Companies are investing billions of dollars in building their data assets in the hope of becoming data-driven.

What leaders don’t realize is that having the data and using the data for business are two very different things. This is not to say that they won’t benefit from good data. Everyone can benefit from good data and nicely done analytics but that makes you ‘data rich’ not ‘data-driven’.

Becoming Data-first or AI-first is the new battle cry. But what does in really mean? When Jeff Bezos wanted to steer Amazon as an ‘API-first’ service-oriented company in 2002, he released the following mandate:

• All teams will henceforth expose their data and functionality through service interfaces.
• Teams must communicate with each other through these interfaces.
• There will be no other form of inter-process communication allowed: no direct linking, no direct reads of another team’s data store, no shared-memory model, no back-doors whatsoever. The only communication allowed is via service interface calls over the network.
• It doesn’t matter what technology they use.
• All service interfaces, without exception, must be designed from the ground up to be externalizable. That is to say, the team must plan and design to be able to expose the interface to developers in the outside world. No exceptions.

The mandate closed with the instruction ‘Anyone who doesn’t do this will be fired. Thank you; have a nice day!’

This may sound too harsh but Bezos meant it and he wanted to make it clear what was at stake. Amazon’s subsequent transformation as A-Z marketplace and the development and growth of Amazon Web Services are the witness to the effectiveness of his relentless approach. If you truly believe that you want to be Data/AI first, then ask yourself — what is at stake for the business? Is it growth, innovation, competitive advantage, margin, customer trust or a combination of all these and more?

Another question you must ask is if Data/AI is not your first priority today, then what is?

In traditional companies, today rules and processes are at the core of all business decisions. When business rules define what needs to be done and processes define how it needs to be done, entire business becomes a rule-processing machines where people exist to just run it.

The true meaning of Data/AI first strategy is that you put information or prediction at the core of all your business decisions.

Data/AI first approach means breaking down and converting this machinery into business tools and empowering people with the right information and autonomy to decide business outcomes and take business decisions.

Anne Thomas, vice president and analyst at Gartner said, “Digital business demands a rapid response to events. A convergence of events generates a business opportunity, and real-time analytics of those events, as well as current data and wider context data, can be used to influence a decision and generate a successful business outcome.”

The precursor to data/AI driven business strategy is ability to handle business events. Events are the future of business intelligence and business processes. The traditional management approach uses reports and transactions of the past to take decisions of the future — it is reactive by design. It is like driving looking in the rear-view mirror. But the events happen in the present and need intelligent response in the present.

“The degree of agility or responsiveness of a company can be measured by its average lead-time in intelligently and decisively responding to critical business events both positive e.g. new customer order, customer appreciations, new business opportunity, new product idea, new regulations and negative e.g. machine failure, raw material delay, customer churn, fraudulent transactions, cyber attacks, low inventory etc.”

The rule-based strategy is like procedural programming where the procedure and rules take precedence over the business outcome or event response. Processes and tasks are pre-defined based on a set of business rules. An events-driven strategy is like object-oriented programming where business outcome and response to business events take precedence. Business processes are a set of ‘event responses’ derived from the lessons learnt from processing events and solving problems to get the desired outcome.

“Application of Data and AI facilitate business outcomes in event-driven business strategy by providing context and understanding to events and how they impact the expected outcome.”

Event Driven Architecture (EDA) is a key technology approach to delivering this goal,” says Thomas. “Digital business demands a rapid response to events. Organisations must be able to respond to and take advantage of ‘business moments‘ and these real-time requirements are driving CIOs to make their application software more event-driven

Data/AI provide the ‘business context’ which when merged with the event stream allows identifying the underlying situation, analyze its root cause and develop potential next actions. Understanding of events and their causes is the fastest route to achieve the expected business outcome. The fastest route to any given problem cannot be known beforehand. It is an iterative process that evolves with the events, information, decision, action, result (next event) and so on. It is like a game of chess where every next action depends upon the past actions and response from the environment. Over time, some of the event sequences can be standardized as named strategies or standard operating procedures (SOP).

In the pre-digital era, business processes were stable and outcomes were predictable. This allowed companies to develop lean and robust processes that served their purpose most of the time. In today’s world, variability and volume of business events is so high that it renders futile any attempt to design an overarching process. Responsiveness beats efficiency to the second place. Lack of responsiveness mean lack of competitiveness and lack of trust from the customers. There is no fun in being the most efficient ‘out of business’ company. And that’s what blind adherence to static processes leads to.

In the digital era, analyzing and responding to a critical event takes highest priority. Solving the problem is important, the steps taken to solve it are not. Contextual guidelines, policies and heuristics are more valuable to problem solving than standard operating procedures.The effectiveness of the problem solving activity can be strengthened by empowering people with ‘data-rich’ business systems and tools. The strength of a business system is to pull in streams of structures and unstructured information to drive and manage the lifecycle of business events to achieve business outcomes. But if a person isn’t able to take the decisions and actions in the real-time, all the data and AI in world won’t help.

Ultimately, it’s the ‘ability to respond to events’ that distinguishes data-driven companies from data-rich companines. It is the repsonse that really counts.

“This business event lifecycle is ultimately being driven by big data. Multiple points of input are correlated against (run against) business rules, policies, computer inference, and from this, dashboards and reports of the results are provided to enable human inference and human decision-making, allowing action to be taken or captured, to which further information is applied informing the next step to take. Thus, the “process” is not a hard-wired workflow per se, but a template for guiding actions and decision-making through a host of input run against a set of rules and sub-tasks defined within the system.” Excerpt from the book ‘How Knowledge Workers Get Things Done: Real-World Adaptive Case Management’ by Nathaniel Palmer and Max Pucher

The only way to be data-driven is to use data at every possible decision point for every important business event to informed action and create more data for learning. It means being responsive to critical business events and to incentivize experimentation, creative problem solving and continuous learning. It means getting rid of rigid ‘blind’ rules and replacing them with analytical insights that have real contextual significance in real-time. This requires a change in attitude, culture and management philosophy of the organization. If you are not simultaneously working on these ‘soft aspects’ of business, investments in tools and technology alone cannot cause a transformation.

About the Author:

Somil Gupta is an AI Strategy Advisor for Nordic Manufacturing and Retail Companies. Before starting this practice, Somil led Digital Business Development for digital and AI solutions for Bosch in the Nordics.

In this role, he was responsible for taking Digital Solutions to market, consulted C-Suite executives in understanding the disruptive impact of digital technologies e.g. IOT, AI, Blockchain and AR/VR and plan organization-wide systems and processes to create business value in customer’s business segments. Somil specializes in Agile Business Development, Agile Business Operations and Technology monetization.

Using writing, consulting and trainings as a way to share his expertise, Somil helps leaders and managers to get ready for disruptions and ride the wave of change and leverage technology strategically in the fast changing world.

Follow Somil on Linkedin: https://www.linkedin.com/in/somilguptaai/

Lean-Agile Software development has taken the world by the storm and it’s encouraging and delightful to see companies adopting it around the world. Agile brings a never seen before speed and responsiveness in the software delivery. More that that, it has put the customer at the center of the software development so that every development and release is based on immediate and real customer feedback. Or is it?

It is interesting to note that there are two very important topics in SAFe that are still not very concrete. These two are the role description of the Business Owner and the process of ‘Release on Demand’. Everyone will agree that the role of the business owner is the most important role in the agile development. BO is supposed to provide inputs on what features are needed by the customer and what economic value should those features have and when shall they be released. But there is no way to ensure that what BO decides is really the voice of the customer. The role of the BO has been described in a great detail ‘within’ the SAFe framework. But their interaction outside the SAFe framework and their decision making model is still undefined. How does a Business Owner know what features must be developed? How do we ensure that they are the right features? And most importantly, who in the organization should be the business owner?

The business owner as defined in the agile context should be someone who must be closer to the customer. Someone who has ‘ears to the ground’, so to speak. The most suitable candidates for this role would be the sales and service teams. I have not seen many organizations where sales and service teams are fairly represented in agile development let alone owning it. Usually, this role is take over by the Business Unit heads who act as the customer proxy but is that truly agile? This is my view that the biggest reason why agile programs fail is due to this very reason, business doesn’t always know what customers want. They either guess or use traditional methodologies to define requirements which cannot be tested and validated. The business hypothesis written down in the epics at the beginning of the initiatives remain hypothesis because no one bothers to test them.

Then there is a challenge with the process of taking product and features to market. Most sales and marketing teams are still not agile. Their ability to continuously learn and take new products out to customers is severely limited. Important activities like sales enablement and go-to-market strategy are developed as an afterthought. As a result, digital products adorn the shelves waiting for real customer feedback. Without products getting sold and used by the customers, the feedback that developers get is not the real feedback. And the work that they do in the absence of real feedback is not development, it is rework.

But that doesn’t sound right! You might think someone would’ve talked about it. Why doesn’t this show up in management reviews? Well, my guess is that SAFe community has thought about this problem and devised a workaround called ‘Release on Demand’. The coupling between deployment and release is disconnected in Agile and DevOps. The intention is to allow the developers to release their code in production where it can be released by the business as they deem fit. While this model works great for small improvements and enhancements, it is not a viable model for major feature releases or testing out completely new ideas.

Usually before a new feature is launched, it has to go through many checks and approvals. A typical app would require legal clearance, Data Protection clearance, pricing and margin approval from finance, launch planning from marketing, sales enablement acceptance and in some cases operations and supply chain planning. These organizational processes are not agile. They can be, but they are not. And they are not in the purview of SAFe which already has its hands full with orchestrating across the organizational technical capacity. They have gone to great details to ensure that things work well within the SAFe framework. But every framework needs to have boundaries. In SAFe, there is a clear boundary between development and business. And the complete job description of Business Owners and definition of Business Development processes is outside that boundary. The onus really is on the business to feed the Agile machinery with the right requirements and feedback to get maximum possible value from them. So how does one go about the commercial aspects of Agile?

The answer to this puzzle is a dedicated Digital Business Development function whose role is to make the commercial side of the organization agile. Digital Business Developers are agilists at heart who like to run with their development colleagues and clear the path for them and their products so that they can reach the real customers and get the real feedback. Digital Business Development owns the agile go-to-market process between the development and release. They collaborate with business owners to define the business strategy and collaborate with the Agile development team for its implementation. They are ones who take the lead on taking new products and solutions to markets and front-end digital sales along with the sales and service teams.

Within Digital Business Development, there is a clear distinction between Business Sponsor and Business Owner. A business sponsor could be any one — Business Unit Head, CDO, CEO or CIO but the business ownership always lies with the Digital Business Development team. It is ultimately their responsibility to lobby with the management to get commercial approvals and work with the sales teams to sell and deploy it. They are thought leaders who launch organization’s brilliant ideas out in the world. And finally, they are responsible for monetization of digital investments through sales, partnerships and other forms of business development activities.

I am happy to see that many organizations have started creating the Digital Business Development roles and some of my colleagues have started this journey. I will be writing more on this exciting new function in the upcoming articles. If you’d like to know what Digital Business Development can do for your business from innovation to monetization, feel free to reach me at the below coordinates.

About the Author:

Somil Gupta is an AI Strategy Advisor for Nordic Manufacturing and Retail Companies. Before starting this practice, Somil led Digital Business Development for digital and AI solutions for Bosch in the Nordics.

In this role, he was responsible for taking Digital Solutions to market, consulted C-Suite executives in understanding the disruptive impact of digital technologies e.g. IOT, AI, Blockchain and AR/VR and plan organization-wide systems and processes to create business value in customer’s business segments. Somil specializes in Agile Business Development, Agile Business Operations and Technology monetization.

Using writing, consulting and trainings as a way to share his expertise, Somil helps leaders and managers to get ready for disruptions and ride the wave of change and leverage technology strategically in the fast changing world.

Before we get into the #PeopleFirstAI, let’s first understand the prevalent paradigm today. I call it Data-First AI. Most enterprise initiatives today are run by the data first paradigm. It started with the famous quote Data is the new oil and how every organization must become a data-driven organization. Let me break it to you as gently as I possibly can. You are already a data-driven organization! Unless you make business decisions on the flip of a coin, you are data driven. If you use any type of ERP or even excel based transaction tracking system, you are data-driven. So what has really changed?

The change has just come not from the data per se. Of course, you now have access to a higher volume and variety of data that ever before. The real change has come from the availability of analytical insights and critical business data not just to managers but to individuals deeper in the organizational hierarchy. Before the data revolution, access to data was proportional to your position in the hierarchy. With the new data-driven paradigm, that access can be made available to anyone on a need-to-know basis.

Trending AI Articles:

1. Bursting the Jargon bubbles — Deep Learning

2. How Can We Improve the Quality of Our Data?

3. AI, Machine Learning, & Deep Learning Explained in 5 Minutes

4. Reinforcement Learning for Autonomous Vehicle Route Optimisation

Organization and business are all based on human interactions and the data is the outcome of these human interactions. If we try to change the data without changing the human interactions is like trying to correct the decisions without changing the decision-making competence. The unfortunate reality is this — people are not that good at interpreting data and rationally acting on it. We are much less rational than what we imagine ourselves to be. If you don’t believe, think about your last impulse purchase. What came first — the decision or the logic? If you are like most people, you made the decision and then try to justify it with logic. Logic and rationality is needed by our conscious mind. But the ‘real-time’ instinctive decisions are taken from the subconscious mind that doesn’t necessarily requires logic. It goes by what feels right. And this is where the data-first AI approach starts faltering.

Now coming to #PeopleFirstAI, lets do a thought experiment. Imagine you are responsible for making an important decision — inventory level in a regional warehouse. The cost of bad decision (stock-out or overstocking) is $1Mn and benefit of a good decision is $5Mn. The system gives you the recommendation for the level but somehow it doesn’t feel right. What would you do? If you decide to follow your gut, you might make a mistake and will have to own the wrong decision. Is your organization processes mature enough to allow that learning curve? And if you blindly follow the system, then who is eventually accountable if the system makes mistakes (and it will make a lot of them)? One of the main goal on #PeopleFirstAI is to resolve this problem of accountability.

The biggest issue of AI adoption in Enterprises is not the prediction, it is the judgement. The data-first approach focuses on the use-case, modeling, data management and deployment activities needed to developing and making the model available. The judgement aspect of the prediction is largely missing in this view. The prediction engine itself cannot create value unless there is a robust decision-modeling system in place to support the prediction engines. The biggest implication of enterprise-wide AI adoption is decentralization of decision authority. For the last five decades, there have been several debates on the merits and demerits of decentralization. With AI, there is no more debate. It has to be decentralized. Otherwise, what is point of using the predictions if the front-line person has to first ask for permission! The value is not created by making predictions, the value is created by acting on them. But if front-line employees are not properly augmented and capable of taking those decision, AI will remain just a fancy prediction tool.

The next important aspect from leadership point of view is the so-called ‘AI-First strategy’ for ‘AI-first organizations’. It’s a popular battle-cry from the CXO office to go AI-first but very few companies understand the implications of AI-first. In a 2018 article, Ajay Agrawal and Avi Goldfarb (author of prediction machines) and Joshua Gans (Google) explain, “Adopting an AI-first strategy is a commitment to prioritize prediction quality and to support the machine learning process, even at the cost of short-term factors such as consumer satisfaction and operational performance.”

Are you really AI-first? Would you be willing to incur short-term costs (and maybe even losses) for a few quarters so that you can train your AI to make more accurate predictions? Are you willing to experiment with AI for your most important customer risking customer satisfaction in an existing business? Then what exactly is the meaning of AI-First? Is it really a strategy or just wild aspirations? And if it is a strategy, could there be another way?

What is #PeopleFirstAI and how does it help?

According to a recent study by McKinsey, only 6% companies have embedded AI into formal decision making and execution processes of their frontline employees and only 16% participants believe that their employees actually trust AI-generated insights. #PeopleFirstAI strategy is a human augmentation and business process oriented strategy that focuses on frontline AI adoption, employee readiness and process maturity. Unless AI is ingrained deep into core business logic and critical business processes, AI-first will remain a pipe dream for most companies.

Another point that differentiates #PeopleFirstAI with other approaches is the focus on transformation and orchestration of core business processes. According to the same McKinsey study, only 21% companies have embedded AI in several parts of the business. Adoption of AI into peripheral processes will not create any significant transformation. It is the orchestration of various value stream processes across the organization that creates high impact. Additionally, inability to transform business core leaves companies vulnerable to disruption because the disrupters don’t attack the peripheral processes — they attack the core business processes. The study concluded that one critical factor of using AI effectively is an organization’s digitalization journey is transforming the core parts of its business. The biggest challenge of any organization is not to automate a single process chain say invoice processing, but to orchestrate it with other business process e.g. payment processing, vendors rating, procurement etc. This is another area where Data-first strategy misses the big picture by focusing on isolated (and sometimes peripheral) use-cases but not on the overall process maturity and orchestration.

A people-first strategy starts with accountability and decision-modeling. It first aims to strengthen and augments human capability before integrating automation. A people-first strategy understands that even the best AI system will be most accurate for only roughly 20% of the most-likely scenarios. The less likely, but equally significant scenarios still need to be dealt by humans. And lastly, a people-first AI strategy accounts for ‘in-process’ or ‘on-the-job’ training of AI. It considers AI not as a business expert rather as an intern who is eager to learn and perform but needs to be sufficiently trained first. Isn’t that a more realistic representation of AI’s current capabilities? Isn’t that not the real AI-First strategy?

Finally, the most important core competence needed for any organization today is resilience. I define organizational resilience as the ability to handle variability — both planned and unplanned without sacrificing scale, productivity and profitability. Planed variability is introduced in the form of broader product portfolio and personalized customer service. Unplanned variability is introduced by demand uncertainty and supply-side disruptions e.g. machine failure, logistics failure and supplier failures. Both these types variability takes a toll on productivity and profitability.

In 2016 study showed that as much as 33% of all organizations reported losses of more than $1 million over that year due to supply chain disruptions. A large part of this costs could have been avoided it the organizations were resilient. It is an incredibly important and difficult goal to achieve and yet there isn’t much discussion on it in the business forums. AI has been found extremely valuable to detect disruptions at global scale. But detecting disruptions is just one aspect of resilience. Ability to effectively manage disruptions also requires process and business-level maturity. To add to the complexity, if AI is not properly integrated into the business processes, it can even add significant uncertainty and variability due to modeling tolerances and errors. That’s where #PeopleFirstAI complements the data-first strategy and addresses the problem of resilience in two ways

1. By augmenting human capabilities to interpret and act on data

2. By strengthening the process capability and resilience to handle variability

Considering the challenges that most organizations are facing or will face, building organizational resilience must be the top priority for the leadership. And to develop this resilience, workforce augmentation through #PeopleFirstAI strategy is absolutely mandatory. A companies AI maturity will not be judged by how accurate predictions it can make, rather by how quickly and intelligently can its front-line employees act in the face of disruption. Resilience is not an accident. It is a critical competence that needs to be deliberately developed over time. In the next course of articles, I will explain #PeopleFirstAI in greater detail. If you’ve enjoyed this article and understood the necessity, do share the claps. If you want to know more, feel free to contact me.

Disclaimer: The views shared in this article are personal and do not represent those of any affiliate.

About the author:

Somil Gupta is a leading strategy advisor for AI to Nordic MNCs. He is also the author of the upcoming book Profit AI. Somil has led the business development for digital and AI solutions for Bosch in the Nordics. In this role, he consulted C-Suite executives in understanding the disruptive impact of digital technologies e.g. IOT, AI, Blockchain and AR/VR and plan organization-wide systems and processes to create business value in customer’s business segments. Somil is passionate about socio-economic and cultural impact of AI and other new age technologies on business, jobs and life. He believes #PeopleFirstAI and enhancing the human element in technology and creating a human-centric technologically-enhanced society. He has spoken and written on human augmentation and is the go-to guy when it comes to developing people-driven technology readiness and monetizing new technology for business and personal success. Using writing and speaking as a way to share his expertise, Somil coaches and prepares leaders, professionals and businessmen to get ready and ride the wave of change and leverage technology strategically in the fast changing world.

Don’t forget to give us your 👏 !

Why #PeopleFirstAI is mission-critical for Enterprise AI Strategy and how can it accelerate AI… was originally published in Becoming Human: Artificial Intelligence Magazine on Medium, where people are continuing the conversation by highlighting and responding to this story.

Artificial Intelligence and its impact on employment or ‘Jobs’ is a hotly debated topic in almost every forum be it business, social or governance. There have been several studies on the topic and a myriad of articles and publications. The 2013 Oxford study predicted that more than 47% of the current jobs are at risk of automation. You can easily find hundreds of articles with much deeper analysis on what kind of jobs will be impacted and in what time frame. Most of these articles will have embedded advertisement of training programs and others modes of income. But how likely it really is and is it really going to be that bad?

I believe that AI and automation will take away a significant percentage of ‘jobs’ and people should be happy about it. I am a firm believer in the power of AI to augment human capabilities and transform world economy. But it is not blind faith. There are several reasons why AI can actually be a force of good in careers despite its seemingly negative initial impacts. And these reasons are often ignored by mainstream media because well, negativity and fears sell better than positivity. You probably wouldn’t have opened this article if the title was changed to ‘Why AI will bring prosperity to your lives’. You see what I mean!

Trending AI Articles:

1. Deep Learning Book Notes, Chapter 1

2. Deep Learning Book Notes, Chapter 2

3. Machines Demonstrate Self-Awareness

4. Visual Music & Machine Learning Workshop for Kids

Before we get deeper into the phenomena of ‘job loss’, lets first understand the history of jobs. The original meaning of job was ‘a piece of work that needs to be done’. By 1650’s, the meaning changed to “work done for pay” or a low mean lucrative busy affair or petty, piddling work; a piece of chance work. Since, this was before the concept of skilled or specialized jobs, it implied a task that can be done by anyone. The present understanding of the word is a remnant of the industrial era and referred to the ‘set of repetitive activities done under a fixed procedure/sequence’. Even today, the word is used in the same parlance in most industrial setups e.g. a ‘paint job’ or a ‘job schedule’. When you give your car for service, the service engineer prepares a ‘job card’ for your vehicle that contains a sequence of activities that needs be done. The procedure or sequence of steps is as important as the steps involved in the task. That’s where the term ‘job description’ comes from. It is a list of repetitive set of tasks that needs to be performed in a predetermined sequence to produce a well-defined output.

The concept of fixed actions in a fixed sequence is derived from the design of machines because this is why machines were created. In the pre-industrial economy, all tasks were done by human labor. It took a lot of time and the output was highly variable. The inventors of machines at the time would spend countless hours understanding the actions and how can they be done faster and with less variability through machines. They tried to ‘automate’ the jobs! Introduction of steams engines and later electricity helped design faster and stronger machines that could do these jobs much faster and with little variance and that led to displacement of artisans and craftsmen forcing them to take up ‘employment’ and execute the jobs with machines.

Now fast forward to 21st century and to modern workplace. We are transitioning from the industrial age to the information age. Thanks to the computers, our work has become more ‘informational’ than ‘industrial’ in the last 50 years. But the remnant of that age — jobs, training, hierarchy etc are still all around us. If you really think carefully, you will realize that AI and automation are not really ‘taking away’ the jobs from us. They will simply ‘take back’ what was rightfully theirs. For centuries, Humans have designed machines that do the repetitive sequential jobs under human supervision. The way I see it, AI will simply take out the robots out of humans freeing us to do more creative and meaningful work with the help of the new intelligent machines aka Bots.

Limited Intelligence

Even though the bots are highly intelligent in the ‘functional’ aspects of our work, they have limited intelligence in the contextual aspects. In my earlier article, I have explained how AI operates in a very narrow context space. There are very few tasks in modern workplace that are strictly functional. If there were, they would have been automated already. There was a time when every organization had an army of assistants or secretaries who would spend countless hours on phone trying to find the right slots for meetings and appointments and on taking dictations and printing it on typewriters. Then came electronic calendars, word processors and printers that automated that part of the job. But did assistants disappear? Oh, they came back stronger than ever. Freed from the robotic tasks of ‘fixing’ the meeting, today they optimize executives’ precious time. And it’s not just the secretaries, every jobs has benefited by giving back robotic tasks back to the machines. The accountants became financial advisors, the shift supervisors became productivity managers and we can go on and on.

When you take away the robot inside a person, what remains is context and creativity. When you take away the repetition, you make way for innovation.

Mass customization and the challenge of variability

Across the industries, customers want personalized products and merchandise. Marketplaces like Amazon have a long list of customization accessories like decals, stickers and back covers for smartphones and laptops and designer versions of mature products like mice, bags and speakers. The demand for customized designer products in not limited by lack of demand but by the lack of supply. Most large companies have adopted the lean manufacturing principle to increase productivity, quality and cost competitiveness at the cost of customization. Customization is costly because it reduces plant throughput and increases the cost of quality and service. AI and Automation will drastically reduce the time and cost of customization making more customized products available at lower price. As significant portion of robotic tasks will be taken over by bots, more and more human effort will be required in customization and design — art.

Informational age will split the work content clearly into what is robotic and needs to be done by bots and what is creative and needs to be done by humans. The functional aspects like ‘preparing a balance sheet’ and contextual aspects like ‘providing financial advisory’.

Human Learning

Now lets come the core cause of the mass fear against AI. For years and decades, the only source of learning for most people were academia and corporate trainings. Both these institutes are remnant of the industrial era and use the pedagogy and instruction design of the industrial age where a person need to be ‘taught’ i.e. imparted the necessary skill through repetition of instruction. Repetition is the keyword here and I hope you’d have got where I am going with this. Informational age changes it completely. We are in the agile age now. You conceptualize, design, develop, monetize and move on. There are no longer repetitions in the macro-world and soon there will be few in the micro-world. Repetitions are the domain of the machines.

The core cause of fear and feeling of doom is the system of learning. Just like everything else in the software world, even the learning has gone agile. The reason software professionals don’t fear AI as much as other professionals is because most of them have adopted or in the process of adopting ‘Agile Learning’. They know that no university or company can train them continuously with every new updates. Dozens of my IT colleagues and friends know that they need to learn continuously and keep ‘re-certifying’ their knowledge even in those areas where they are already experts. In fact, you can no longer call yourself ‘an expert’. That term is dead and it refers to old, stale knowledge that may not even be relevant. Modern ‘Expertise’ is a constant learning process where you hustle every day to maintain your edge.

Assisted Self Learning

The only way to get over the fear of job loss and unemployment is to learn continuously every day. You must commit to daily self-learning for at least 1 to 2 hours every day just to stay sharp. You need probably more to get better. In the industrial age, quality and productivity were the gold standard. In the informational age, it is learning and agility. The information is out there for you to consume. Still, some of us would find it difficult to start and adopt the process. For such aspirants, I will suggest Assisted Self Learning. Find yourself a learning coach who would help you get started and keep you on track. I have had the privilege to coach a few professionals and get them started and the experience was enriching and illuminating for all of us. Feel free to contact me to on LinkedIn, Instagram or telegram. Subscribe to my hashtag #PeopleFirstAI on LinkedIn and Instagram. I also invite you to join my free meetup group.

AI and Bots are the new machinery of the informational age and you don’t need to be afraid of it. You just need to master it. Welcome to the knowledge age where Knowledge is the only currency and agility the only criteria. Happy Learning!

Disclaimer: The views shared in this article are personal and do not represent those of any affiliate.

About the author:

Somil Gupta is a leading strategy advisor for AI to Nordic MNCs. He is also the author of the upcoming book Profit AI. Somil has led the business development for digital and AI solutions for Bosch in the Nordics. In this role, he consulted C-Suite executives in understanding the disruptive impact of digital technologies e.g. IOT, AI, Blockchain and AR/VR and plan organization-wide systems and processes to create business value in customer’s business segments. Somil is passionate about socio-economic and cultural impact of AI and other new age technologies on business, jobs and life. He believes #PeopleFirstAI and enhancing the human element in technology and creating a human-centric technologically-enhanced society. He has spoken and written on human augmentation and is the go-to guy when it comes to developing people-driven technology readiness and monetizing new technology for business and personal success. Using writing and speaking as a way to share his expertise, Somil coaches and prepares leaders, professionals and businessmen to get ready and ride the wave of change and leverage technology strategically in the fast changing world.

Don’t forget to give us your 👏 !

Why AI will take your ‘Jobs’ and you should be excited about it was originally published in Becoming Human: Artificial Intelligence Magazine on Medium, where people are continuing the conversation by highlighting and responding to this story.

According to the new paper, racial biases could inadvertently be built into healthcare algorithms if proper care is not taken while designing the system. Healthcare delivery models may also be skewed in the absence of sufficient genetic data points or to favor ‘commercial benefits’ instead of offering proper care. The big question is who should take accountability for such mishaps? Should it be the physician or the hospital that are using it? Or should it be the Healthcare AI system vendor? And the bigger question is how we can avoid the risks and liabilities of such biases when we know that they are inherently unavoidable.

As quoted in the paper by Implementing Machine Learning in Health Care — Addressing Ethical Challenges by Drs. Danton S. Char, Nigam H. Shah and David Magnus. The report was published by Financial TImes a few days ago.

“As clinical medicine moves progressively toward a shift-based model, the number of clinicians who have followed diseases from their presentation through their ultimate outcome is decreasing. This trend underscores the opportunity for machine learning and approaches based on artificial intelligence in health care — but it could also give such tools unintended power and authority. The collective medical mind is becoming the combination of published literature and the data captured in health care systems, as opposed to individual clinical experience. ”

Trending AI Articles:

1. Basics of Neural Network

2. TensorFlow Object Detection API tutorial

3. TensorFlow Object Detection API: basics of detection (2/2)

4. Understanding and building Generative Adversarial Networks(GANs)- Deep Learning with PyTorch

What the researchers are highlighting in this paper is the future scenario when clinicians will become increasingly dependent upon the AI-based expert systems for not just diagnosis support but for advice on treatment. And then racial and other biases can play out in unpredictable and uncontrollable ways. As more and more clinicians depend upon the ‘system’, the lesser individual power any one physician will have to identify and correct the errors committed by AI. In the absence of this constant critical feedback, these errors and biases will only get reinforced into the system and it will become even more difficult to correct them. The solution this paper offers is to bind the healthcare systems by the same code of ethics that have been guiding the physicians for centuries. But how do we ‘teach’ AI something it doesn’t inherently ‘understand’ — value and respect for human life? Or is there another way?

Following a human-centered design of AI systems ensures human oversight and accountability to key decisions and actions. When the stakes are high, Human decision-making becomes more empathetic than rational and we cannot teach that to the AI model. AI operates in a very narrow context space consisting only of scenarios that it has been trained on. A simple Pareto analysis will show that only 20% of the scenarios will occur 80% times. That means the relative probability of the majority 80% possible scenarios is so low that either the model will not be able to detect such scenarios or will not be able to prescribe the right course of actions. Potentially, this mean the AI system can err 20% of times simply because it lacks the sufficient data points to make an accurate prediction. When the stakes are high — as they are in the healthcare space, 20% is scary. For a lot of medical conditions, even 1% error rate would be a nightmare for the care providers.

Keeping humans in loop can provide the necessary oversight and accountability as Humans are much more efficient in handling those 20% cases which require more context than computation. But humans can only take action if the AI ecosystem is designed around the human needs. Automation and Augmentation are not exclusive, in fact they inclusive. No AI system can work efficiently unless the teams operating them are also augmented with knowledge and vice-versa.

From the design point of view, Physicians and care giver who want to use these advanced tools must also understand the risks and limitations of these systems. While they need not become machine learning experts but they need enough education about these machine-learning systems to understand how it works — the underlying hypothesis, assumptions, set of models and the datasets.

“Remaining ignorant about the construction of machine-learning systems or allowing them to be constructed as black boxes could lead to ethically problematic outcomes.”

But like in every other AI applications, the problem of transparency is a human problem rather than a technical one. The two things that create competitive advantage for the vendors of Healthcare AI systems are their proprietary model and datasets. Additionally, it’s their exclusive knowledge and expertise on these systems that gives them the relative power against the physicians. More transparency and education means losing that competitive advantage. No vendor would willingly give away their intellectual property to educate their users. And the more physicians and clinicians understand the core technology, the more power they have over the vendors. Vendors and their investors don’t like that at all. So then how to resolve this deadlock?

Human-centered design and PeoplefirstAI can circumvent this problem by creating enough levers and control interfaces in the AI system so that humans can easily control it. There is a lot of research happening today in the ‘Explainable AI’ space to create the necessary trust and openness in these systems. Giving humans control over control parameters of the AI system along with the ability to visualize and predict the intermediate processing steps can help create a more controllable AI system. The designers need not share their models and datasets with the physicians. Instead, they just need to show them ‘how’ the systems works and how can they control the process. It is a well-established fact that humans fear what they don’t understand and readily embrace what they can control. Yes, the vendors will lose control over their products in the process. Most AI systems use the real-world usage data as a feedback to train their models. With physicians controlling the AI control parameters, it will become much more difficult to integrate this data back into the models. Additionally, the vendors will also lose some control over how the AI works in real-scenarios as a large part of operational and contextual data (on what basis physicians took the decisions) might not be available to them.

Giving up total control over their products might look like a losing bargain to the Healthcare AI vendors but in the long term it will only create openness and trust to drive widespread adoption with lesser risks and liabilities. The winners and losers of the AI game will not be decided on the superiority of their technical products. It will be decided upon who can create the first system that physicians and hospitals find most comfortable and safe for their patients.

In conclusion, the AI systems are complex and susceptible to unpredictable biases and risks that creep in unnoticed due to hidden patters inside the datasets. A blackbox AI solution may look tempting at first to the vendors but considering the risks, it is not a sustainable business model. And healthcare industry is a very special case because the stakes are very high. Even minor errors could create massive risks right from denial of effective care to potentially endangering the lives of countless patients. It is therefore critical to have an open and transparent understanding of inner working of the Healthcare AI systems by the physicians and clinicians. Human-centered design processes can create control levers inside the Healthcare AI system giving control, oversight and accountability to humans and augmenting their capabilities in the process. An open, controllable and predictable AI can foster a higher acceptability and adoption.

Don’t forget to give us your 👏 !

Can AI Discriminating against Patients: How Human-Centered Design can make Healthcare AI safer was originally published in Becoming Human: Artificial Intelligence Magazine on Medium, where people are continuing the conversation by highlighting and responding to this story.

It creates circular logic problem for the companies –

1. We want AI to prevent errors by learning from the data

2. The data used for training the AI contains the same (or similar) errors it attempts to prevent

If the data that is used for training AI contains errors, biases and it derives all its ‘intelligence’ from the data, AI will simply learn how to make those errors. This fact was highlighted by Facebook’s automatic advertisement targeting system that would selectively show the ads to a skewed sub-group of the target audience even though the ad was intended to be inclusive by the advertisers. For example, job ads targeted to both men and women might still be seen by significantly more men.

Training AI is one thing but using it in to solve real world problems is a different ball game. It is important for business professionals that AI is not a lateral expert who can solve their problem. Rather it’s a trainee that needs to be taught on the job before giving it any major responsibility. There is only so much it can learn from the past legacy data if it is not put into the front lines to solve the real-world problems. More than training, AI needs planning and risk management so that it can be deployed.

A Prediction Engine

Economists Ajay Agrawal, Avi Goldfarb, and Joshua Gans call AI a ‘prediction machine’ to explain its economic role in the real-world use-cases. People argue that AI is not just used for prediction but also for classification but from AI’s point of view, classification is also a prediction problem where you predict the class of the input object. I prefer to call it ‘prediction engine’ because like an engine AI models create predictions based on the data and also like engines, they cannot serve any useful purpose on its own. For engines to create real value, they need to be put into a machine. If you put it into a vehicle, it can be used for transporting goods and people (different machines) and if you put it into a generator, it can produce electricity. Just like an engine, the value of AI depends upon the application it is put into and what problems it is used to solve. AI that is trained with the same customer purchase history will produce a different result if it is deployed as a recommendation system versus if it is used for opportunity qualification inside a CRM. It is not the AI model that drives value, it’s the role it plays in different contexts of business that drives its value.

Not a silver bullet

Since AI uses hidden patterns within the data to make predictions. If the data quality is bad, the models are inaccurate. With inaccuracy comes the risks and if business are not prepared for the risks, AI will do more harm than good. Consider the case of the predictive maintenance where AI is expected to have a real impact on preventing the downtime. While everyone focuses on the prevention, there is an equal chance that AI can introduce downtime. If an AI is wrong in its prediction (and it will be), at least two cases are possible

1. AI predicts a failure but it doesn’t happen and you still service the equipment

2. AI doesn’t predict a failure but the machine still fails

In both the cases, AI introduces downtime which may not have occurred in the absence of AI. In the first case, inaccurate prediction causes unnecessary ‘planned’ downtime since the machine needs to be shut down for maintenance even though there is no problem. The second part is even trickier. The more field engineers rely on AI, the lesser is their incentive to apply their judgement. AI will only benefit them if they are able to delegate this diagnostics to the AI to do something more valuable. If they can’t find something more valuable, AI will only create process overhead for them. If they are able to delegate, they have to also accept the risks that come with it. Successful adoption of AI means less oversight and higher probabilities of misdiagnosis causing additional downtime risks than what could have been if the field engineers were vigilant. Data Scientists evaluate these risks in a framework called ‘the confusion matrix’ used to calculate the ‘confidence score’ or ‘accuracy meter’ of AI models.

Risk Mitigation and Strategy

All is not gloomy as AI does add tremendous value to business processes. A very simple way to assess the utility of AI is to understand its economic value. In most prediction cases, AI gives you the value of ‘additional time’ by substituting human time with machine time. Machine time is more scalable and far more valuable because the additional cost of deploying one more AI agent is negligible and its productivity is a thousand times better than humans. But the speed comes at the cost of accuracy and risks. And the value of time is only valuable if it can be utilized effectively. Going back to our predictive maintenance example, if AI predicts a failure 5 days in advance, you simply get a bonus 5 days to act. How you use this bonus time determines the value of AI in your application. If you don’t do anything, there is no value. If all the resources needed for service — time, material and expertise is arranged in 5 days, then breakdown can be avoided. If even one resource is missing, there is no benefit.

Let’s say that all the resources can be made available in time, then the question comes to cost and risks. What kind of costs and risks are acceptable to you and that determines your strategy. If expertise, material and downtime is readily available on demand and immediate service is acceptable to the customer, then a company can keep their teams on standby waiting for the failure to become more certain before they take an action. In such cases, the value of prediction is limited.

But if availability of downtime and expertise is not readily available (e.g. the machines needs to go into a remote site like an underground mine or forest or underwater or offshore), then the prediction is more valuable and cost of inaccurate prediction is much higher. In such case, customers wouldn’t mind a preventive service ahead of deployment even though it is redundant. Same if the machine takes a long time to repair or offsite expertise or special tools or transportation. Each of these factors make reactive service less attractive and make AI-based prediction more valuable.

Note to the business managers

Business managers need not understand how AI works. You don’t need to know how something works as long as you know how to (or how not to) use it. How many people really know how computers or smartphones work? But you use them every day! It is the same with AI.

The important things that business managers need are

1. That you know what AI can and cannot do for you

2. The right mechanism to access its awesome power for your problems

3. A robust strategy to manage costs and mitigate risks

Conclusion

AI is a valuable tool if you can utilize its output — prediction and its economic value — time. It is neither good nor bad but it can be dangerous if used without care and caution. The real value depends not on the AI but on the mechanism that leverages it. Finally, AI is not inherently intelligent. You have to put in the right data and train it in the real-word workflows to make it work

Why AI isn’t what you think and what can you do to make it work for you was originally published in Becoming Human: Artificial Intelligence Magazine on Medium, where people are continuing the conversation by highlighting and responding to this story.