How Can Artificial Intelligence Guide a Business Response to COVID-19?

Disruptive IT Trends9 minutes readOct 22nd, 2020

Cycle calls out Formative AI, an emerging technology capable of driving businesses through rapid change, as long as humans remain at the wheel

Artificial Intelligence (AI) uses the past to predict the future. The unprecedented nature of the current pandemic would therefore seem to be a poor fit for AI if a business wants to forecast future operating conditions during and post COVID-19. Back in March 2020 at the start of the pandemic, that was absolutely true; it is not true now, several months on, and business leaders are now turning to AI to dynamically respond to rapid changing conditions. This article explains why our company President & COO gate crashed an internal AI project demo, how we are using AI at Unisys to plan for a “new normal” operational future, and why this use of AI is part of an emerging trend recently identified by Gartner.

“Ignore me,” Eric Hutto said in a reassuring voice as he joined our team call, “I’ll just be a fly on the wall”, precisely two minutes before he launched into a series of difficult probing questions about the data, the AI model and the robustness of its predictions pre, during and post COVID-19. No pressure! He was clearly very interested in this in-house AI forecasting tool we had developed at double-quick speed for the leadership team of one of our main business lines. Eric is the President & COO of Unisys, where I work as a Data Scientist. Unisys builds high-performance, security-centric solutions for the most demanding businesses and governments and prides itself on over 145 years of game-changing innovation. Consequently, Unisys is awash with really impressive and far more substantial innovation projects, so why was Eric so interested in our rapidly built, and to be honest, fairly modest AI innovation project?

One clue comes from Eric’s recent article on the need for speed, which aside from convincing me never to get in a car with him, argues that fast-moving organizations need dynamic drivers. Extending this analogy, few would dispute that dynamic drivers need the clearest possible view of the road ahead and, in a modern business context, that means data-driven decision making – not only day-to-day but also in a crisis. Unisys responded impressively quickly to the emerging pandemic, shifting 95% of its global workforce to remote employees in 48 hours, without service reduction or disruption to customer businesses. Such decisive action is doubtless one reason why the company has not been as heavily impacted as other, less fortunate or less dynamic, companies. Those rapid decisions were driven by the best available global and regional data on COVID-19 at the time; they were data-driven decisions, of course taken with a final human assessment of the balance of risk over the available courses of action.

So, much as it hurts the egos of our project team, I suspect that Eric wasn’t especially interested in our AI innovation per se; he was, however, self-evidently interested in whether this type of AI could improve the view of the road ahead and give his leadership team the data they needed to manoeuvre confidently under rapidly changing conditions. He wanted to know whether AI models like ours could adapt their daily, weekly, monthly, quarterly and annual forecasts to the “new normal”, adjusting the “old normal” trends and cycles learnt from many years of pre-pandemic business data, by learning dynamically from data we had accumulated since lockdown. The answer is yes, but there are a couple of challenges, the first of which we’d already solved at the time of the demo; the second can’t be solved by technology alone, it needs the steering hand of a human.

The first challenge stems from the fact that an AI model is normally trained to answer a very specific question. Indeed, to answer the specific question initially posed, we successfully trained an AI model to forecast months ahead with less than 5% error with a high degree of confidence (95%). The snag was that this was only one example question, and the leadership team had dozens of similar but slightly different questions. Moreover, the set of questions was likely to shift and change over time as the pandemic unfolded and business circumstances changed. Unfortunately, whenever we trained our AI to answer a new question, we discovered that although it performed admirably on each new problem, it simultaneously lost the ability to accurately answer all the other questions. To overcome this challenge, we abandoned the idea of training a single AI model and instead built a tool that dynamically trained a brand-new model every time the leadership team asked a new question. In fact, behind the scenes, the tool trains 27 slightly different candidate models and chooses the best performing one. This adaptive use of machine learning (ML) is something that Gartner have since identified as part of an emerging trend and, as I will shortly explain, goes some way towards addressing our second challenge in forecasting during COVID-19.

Gartner’s recent press release lists five major trends they claim will drive technology innovation for the next decade. Their analysts identified these top trends after surveying more than 17,000 emerging technologies and trends, from which they selected a top 30. Scanning through the list of 30 technologies positioned along the familiar Hype Cycle expectation curve, I was pleased to see at least 10 AI-related technologies had made it through from the wider list of 17,000 emerging technologies surveyed. One major trend, Algorithmic Trust, sprang out as something I expected to see, having only recently written about this topic myself. However, another top trend, Formative AI, I’d never even heard of before. Somewhat suspicious that the authors had just invented the term, I read their definition: “a set of emerging AI and related technologies that can dynamically change to respond to situational variances.” They continue, “The most advanced can generate entirely novel models that are targeted to solve specific problems.” This sounds a lot like our internal AI forecasting tool. Maybe Gartner is onto something real after all.

Identifying emerging trends before they become widely known and then raising awareness of them is arguably one of the main contributions of the Hype Cycle. So, I shouldn’t really have been surprised to see an AI technology that I’d never heard of – but I was surprised to find that we were already using it in a high impact application! Evidence, if ever it were needed, that some disruptive technologies can traverse the Hype Cycle with surprising speed. In the words of Brian Burke, research vice president at Gartner,

Emerging technologies are disruptive by nature, but the competitive advantage they provide is not yet well known or proven in the market. Most will take more than five years, and some more than 10 years, to reach the Plateau of Productivity. But some technologies on the Hype Cycle will mature in the near term and technology innovation leaders must understand the opportunities for these technologies, particularly those with transformational or high impact“.

Eric has possibly never heard of Formative AI either, although it wouldn’t surprise me if he had. Either way, I’m sure he’ll be pleased to know we’re making good use of an emerging technology identified by Gartner, what he really needs to know is whether AI can give his team a clearer view of the road ahead. And this brings us back to the second challenge in using technology: an AI model learns from the data you train it on, including any omissions, biases and anomalies. Consequently, when we train an AI model on the COVID-19 impacted data, since March 2020, we get a forecast for business in the “new normal” that looks very different from the “old normal” forecast we get by excluding the pandemic data. When we include old normal and new data then we get a forecast that merges these two “normals”. The snag is that we have no historical business data about how fast a recovery from a pandemic happens and so the model just makes a best guess by blending the old and new trends and cycles – the more old data you include the more this blend looks like the old normal; the less old data you include, the more this blend looks like the new normal.

In terms of Eric’s driving analogy, this means that our AI models give his leaders clearer views of two very different roads ahead – the old normal and new normal routes. Ultimately though, it comes down to the skill and experience of the driver to steer their part of the business towards the old or new normal at the right time – or, perhaps more likely, towards a recovery that lies somewhere between the two. Our Formative AI tool is therefore closer to a driver aid, a forecasting satnav, rather than a self-driving car.

If Gartner is right, as I believe they are, then we can expect to see a lot more such tools in future as businesses and governments turn to AI and ML as a data-driven mechanism to forecast contrasting future scenarios. True, there have been valid criticisms of the technologies underpinning Formative AI – especially the risk of ‘baking in’ human biases – but as Bill Mew, a campaigner for digital ethics points out, the greater risk is the risk of complacency in business and government response to COVID-19,

Much of the focus on training AI and ML has been on eliminating bias – preventing us building our own attitudinal biases into the algorithms. One bias in particular that has been demonstrated in our Covid response has been ‘complacency bias’, our inability to recognise risk until it is too late. We saw this with the financial crisis in 2008 with credit risk and with the current pandemic with health risk (where ‘complacency bias’ may even result in a second wave). Governments and businesses now need to focus not only on how to recover, but also how to avoid making the same mistakes again. Risk appreciation, in regard to looming threats like climate change and cyber crime, will only be effective if ‘complacency bias’ can be eliminated and if we can learn to listen when the alarm bells ring.

So, whether we consider the biases we bake into AI models or the directions we take guided by AI tools, it is important to recognize that the human hand remains very much on the wheel. Armed with that knowledge, placing Formative AI tools in the hands of senior leaders empowers them to combat the risks of ‘complacency bias’. In Eric’s terms, our Formative AI forecasting tool empowers his dynamic drivers to steer a more informed course through COVID-19 and the many possible roads beyond.

Tags-   AI artficial intelligence COVID insight pandemic

About The Author

Simon Price

Simon Price, Ph.D. - Experienced Data Scientist with a wide range of machine learning (ML), artificial intelligence (AI) and natural language processing (NLP) project deliveries for industry, public sector and government clients.