subconscious — free from any social restraints or inhibitions that rule the conscious.</p><p>So what memories are stored in my network? Well do you remember your trip into work say last Thursday? Probably not, unless something pretty special happened that elicited an emotional response.</p><p>Unfortunately (or maybe fortunately for some..) you can’t audit all of your memories stored in this network, but there have been several studies providing the rules as to what and where certain memories get stored.</p><p>You can imagine the primary process as an electrical network of all emotionally charged memories. And memories are stored based on their emotion.</p><p>When one node is activated, it sends a signal to other connected nodes, triggering other memories based on the same emotion.</p><p>Which is why sometimes when you’re daydreaming and you’re thinking about one thing, a few seconds later you could be thinking about something completely different — chances are they are related in emotion rather than context.</p><p>So the more emotionally charged memories you have, the more nodes you have connected, and the more you have available for free association between these and your knowledge base.</p><p>It may even explain why so many great artists are ‘tortured souls’. They have an enormous amount of emotionally charged memories along with being masters of their craft.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*w_zlB6bqMY68v8v6OgXXjQ.png" /></figure><p>Ok so technically…if we feed a non-organic algorithm all of these memories with there particular emotion tagged, it should be able to achieve the same results?</p><p>Not quite! Because your memories are unique to you!</p><p>Two people could have the same experience, but have completely different memories based on their emotional state at the time.</p><p>Your personal developmental history, your environmental stimulus, your emotion state — all contribute to what and how memories are stored.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*qBv0QxIA92Y8_26z" /></figure><p>And look I tried to do the math — but let’s just agree that until we reach a sophisticated version of a whole brain simulation — this isn’t going to be possible.</p><p>Just like machines are fundamentally built on creating large knowledge bases, our ability to tap into our emotion laden memories is in our very core and has be in development for 6 millions years…</p><p>If you look back over time again, you can see how the size of our brains, actually correlates to our creativity. Because the bigger the brain — the more storage we have for more complex memories.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*y1GFDx1-uhChyWpUV0Ugew.png" /><figcaption><a href="http://www.evoanth.net/2016/04/28/underestimating-human-brain-size/">http://www.evoanth.net/2016/04/28/underestimating-human-brain-size/</a></figcaption></figure><p>When we had no real signs of innovation 3.4 million year ago, our brains were estimated around 450 cubic centimeters (roughly the same as some chimpanzees).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bxhNtayM02HgUGm9XTRWqA.png" /><figcaption><a href="http://www.evoanth.net/2016/04/28/underestimating-human-brain-size/">http://www.evoanth.net/2016/04/28/underestimating-human-brain-size/</a></figcaption></figure><p>But by the time we invented those really simple stone tools — are brains had almost doubled to 930 cubic centimeters.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QA-g_1t_RDyxZaeWayPGtw.png" /><figcaption><a href="http://www.evoanth.net/2016/04/28/underestimating-human-brain-size/">http://www.evoanth.net/2016/04/28/underestimating-human-brain-size/</a></figcaption></figure><p>Our brains didn’t really grow much for 1.6 million years (the same amount of time we didn’t really improve a great deal on those simple stone tools) until about 100,000 years ago our brain ballooned to 1,330 cubic centimeters — the same time we got out of our creative rut!!</p><p>We had 100 billions of neurons for stimuli to be encoded. In addition, more neurons could participate in the encoding a particular memory, with a finer relation to an emotion. And hence, more available for free association!</p><p>So when it comes to the ability to tap into emotional memories inside the primary process network, it’s the non-organic algorithm who wins this time</p><h4>But AI can be used to create associations between seemingly unrelated content.</h4><p>This is where non-organic algorithms shine. Not only can AI be used to ‘understand’ huge amounts of information like we spoke about before, but can also identify hidden patterns or associations in this huge amount of data.</p><h3>R/GA Spark</h3><p>R/GA Spark is an internal tool we built at R/GA that uses AI to find associations amongst huge data sets.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cZRROzoWNomsXhvTTXKhtQ.png" /></figure><p>So the same way Stitch Fix trained their model on comments etc, our model ingested 10,000 articles — used word2vec to plot in high dimensional space</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aC4qRoo3hEq5Xzy9Iir3Pg.png" /><figcaption>Conceptual visualization of document clusters.</figcaption></figure><p>So you can imagine 10,000 articles, clustered together by semantic meaning.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WipN97pzSioXXM3ndDNo-A.png" /><figcaption>Assigning a number for each word in our vocabulary.</figcaption></figure><p>The vocabulary is then built up of all the words found in the articles, labeled 0-x. And we used the article contents in word2vec to group them semantically.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PsnSUBc9pyhMY4Olk7CXPQ.png" /></figure><p>So now we want to use the model. To do this, we first expand the 2 topics, by looking at our knowledge base. As an example, we ask the AI what words surround say —’ travel and vacations’ and what words surround ‘workplace environment’.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/804/1*dPI5ykPvIo0jayV1xw9Vgg.png" /></figure><p>Again, we haven’t taught the AI the meaning of any of these words!!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/890/1*X5YS-CsOBx4UZj2qSknWmw.png" /></figure><p>Then we iterate over each word combination and look for content that is related to any combination of these words. This is something that you just can’t do manually.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pLrWsy85REwskcvNrDMxzA.png" /></figure><p>And we find the article (or articles) that are closely related to those 2 topics — semantically. In this example we get the article ‘The Digital Nomad Life: Combining Work and Travel’.</p><p>In summary, there is a clear creative advantage in what we have sitting in our primary process network. So this is an excuse for all creatives to start having more emotionally charged experiences. Go on holiday to a foreign country, connect with people who are culturally, racially or politically diverse from you, get out of your echo chamber and challenge perspectives that you wouldn’t normally be exposed to.</p><p>And then leverage AI to help find the connections, build bridges of relevance, or uncover patterns that inspire your next creative idea.</p><h3>Illumination</h3><p>This is the phase where the solution to the problem occurs or is recognized. This is often referred to as the ‘aha’ moment. The moment you realize you’ve got the answer.</p><p>I know personally — when I have come up with a great idea, I get that same feeling. It’s a gut feeling that I have best possible combination of associations in my head.</p><p>This gut feeling or intuition “are rapid cognition or condensed reasoning that takes advantage of the brains built in shortcuts” to be able to evaluate the validity of all outcomes produced.</p><p>When, or if, a solution has been found, it is marked by a feeling of completion. This then activates a release of dopamine, to reward the brain and make sure you come back again.</p><p>So we have the primary process at work here again.</p><p>And going back to our ancestors, our large brained hominids could not afford to linger in an associative state for too long, so they developed a way of switching back to analytical thought by altering concentrations of dopamine and other neurotransmitters.</p><p>The pleasure that comes when the solution is found, or the composition is formulated, helps the creator know when to stop or which of the myriad of ideas and configurations is correct.</p><p>Without this mechanism in place, how would we know when to stop producing combinations?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/604/0*7JwQrOpM1lHVGsNx" /></figure><p>The brilliant thing about divergent thinking is, that it’s completely irrational and illogical. There are no defined rules, no explicit definition of the correct answer. Therefore no programmable parameters we can hard wire, to tell us when we’ve got it.</p><p>We see this in generative design.</p><h3>The Next Rembrandt</h3><p>The Next Rembrandt is an amazing project demonstrating the power of generative design. This, is not a real Rembrandt painting…</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UtlGN7qsgWt97eB5hlb0KA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*n1Q5rCfNAH4jVHXC93Mg4A.png" /><figcaption>Images: <a href="https://www.nextrembrandt.com/">https://www.nextrembrandt.com/</a></figcaption></figure><p>Look at the brush work. It’s incredible. Unless you are a professional art historian, it would be very hard to tell this apart from something in his body of work.</p><p>So how does this work? First they needed to gather a set of images they can use as ‘inspiration’ for the AI algorithm. They collected 2D and 3D images from multiple sources — around 150 Gigabytes of digitally rendered graphics.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*O2M23SGsZE1ClbJuPbMFTQ.png" /><figcaption>Images: <a href="https://www.nextrembrandt.com/">https://www.nextrembrandt.com/</a></figcaption></figure><p>Then, using image recognition, they were able to determine some key components to a typical Rembrandt painting.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*G_5Lh2lVPgnVYzG-gkCrRg.png" /><figcaption>Image: <a href="https://www.nextrembrandt.com/">https://www.nextrembrandt.com/</a></figcaption></figure><p>Portrait, Caucasian, male, facial hair, 30–40, dark clothing, collar, wearing a hat and facing to the right. So they now have starting point to start building this painting.</p><p>They then began developing the feature of the sitter — based on a similar model.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WxX3lB5t9YcOSIxor7K0Lg.png" /><figcaption>Image: <a href="https://www.nextrembrandt.com/">https://www.nextrembrandt.com/</a></figcaption></figure><p>Looking at the the different variations of a feature, for example an eye, and then letting the algorithm create one on its own from the rules it has learnt from studying all the real Rembrandt eyes.</p><p>And finally, they took a similar approach with height maps of all the 3D Rembrandt images.</p><p>The final piece was then 3D printed to achieve this authentic dimension.</p><p>The result is incredible. They are now considering using this technology in reverse to find fake paintings!</p><blockquote>“It was a struggle because it could be a perpetual art project…When do you say, ‘Now it’s good’?”</blockquote><blockquote>J. Walter Thompson’s Bas Korsten</blockquote><p>I remember reading that the engineers would run the algorithm over night and then come in the next morning to go over the thousands of variations it had created. How does the algorithm know when to stop? It doesn&#39;t.</p><p>That is why your ability to subconsciously recognize when you have the best possible combination, is your advantage.</p><p>Another crucial part of the creative process is execution. AI can be used to automatically validate or make your finished product better.</p><h3>Netflix</h3><p>The Netflix recommendation engine is the best example of this.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5tjV0gMpAxRad7rkVBbNTw.png" /></figure><p>Once the engineers have created the base model, they use feedback from the user to retrain the model on the fly, becoming more and more accurate over time.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*f1LsnU9hz6veGxdxxj5izw.png" /></figure><p>So in practice, when you have an idea — we can use AI to validate it and once we have produced it — use AI to make it even better. But bringing that combination of thoughts and knowledge together and realizing it has the magic quality of a brilliant creative idea….that’s all you baby.</p><h3>The AI Creativity Model</h3><p>So what’s the score board?!</p><h4>Preparation:</h4><p><strong>Non-organic<br></strong>Leverage AI to help us understand, and identify insights in the huge amounts of data we have access to.</p><h4>Incubation:</h4><p><strong>Non-organic + Organic<br></strong>Leverage AI to help find the connections, build bridges of relevance, or uncover patterns that inspire your next creative idea.</p><h4>Illumination and Validation:</h4><p><strong>Non-organic +Organic<br></strong>Leverage AI to validate your creative idea and make it better.</p><h4>There you have it — the key to limitless creativity. The blueprint for you to leverage AI to extend the capabilities of your mind!</h4><p><strong>Creative human + Artificial Intelligence = Limitless Creativity</strong></p><p>Understanding our limitations and the capabilities of new technologies such as AI, allow us to continually create new innovations that start the cycle all over again. Which is why it has never been a more exciting time to be a creative.</p><p>Jenna Niven — Group Creative Director R/GA NY<br>jenna.niven@rga.com</p><p>@jennaniven</p><h3>Don’t forget to give us your 👏 !</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/780/0*jkW7MZWo_2QGmdo0" /></figure><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fupscri.be%2F8f5f8b%3Fas_embed%3Dtrue&amp;dntp=1&amp;url=https%3A%2F%2Fupscri.be%2F8f5f8b&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=upscri" width="800" height="400" frameborder="0" scrolling="no"><a href="https://medium.com/media/c43026df6fee7cdb1aab8aaf916125ea/href">https://medium.com/media/c43026df6fee7cdb1aab8aaf916125ea/href</a></iframe><figure><a href="https://becominghuman.ai/artificial-intelligence-communities-c305f28e674c"><img alt="" src="https://cdn-images-1.medium.com/max/255/1*2f7OqE2AJK1KSrhkmD9ZMw.png" /></a></figure><figure><a href="https://upscri.be/8f5f8b"><img alt="" src="https://cdn-images-1.medium.com/max/255/1*v-PpfkSWHbvlWWamSVHHWg.png" /></a></figure><figure><a href="https://becominghuman.ai/write-for-us-48270209de63"><img alt="" src="https://cdn-images-1.medium.com/max/255/1*Wt2auqISiEAOZxJ-I7brDQ.png" /></a></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=30cc7e26f28d" width="1" height="1" alt=""><hr><p><a href="https://becominghuman.ai/limitless-creativity-how-to-design-in-an-ai-world-30cc7e26f28d">Limitless Creativity: How to Design in an AI World</a> was originally published in <a href="https://becominghuman.ai/">Becoming Human: Artificial Intelligence Magazine</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Data isn’t the new oil. Creativity is.]]></title>
            <link>https://medium.com/@jennaniven/data-isnt-the-new-oil-creativity-is-7de0afe0b0db?source=rss-77f92ddbb1e0------2</link>
            <guid isPermaLink="false">https://medium.com/p/7de0afe0b0db</guid>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[nlp]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[design]]></category>
            <dc:creator><![CDATA[Jenna Niven]]></dc:creator>
            <pubDate>Thu, 11 Apr 2019 20:06:56 GMT</pubDate>
            <atom:updated>2019-04-11T20:06:56.356Z</atom:updated>
            <content:encoded><![CDATA[<p>In 2006 Clive Humby coined the phrase “data is the new oil”.</p><p>And he was right because around 2012 machine learning had some major breakthroughs which catapulted its usage from the depths of a research lab, into businesses everywhere. <a href="https://www.forbes.com/sites/gilpress/2019/04/03/7-indicators-of-the-state-of-artificial-intelligence-ai-march-2019/amp/">Today, business value derived from AI reached $1.2 trillion worldwide and is set to grow to $3.9 trillion by 2022 (Gartner).</a></p><p>What has data got to do with it? Well, machine learning models are prediction engines. And they make predictions based on what they have seen before. The more data they have seen, the better the prediction. Hence why you need a lot of data to train a model from scratch that performs well.</p><p>The more proprietary data you have (that’s useful) the bigger the competitive advantage. So it made sense that we have been ‘labeling’ (ML joke..😂) data as the new oil.</p><p>What<em> I </em>certainly didn’t expect to happen though, was the extent to which the AI community was willing to share code and high-quality, labelled data sets!</p><p>In 2010, the first version of the ImageNet dataset was released. ImageNet contains more than 14 million hand annotated images, of more than 20,000 categories.</p><p>In 2017 a dataset called EMNIST was published. It contains 280,000 images of handwritten digits and characters.</p><p>And according to their <a href="https://developer.amazon.com/blogs/alexa/post/30dc5515-3b9f-4ec2-8f2a-ac98254625c6/topical-chat-dataset-helps-researchers-address-hard-challenges-in-natural-conversation">blog</a>, in September 2019 Amazon will release a dataset called Topical Chat. This dataset will contain more than 210,000 utterances (over 4,100,000 words) derived from conversations collected by Alexa.</p><p>Even more exciting, is the availability of pre-trained machine learning models. This means basically all the grunt work has been done for you. For example, in 2018 Google released a pre-trained NLP model called BERT (Bidirectional Encoder Representations from Transformers). BERT has been trained on 3,300 million words — all you have to do is train on top with a small amount of data based on whatever the task is.</p><p>So thanks to the open AI community, all of sudden high-quality data doesn’t seem that hard to come by. But if we all have the same pre-trained models and datasets, where has the competitive advantage gone?</p><p><strong>Creativity will soon become the competitive advantage and the ‘new oil’ of the AI revolution.</strong></p><p>As Fai-Lee says in his book <a href="https://www.amazon.com/AI-Superpowers-China-Silicon-Valley/dp/132854639X">AI Superpowers</a>, we are moving from the research phase to ‘The Age of Implementation.’</p><blockquote>“Deep-learning pioneer Andrew Ng has compared AI to Thomas Edison’s harnessing of electricity: a breakthrough technology on it’s own, and one that once harnessed can be applied to revolutionizing dozens of industries.” (AI Superpowers, Kai-Fu Lee)</blockquote><p>The rate of new AI breakthroughs will slow, and the innovation will lie in the application of the technology.</p><p>The innovation will come from creatives who:</p><ul><li>Understand the problems that need (and can) be solved.</li><li>Understand what machine learning models are out there and their applications.</li><li>And finally, understand how one or more of these stitched together can form a product or service that solves the problem.</li></ul><p>Currently at R/GA, the team and I are working on a product that aims to help give news readers access to a pool of diverse journalists’ perspectives.</p><p>The solution uses NLP to find out what is trending in the news, how the majority of journalists are talking about these topics, and then find diverse journalists’ that give an alternate point of view.</p><p>Not an easy task when you break it down.</p><p>Here are the models we needed:</p><ul><li>Topic model: to find the trending news topics</li><li>Word vector: to understand what each journalist is talking about on that topic</li><li>Sentiment model: to understand how each journalist is talking about certain words, whether that be positive, negative or neutral</li></ul><p>Each of these models alone would require an enormous amount of data to build from scratch. But a variety of each of these models are available pre-trained.</p><p>Topic Model example: <a href="https://medium.com/MALLET%20topic%20model%20package">MALLET topic model package</a></p><p>Word Vector Model example: <a href="https://github.com/dav/word2vec">Word2Vec</a></p><p>Sentiment Model example: <a href="https://stanfordnlp.github.io/CoreNLP/">Stanford CoreNLP</a></p><p><em>Note: Choosing which model to use is quite a complicated (and creative!) process and does take a lot of R &amp; D.</em></p><p>Each of the models individually are VERY cool, but the VERY VERY cool part is when you see how they are all working together and — fingers crossed — helping to bring more diversity to mainstream media. <em>Launching soon!</em></p><p>So if your AI strategy only consists of hiring data engineers and acquiring data — you’re doing it wrong. You should be looking for creatives who will, in collaboration with the data engineers, turn the breakthroughs in machine learning into products and services that will change the world.</p><p>And if you are one of those creatives, get in touch!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7de0afe0b0db" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Thinking Ahead 2018]]></title>
            <link>https://medium.com/@jennaniven/thinking-ahead-2018-b526d385e125?source=rss-77f92ddbb1e0------2</link>
            <guid isPermaLink="false">https://medium.com/p/b526d385e125</guid>
            <category><![CDATA[technology-trends]]></category>
            <category><![CDATA[tech]]></category>
            <category><![CDATA[future]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Jenna Niven]]></dc:creator>
            <pubDate>Mon, 01 Jan 2018 08:10:16 GMT</pubDate>
            <atom:updated>2018-01-01T08:10:16.559Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>Do you think AI technology like home assistants is a trend or a fad?</strong></h3><p>Definitely not a fad. Interactive Voice Recognition systems (IVR’s) have been around since the early 2000’s just in different forms. You know the old ‘say yes for account balance’…. What we’ve seen is the underlying technology become a lot more sophisticated — really quickly. And new forms of IVR’s appearing such as an Alexa or Google Home. So it’s definitely not a fad — but we will see it evolve quicker than we’ve seen before thanks to the democratization of technology.</p><h3><strong>Are there any interesting examples from this year that demonstrate why the above is in fact a trend?</strong></h3><p>I think the democratization of tech is one of the reasons why. I mean the number of Alexa skills has exploded — in two years there has been 20,000 skills created….due probably in part to the developer ecosystem Amazon has created. I mean even I can build a skill with my limited programming knowledge!</p><h3><strong>Do you believe we’ll get virtual assistant fatigue in the next 12 months?</strong></h3><p>I think people will just begin to expect this kind of experience from every interaction that they have with brands. From buying groceries to lodging an insurance claim — experience transferal will ensure not fatigue but demand.<br>Although I do expect there to be some consolidation in the general purpose assistant market — Alexa or Google Home..who will win?!?!</p><h3><strong>What will we see happen next? How long will it take?</strong></h3><p>To my point earlier, there really is an arms race in the general purpose AI space on 2 fronts:<br>- Get more people developing applications on your platform<br>- Get more user’s buying and using your platform — aka more data</p><p>Who ever wins on both these fronts will ultimately see their market share increase. And as the gap increases it will be harder and harder to close as these general purpose assistant become ingrained in our lives and behaviors.</p><h3><strong>What’s the sleeper threat for 2018 — AI/Machine Learning, or any other industry — that we should be looking out for?</strong></h3><p>I see 2 sleeper threats.</p><ol><li><strong>Algorithmic bias<br></strong>For about a decade now we have left algorithms to decide what emails we read, what music we listen to, what schools we are allowed to go to, what jobs we can apply for….even what prisoners we release on parole! And only now are we slowly starting to think ‘ um, how does this thing actually work? What is it not showing me? What am I missing out on?’ Which has uncovered incredible amounts of unconscious bias. AI amplifies this bias. So we need to take the steps now to ensure that what we have already created and what we create in the future, are ethically and morally buttoned up.</li><li><strong>Along the same vein — but I am going to go a little ‘Musk’ on you….</strong><br>A question I have been asking myself lately has been… If I had a life threatening ailment and had to choose between an average Dr who has an incredible AI he uses to treat and diagnose, or the leading Dr in that field who has written many a papers on the topic — I would choose the average Dr with the AI…. which actually really surprised me. I would let an algorithm make an important life decision for me, as I know it would be able to consume and analyzes more information than one leading specialist could ever do in his lifetime 10x times over.<br>This leads me to think — what will happen to human civilization if we relinquish our right to choose based on the fact that these algorithms would make the best decision for me — by fact, as it can take into account way more data than us humans can….? Hmmm, not sure but an interesting thought to leave you with!!</li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b526d385e125" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A High-level Framework for Designing with AI]]></title>
            <link>https://medium.com/r-ga-by-design/a-high-level-framework-for-designing-with-ai-544285dbc9f?source=rss-77f92ddbb1e0------2</link>
            <guid isPermaLink="false">https://medium.com/p/544285dbc9f</guid>
            <category><![CDATA[design]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Jenna Niven]]></dc:creator>
            <pubDate>Thu, 27 Apr 2017 21:11:19 GMT</pubDate>
            <atom:updated>2017-04-27T21:11:19.259Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WFud_Y7ZI5UOzBvCxbX2rw.png" /></figure><p>New technologies come and go, but not since the internet have we seen such a dramatic game changer as AI. AI promises to disrupt just about every industry you can think of. In other words, just about every one of our clients have been, or will be, affected by AI — and much like the internet and our shift into digital design, we must also start to augment our design processes to be better able to meet these needs.</p><p>However, due to the exponential growth of this technology, we currently lack a shared understanding of the principles of AI and common vocabulary is often misattributed. For example, AI does not have one unified definition and should be a term used to encompass many concepts associated with AI: Neural Networks, Deep Learning, Machine Learning, Natural Language Processing — just to name a few — but the term AI can sometimes be used to describe things like ‘chatbot’ which doesn’t necessarily sit under AI explicitly.</p><p>It’s important though, that we begin to define and agree upon what we understand to be the fundamentals of AI, as the process in which to build successful products or services is dependent on the interconnectedness and alignment across our disciplines.</p><p>So the impetus for this article is two-fold;</p><ol><li>Create a framework for thinking about designing with AI, and</li><li>Create a reference to better understand the fundamentals of AI and how the components fit together.</li></ol><h3>The Framework</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/578/1*mviJUn3CCSnT9mjt3bmNLA.png" /></figure><h3>Input and Output</h3><p>It’s no surprise that input/output can come in a variety of different form factors. Here we essentially need to figure out:</p><ol><li>What data we are trying to collect,</li><li>The method in which we are collecting the data, and</li><li>How we are communicating to the user.</li></ol><p>To help figure this out I use the following equation:<br><strong>Brand + Data + UI = Input/Output</strong></p><h3>Brand</h3><p>Branding is an incredibly important component when designing with AI, particularly when we are using derivatives of the same algorithms/API’s. How will a user feel after an interaction with your product and is this in line with the overarching brand strategy? The type of brand personality you infuse play a huge role in the differentiation of your product or service, the level of engagement you achieve, and the way your users interact with it.</p><blockquote>“Our approach on personality includes defining a voice with an actual personality. This included writing a detailed personality, and laying out how we wanted Cortana to be perceived. We used words like witty, confident and loyal to describe how Cortana responds through voice, text and animated character.” — <a href="https://blogs.windows.com/windowsexperience/2015/02/10/how-cortana-comes-to-life-in-windows-10/#ytorFrGLtStvgYhS.97">Marcus Ash, Group Program Director, Cortana</a></blockquote><p>There are many ways your brand can filter through your product — the obvious being voice user interface (VUI), but it could also be as simple as a series of beeps or the way you write your conversation dialog.</p><h3>Data</h3><p>Thanks to the IoT, there is now more data available than ever before. In fact, by 2020 there is expected to be more than 44 zetabytes of data in the world — that’s 15x every grain of sand on earth! So your data input could be anything from a simple text input, to NFC, to your smartwatch.</p><p>Depending on how you are using the data you collect, you may require structured data or unstructured data. Structured data basically means data that is already organized into a structure like a database, csv or json file and has labels associated with it. Unstructured data on the other hand, is data that has not been classed or categorized. For example a set of random images or a call center recording.</p><p><strong>AI Components<br></strong><em>Structured Data, Unstructured Data</em></p><h3>User Interface</h3><p>Given that you may be dealing with a connected ecosystem — you may have different form factors to communicate through. Some may have screens or other ways of communicating such as flashing LED lights or some may not have the ability to communicate at all. What is the most appropriate communication interface?</p><p>VUI is a hot topic at the moment. Voice recognition accuracy has significantly improved over the last 12 months and it’s expected that at least 50% of all searches will be done via voice or image by 2020. However, we still need to think about whether the methods you are using to collect the data fit in with existing heuristic patterns for your product. And, how will you account for fall-back or secondary options?</p><blockquote>“No one wants to wait 10 seconds for a response. Accuracy, followed by latency, are the two key metrics for a production speech system…” Andrew Ng, former Chief Scientist at Baidu</blockquote><p>Another common example is free form text through a conversational interface such as Facebook messenger or SMS. This allows the user to converse in a way that is natural to them. Not only does this reduce friction, it also allows us to see if the product we have created is actually meeting the real needs of its users!</p><p><strong>AI Components<br></strong><em>Voice User Interface, Conversational Interface</em></p><h3>Recognition</h3><p>The recognition phase, is where we begin to identify this anonymous input. The term ‘understanding’ in the human world is very different from the definition in the computer world. In the simplest terms, whatever we put into a machine invariably gets broken down into a data point that, on its own, has no meaning. However, once we look at that data point in the context of what we call a ‘class,’ meaning or understanding is attributed.</p><blockquote>“Speech recognition systems that map audio signals to words are the canonical example of assessment. The input is a raw audio signal and the output is a set of words that can then be used to determine meaning.” —<a href="http://ai.xprize.org/news/periodic-table-of-ai"> Kris Hammond, IBM Watson AI XPrize</a></blockquote><p>There are 2 ways we can assign classes to the data we have collected — supervised or unsupervised. Supervised means we manually assign a class or label to a data point while unsupervised means using one of a variety of algorithms, such as TSNE, to automatically cluster data based on how it perceives the relationship between the data.</p><p>Here is where it starts to get complicated; data points don’t have to be purely 2 dimensional. A data point could be multidimensional; it could have several variables attached to it.</p><p>A good example is if we think about a person. You could use many different features to describe a person — their height, eye color, favorite food, hobbies. Imagine you did this for all of your work colleagues and then had to try and figure out the commonalities between them all. This is where AI comes in handy — it doesn&#39;t see the meaning of these features, it just sees them as numbers. These numbers can then be plotted in a three dimensional space where ‘people’ that it thinks are related are clustered more closely together. We can then use this structure to begin to infer meaning.</p><p><strong>AI Components<br></strong><em>Speech Recognition, Audio Recognition, Facial Recognition, Image Recognition, General Recognition, Text Extraction, Concept Extraction, Supervised Learning, Unsupervised Learning</em></p><h3>Reasoning</h3><p>Reasoning is intrinsically linked to the recognition of the input, and where the term ‘understanding’ is again used in the context of the machine. Once we have mapped the data in whatever way is appropriate, we can begin the interpretation. Moving from categorizing a series of words, to being able to understand the meaning of those words, derive sentiment, and extract entities.</p><p>Here is an example using <a href="http://www.textrazor.com/">Text Razor</a>:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/980/1*IxCfyjJvFy4BxyBO7obsIw.png" /></figure><p>In this case, the Text Razor model can understand that ‘House’ in this context, actually represents a company or organization rather than a building.</p><p>As you can see, the interpretation of the input can be heavily influenced by the context — this is where things start to get interesting! Previous conversations can help narrow down your user’s intent.</p><p>In essence, this is facilitated by machine learning. According to Kris Hammond, there are three types of learning:</p><ol><li>High volume data learning (being able to learn by analyzing huge amounts of structured or unstructured data)</li><li>Correlation learning (understanding the relationships between data) and</li><li>Incremental learning (learning from the execution of a previous task).</li></ol><p>The reasoning phase should provide a synthesis of data based on confidence scores or a prediction of future states.</p><p>Confidence scores are a rating that may be provided with the output of the reasoning, based on how well a machine thinks it has performed. The most well-known use of a confidence score was in the infamous Watson Jeopardy scene where the AI replied ‘What is Toronto?????’ in response to “Its largest airport is named for a World War II hero; its second largest, for a World War II battle” in the category of “U.S. Cities.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/0*VHLEpGDU0EHjai35." /></figure><p>After dissecting the code, Watson developers saw that it had come to two conclusions — one being ‘Toronto’ and the other being ‘Chicago’ (the correct answer)—however both had a very low confidence score of about 30%. Because it was unable to make a clear decision, it expressed its low confidence with question marks.</p><p><strong>AI Components<br></strong><em>Speech Identification, Audio Identification, Facial Identification, Image Identification, Data Analytics, Synthesis, Predictive Analytics, Language Understanding, Machine Learning, Concept Expansion, Sentiment Analysis</em></p><h3>Response</h3><p>And finally, here is where we start making decisions of what is to happen as a result of the interpretation of our data. We could respond by communicating with the user in the form of dialog, or an automated email. Or we could take action, like turning the lights a different color or ordering a book off Amazon.</p><p>You could also determine your response based on the confidence score of your interpretation. If you have a high degree of confidence you could use an implicit confirmation such as saying “ok, setting additional timer for 3 minutes” and also completing the action of setting the timer. However, if you only have a small degree of confidence, you could use the response to illicit further confirmation by asking “You would like to set a timer for 3 minutes, is that correct?” before completing the action.</p><p><strong>AI Components<br></strong><em>Language Generation, Decision Making, Relationship Learning, Knowledge Refinement</em></p><h3>Putting it all together</h3><p>Now that you understand each phase, here is a dissection of a fictional product based on the framework.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/916/1*7UbRXJHb-eLMojYgIjxThw.png" /></figure><h3>A final thought</h3><p>The internet did not disrupt the commerce industry. The applications we developed leveraging the internet’s many capabilities, such as encryption and globalization, were the defining factors in transforming an entire sector. AI will prove to be the same. The winners will be the people who truly understand the capabilities of the technology and are able to develop new and creative ways of applying it.</p><p>Hopefully this framework is a good starting point on your discovery and learning of the amazing potential of AI.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=544285dbc9f" width="1" height="1" alt=""><hr><p><a href="https://medium.com/r-ga-by-design/a-high-level-framework-for-designing-with-ai-544285dbc9f">A High-level Framework for Designing with AI</a> was originally published in <a href="https://medium.com/r-ga-by-design">R/GA by Design</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[An Easy Explanation of Artificial Intelligence]]></title>
            <link>https://medium.com/@jennaniven/an-easy-explanation-of-artificial-intelligence-acca15f56aea?source=rss-77f92ddbb1e0------2</link>
            <guid isPermaLink="false">https://medium.com/p/acca15f56aea</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[chatbots]]></category>
            <category><![CDATA[neural-networks]]></category>
            <category><![CDATA[advertising]]></category>
            <dc:creator><![CDATA[Jenna Niven]]></dc:creator>
            <pubDate>Wed, 04 Jan 2017 09:48:47 GMT</pubDate>
            <atom:updated>2017-01-04T09:48:47.733Z</atom:updated>
            <content:encoded><![CDATA[<p>I was recently involved in a project which involved a cross pollination of campaign and product ideas. With a good brief, a technologically sophisticated client, and lots of unstructured and structured data — the project was a perfect opportunity to leverage AI to build a really cool product aimed at meeting the brief while being enhanced by a clever campaign message. The only problem…. the creative lead had no idea what Artificial Intelligence is and he isn’t alone!</p><p>Brilliant idea, after brilliant idea (I may be a bit biased about said brilliance…) were left with blank, frustrated stares. “I don’t get it.” And “Where’s the magic?” were common phrases parsed through the creative process and although the client ultimately understood and loved the idea that was presented — getting through the internal agency jungle wasn’t as easy as it should have been. These days, ideas leveraging technology in new exciting ways should be a no brainer!</p><p>This experience is the impetus for this short article: An Easy Explanation of Artificial Intelligence. Hopefully this brief description will give people in advertising, and in general, a basic understanding of what AI is and capable of doing so, in 2017, we can;</p><p>1. Get past AI novelty use cases<br>2. Move beyond falling back to traditional ways of creating, and<br>3. Finally build some amazingly smart, creative AI solutions for our clients.</p><h3>An Easy Explanation of Artificial Intelligence</h3><p>Artificial Intelligence (AI) is a term used to generally describe a ‘new’ way of computing. For comparison, the old way requires us to tell the computer EXACTLY what we want it to do. The computer needs us to program very complex and exact instructions for it to complete a task — this is a programmable system. The ‘new’ way is based on what we call probabilistic logic. This means we don’t need to tell the computer exactly what to do, we can give it a set of parameters and data set, and based on what ever type of algorithm you are using, it can compute a set of answers with assigned accuracy (or probability) on its own.</p><p>Artificial Intelligence encompasses many terms you may have heard kicked about: Neural Networks, Deep Learning, Machine Learning, Cognitive Computing, Augmented Intelligence, Natural Language Processing — just to name a few. However, one noticeable absentee is bots. While bots (robots, chat bots, drones etc) could all possibly use AI in one way or another, they are merely the vehicle or the UI to express the results of an AI algorithm and should not be confused for AI itself. For example, most chat bots use decision tree logic to output a response and are not classified as AI.</p><p>From my perspective there are 3 macro applications of AI:</p><p><strong>1. Discovery</strong><br> AI can discover patterns and insights from the ever increasing volumes of data, that perhaps even the most intelligent humans could not detect.</p><p><strong>2. Engagement<br></strong>AI allows expert information to be delivered to an extended audience in a timely, natural and usable way.</p><p><strong>3. Decision</strong><br> AI can help with decision making by being able to consume large amounts of data and suggest a list of options.</p><p>There are many core AI algorithms already created we can leverage to create products or campaigns. For example T-SNE is a technique that turns high-dimensional data into two or three dimensions, which can then be visualized in a scatter plot — any there are many more.</p><p>AI isn’t a gimmick or technology that can perform amazing, rub-your-eyes-in-disbelief magic. Like anything computational, there are parameters required and obviously technological limitations. Having said that, I believe the potential of AI really hinges on the creative application. How can we use x algorithm to do y that outputs to z…. there really are a million ways.</p><p>Hopefully this short description will go a little way in making your life (and mine…) easier selling in AI in 2017, and we can finally get on with making some really cool ideas a reality!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=acca15f56aea" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Artificial Intelligence in Advertising]]></title>
            <link>https://medium.com/chatbot-news-daily/ai-in-advertising-d490f067a7bf?source=rss-77f92ddbb1e0------2</link>
            <guid isPermaLink="false">https://medium.com/p/d490f067a7bf</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[advertising]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[chatbots]]></category>
            <dc:creator><![CDATA[Jenna Niven]]></dc:creator>
            <pubDate>Tue, 01 Nov 2016 00:02:22 GMT</pubDate>
            <atom:updated>2017-01-19T16:30:49.403Z</atom:updated>
            <content:encoded><![CDATA[<p><a href="https://www.intelligentsia.ai/">Intelligentsia.ai</a> is for people who are curious about AI and want to see the immediate steps ahead as well as understand the big visions. I sat down with Helen and Dave, the founders of Intelligentsia, to discuss my opinions on AI in advertising.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*sz89tYfiK4aLEIjl5LC6qA.png" /><figcaption>Credits: Shutterstock</figcaption></figure><p><strong>Helen:</strong> What impact do you see AI making today on marketing and advertising?</p><p><strong>Jenna:</strong> Advertising and marketing has changed significantly from the days of purely running a campaign over print. We have not only a multitude of channels we can communicate through but also the ability to develop products and systems to meet the exact needs of a particular audience. For example, when we look at improving the effectiveness of a customer support tool, we can now create new ways to communicate with the audience, maybe a chatbot, and use sentiment analysis to gain insight into a person’s temperament at that exact moment. We also have tools for analyzing a whole different set of data, both structured and unstructured, for example call center recordings. This enables us to better understand a customer’s needs and be able to personalize their experience more effectively in real time. So advertising is moving from traditional campaign work to a service that helps access a particular audience in more effective and diverse ways, perhaps with new products, perhaps with new business models. The main impact of AI will be enabling this.</p><p><strong>Helen:</strong> So being able to create new products and solutions at scale and in real time?</p><p><strong>Jenna:</strong> Yes, for example, a web site is now a product in itself, not just a static place used for developing brand awareness. We could create a tool for a website that changes a customer’s experience based on previous visits. Advertising and product development are merging, becoming a more dynamic service not just for brand awareness but creating new systems that support brand awareness and customer experience.</p><p><strong>Helen:</strong> Speaking of website tools, is the chatbot hype ahead of the reality?</p><p><strong>Jenna:</strong> Definitely. The problem is that clients view chatbots quite negatively. This is usually because of a bad experience in the past or some of the recent bad press. There are a lot of developments in natural language processing and in machine learning and chatbots are now a lot more sophisticated. Some clients are taking baby steps with chatbots and using them to learn. The benefit of a chatbot is multifaceted; It’s not only about being able to have a conversation with a customer, but the ability to parse information from the customer and keep a history, to analyze sentiment and to make it’s solutions smarter and more accurate over time. I think the more important issue generally is determining what specific AI capability is right for the client. It’s not always going to be a chatbot, there are a lot of other AI solutions out there, and we start with the client’s need.</p><p><strong>Helen:</strong> Do you see chatbots being able to play a role in generating brand awareness and publicizing and exercising marketing campaigns? Can a chatbot be the voice of a company and go out and bring in a customer or is it still really just about customer service?</p><p><strong>Jenna:</strong> Definitely, a chatbot can be part of a campaign and as a product to support the brand. But this raises important questions like, how do we teach a chatbot to be the voice of the brand, how do we get it to talk in the right tone of voice and use the right language? I haven’t seen anyone do this well yet but it’s an exciting area of development.</p><p><strong>Helen:</strong> How would you describe talking to clients about AI and machine learning? What general level of awareness do you find?</p><p><strong>Jenna:</strong> It depends on the industry. Financial sector clients tend to have a relatively high level of awareness of AI but no experience of practical applications beyond analysis. They haven’t been exposed to ideas for new products and services that are AI based. Consumer sector clients tend to be more aware because they’ve used AI before, say on their website as a recommendation engine. But in general we see most clients being cautious about building new products based on AI because there isn’t enough out there already. This is very much an early adopter’s market — and advantage! We will see more adoption as more solutions using machine learning are trained and deployed. There’s not a lot out there that will blow you away yet. It’s still early days.</p><p><strong>Helen:</strong> Do you think AI will change what is measured in advertising and marketing?</p><p><strong>Jenna:</strong> Yes, and that’s driven by the ability of AI to handle huge quantities of data. We now have so many more data sources; we can do collective analysis, we can drill down in ways that haven’t been possible and we can do all of this in real time. This is a massive opportunity for AI. It’s a huge shift to go from running a campaign, collecting information and analyzing it days or weeks later to being able to analyze in real time, get insights about how to augment the campaign such as changing the messaging, the channel or the time of day. The level of granularity we can now go to during a campaign is a very big shift. For example, take football, where you have a high volume audience. Now it’s possible to see significant change in real time. For these large audiences in a specific time frame, it could now even be possible to go down to a specific moment in the game and target a campaign around the sentiment during that actual play.</p><p><strong>Helen:</strong> Do you encounter resistance to the AI opportunity because of this level of change? Are some clients better positioned than others to move to that level of sophistication more quickly?</p><p><strong>Jenna:</strong> If you take the traditional process; running a campaign, waiting for it to finish, then the manual process for assessing effectiveness of the campaign (which even then doesn’t do a very good job because it doesn’t take into account all sorts of other factors) you’d expect people to embrace the new AI solutions. But, yes, there are areas of resistance based on people not being sure if it will make their jobs more complicated or more difficult to manage. Most commonly, people aren’t able to visualize the workflow process that’s generated by AI. Again, I think it will take a few early adopters to figure it out, explain it to others and be able to demonstrate how it makes their jobs easier as well as their campaigns more effective. It’s a natural fear-based reaction and I think this is a conversation that has to be had over time with people who have experienced how AI enhances their role rather than making it redundant.</p><p><strong>Helen:</strong> Advertising prides itself on being a creative industry. How is the creative process being disrupted?</p><p><strong>Jenna:</strong> The creative concept is always going to be a human activity. I think the backend of the process is what AI best drives; the analysis, the new data sources, the sentiment, what the audience likes, the strategy to support the creative concept, the whole creative process. Everything including the ideation, the use of AI, experience design, prototyping and testing — will change. The people who will succeed are the ones who understand the fundamental concepts of the AI capabilities. But machines won’t come up with creative concept anytime soon!</p><p><strong>Helen:</strong> What happens when a machine generates an outcome based on a black box algorithm but a creative director disagrees with what the machine says?</p><p><strong>Jenna:</strong> Yes, that’s interesting! Who are you going to believe; the experienced ad person who’s been in the industry for 20 years or the machine that’s been around a couple of months but is now analyzing the audience in real time with a lot of sophistication? It’s a bit of a leap of faith. We will have to think in more nuanced ways and think about the context of the campaign. For example, take a brand awareness campaign, which tends to be more long tail, where we don’t see the affect for a while, versus a product campaign that’s measured with a conversion rate and we can see the uptick straight away. I think some campaigns will be more suited than others to the sorts of approaches that AI offers. I can see a bell curve approach where AI can effectively target a specific group, say early adopters, and then build up data on the influence of that group over time. These types of campaigns have been a bit of a gamble in the past but I think we can now use AI to make a better case for that gamble.</p><p><strong>Helen:</strong> What have you found on the vendor side?</p><p><strong>Jenna:</strong> We’ve had the most success with start ups, and I’d say that’s mostly because start ups are more likely to be open to collaboration with a third party. Our focus is keeping the lines of communication open with everyone. Small start-up vendors are very open to being part of the overall solution that we want to offer. We’ve found that’s been well received by clients too. And we’ve had the best success with companies that have a highly specialized offering that addresses a specific need in the market. We’ve worked with one company that has a very specific product around contextual content, it’s very granular and very effective and it’s clear they will be successful.</p><p><strong>Helen:</strong> There’s a lot of focus on bias and fairness in AI. What are your thoughts on how this applies in advertising?</p><p><strong>Jenna:</strong> I view this as a risk mitigation issue. The advertising industry needs to think carefully because personalization can be counter productive and end up too narrow. If the goal is to personalize, how do we make sure we don’t alienate a person or constrain what’s offered to them? On a broader note, I believe AI needs to be developed by a diverse group. I like the analogy of AI as a baby; it needs to grow up in an environment where it’s exposed to a whole range of views, ethics and morals as the way to keep bias at bay. At an advertising level, we need to make sure we are actively aware of stereotypes or if we use AI, we need to make a point of checking that we aren’t reinforcing a stereotype. I think this risk is real and it’s inevitable, but I don’t think anyone has really tackled managing it at the level of workflow yet. We don’t want to miss the edge cases.</p><p><em>Edited for clarity and brevity.</em>​</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d490f067a7bf" width="1" height="1" alt=""><hr><p><a href="https://medium.com/chatbot-news-daily/ai-in-advertising-d490f067a7bf">Artificial Intelligence in Advertising</a> was originally published in <a href="https://medium.com/chatbot-news-daily">Chatbot News Daily</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Data and the Future of Artificial Intelligence]]></title>
            <link>https://medium.com/@jennaniven/data-and-the-future-of-ai-1ad09efc7aba?source=rss-77f92ddbb1e0------2</link>
            <guid isPermaLink="false">https://medium.com/p/1ad09efc7aba</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[design-thinking]]></category>
            <dc:creator><![CDATA[Jenna Niven]]></dc:creator>
            <pubDate>Thu, 27 Oct 2016 12:50:31 GMT</pubDate>
            <atom:updated>2016-11-06T15:58:48.339Z</atom:updated>
            <content:encoded><![CDATA[<p>“Rangers Use Artificial Intelligence to Fight Poachers”</p><p>“New Cure For Cancer Found? Artificial Intelligence Developed Drug Reverses ‘Warburg Effect’”</p><p>“An Artificial Intelligence Robot Finds Jenna a Husband”</p><p>These headlines may sound a bit far fetched, but they are fast becoming a reality.</p><p>The debate of whether we are in an AI ‘winter’ (all talk, no walk) or an AI ‘spring’ (AI is about the change the world) is raging and has both sides of the fence divided. After spending a few days at the MIT EmTech Digital Conference listening to the some of the sharpest minds in AI from companies like Google, Facebook, Baidu, Tesla and more — I strongly believe we are about to reach a tipping point where AI will fundamentally change just about any industry you can think of.</p><p>However, there is one challenge even the biggest players are struggling with, and that’s data. The quantity and quality of data is fundamental to the success of any AI solution and thanks to the IoT, it is predicted there will be about 44 zettabytes of data in the world by 2020 (that’s 57 times every grain of sand on earth…). So the race to collect is on! Here are the 3 main strategies I see in action;</p><h3>1. User Generated</h3><p>100 million users have slaved away on Pinterest to generate and categorize 71 billion images, ripe for AI solutions. Recently, the company has released a tool that enables you to query an object that sits inside an image. For example, if you see a lamp in a living room image — you can select the lamp and search for exact or similar lamps. All thanks to image recognition, machine learning and you, Pinterest User!</p><h3>2. Acquisition</h3><p>January 13, 2014, Google acquired Nest for $3.2 billion (not a typo). Do you think that Google would pay that amount of money simply to know when your house is too warm? The value in Nest lies in the interconnectedness of the data that it collects — sewing together a picture of human behavior. Google can then begin to develop AI solutions that could potentially know what you need, long before you do.</p><h3>3. Business Model Evolution</h3><p>Tesla has revolutionized the car manufacturing release model. Traditionally, car manufactures would come up with a new innovation, manufacture a new model of the car, then release this to market. Tesla, however releases a car to market with hardware already installed, even when it isn’t yet in use. They use this hardware to collect data from your car to test new innovations in a simulated environment. Once they have validated the feature in house, they then install the software in your car and perform in-field validation using real-time data. Finally, when the feature has proved its worth, it is ‘turned on’ so all Tesla cars have this new feature without you having to buy a brand new model!</p><p>So who will collect the most meaningful data? Who will create the most commercial application? Who will find me a husband? It is yet to be seen, but It’s clear the future of AI will be defined by our imaginations…and data.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1ad09efc7aba" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>