Winter is coming: The alarming state of AI in organizations

by Tim R. Schleicher and Oliver M. Triebel

What a move: Daimler freezes the partnership with BMW and collaborates with NVIDIA on building self-driving cars. It was the right decision because Daimler and BMW would have never managed to counter Waymo or Volkswagen – and because AI is not holding up to the promises of business gurus.

Yes, there are these "enlightened investments during a business downturn that pay major dividends in the future," as the 7-time CIO Mark Settle writes for Forbes. Whether the collaboration between Daimler and NVIDIA will become one of them is yet to be seen. And that's because there may be something wrong with AI in organizations.

Many business leaders tell us that organizations are failing AI and AI is failing organizations.

Most AI prototypes never get implemented. Companies spend millions on AI projects that don't create tangible value. The Economist recently acknowledged that we are not only in an economic recession, but also in an AI downturn. Drawing on our own observations, we do think that’s true.

In the mid-1970s and again in the early 1990s, it was frustration with the progress of AI that led to strong AI winters. Just note how in 1974, the disappointment of the US Defense Research Agency (DARPA) in speech-understanding-research led to severe cuts in AI funding. Yet many years later the technology developed back then, i.e. the hidden Markov models, became essential for machines to understand human language. Like in the seventies, we are approaching another AI winter today. But why this time?

1. Claims about AI’s future didn’t turn out well

In 2015, Elon Musk promised that by 2018, we'll finally have fully self-driving cars. Well, we don't. As in the case of cars, AI in organizations is largely framed as a revolution. Yet, what Machine Learning is best at is sophisticated pattern matching. AI can lead to revolutions, but in most cases will "only" improve efficiency, enable hyper-personalization, and foster accuracy. The conception of AI merely as a revolution has led to deep frustration as this promise remains largely unfulfilled. People are slowly understanding that the future with and of AI might look differently than portrayed.

2. Severe tech and data challenges persist

AI's progress in recent years was not only due to big data and better algorithms. It improved largely due to the availability of unseen computing power – an essential precondition for Deep Learning. However, for advanced tasks like autonomous driving, algorithms are becoming increasingly expensive to train and Moore's law is fading. As if that isn't enough, people are increasingly finding big data full of traps. The reason why we are not sitting in self-driving cars is because they still fail to handle outlier situations that have not been covered by training data. Algorithms don't adapt well to things unseen.

3. Leaders shy away from AI

According to Gartner, a large number of AI projects fail because they are too large-scale from the outset, achieve impact too late, and have subcritical employee involvement. More than that, organizations fail to recruit the right talent, and don't manage to build the mindsets and capabilities for AI. Leaders still don't demystify AI and inspire others for it. But without the employees on board, AI projects are prone to fail. AI requires more leadership, not less.

These developments have left AI in organizations in a paradoxical situation – long before the Covid-19 crisis hit organizations with full force: it's been talked about, employees are still afraid of it, and leaders are holding back their endorsement. Besides, IT and AI spending plans are being significantly altered right now. So AI's future remains rather gloomy.

Yet, we have personally experienced how leaders can turn around AI in their organizations and create value in spite of the crisis.

What leaders should do now

With small-scale projects which solve real life, highly relevant problems and an orchestrated mindset change alongside of it, organizations can reap the benefits of AI. To do so, leaders have to frame AI as a means to level up business performance – and select use cases that most likely won't become a revolution. Secondly, leaders need to understand that for many applications, shallow Machine Learning is very well suited. This version of AI ensures data need and computing costs remain low. And finally, leaders need to establish an AI movement which fosters an AI-like learning culture: it enables employees to use data and AI-approaches to master the challenges of today's business environment­ – and thus creates true value.

The Corona pandemic will propel those ahead who are leading in AI already. Those who don't support AI projects now will fall even further behind in our post-crisis world.

This article is based on insights from our work at LEAD Machine Learning.