This post was originally posted on Towards Data Science - Fast track to the other side of the AI hype collapse

Artificial intelligence, Deep Learning, and Machine Learning are the buzzwords of the moment. Is the excitement surrounding all things AI about to take a plunge? Some technologies that were this hyped at their peak saw interest nosedive. They then struggled to come out the other end of the hype-hungover. Getting to solve real-world problems will help mitigate the fall down the hype-chute. It will also help keep interest and investments going, fast-tracking to more productive AI solutions.

At the peak of inflated expectations

A simple search for machine learning on Google Trends paints a clear picture. The interest in the subjects has increased dramatically over the last couple of years. Searches for Artificial Intelligence and Deep Learning display similar characteristics.

At the same time, we see investments pouring into anything with an AI or Deep Learning label. The government of France recently announced that they plan to invest $1.85 billion (€1.5 billion) into AI research over the next couple of years. The US spent $1.2 billion on unclassified AI research in 2016. These AI investments by the governments of USA and France are dwarfed by the amount of money that the Chinese are throwing at it. It is not only governments that are opening their wallets. Last year, Google rolled out Gradient Venture; a massive venture fund with an AI focus. Microsoft also launched an AI fund last year. Toyota kicked one off as well ($100M). Basis Set Ventures ($136m) and Element.ai ($102M) started. AI rock-star Andrew Ng is pooling $150M into a venture capital fund focusing on AI. According to McKinsey, a total of $26 billion to $39 billion was spent on AI in 2016 alone. I am confident that the number for 2017 at least isn’t much lower.

In Gartner’s list of top 10 Strategic Trends for 2018, they place AI at the very top. Accenture is conveying the very same picture in their report Technology Vision 2018.

Big words are being thrown around. We hear about Artificial intelligence almost every day and in all kinds of contexts. We are reading about self-driving cars and seeing machines win Jeopardy, Chess and Go. They tell us that AI is going to be everywhere. Our lives will be more comfortable and better in every way imaginable. We will not be driving cars. Doctors and nurses will be replaced by incredible robots that never miss a diagnosis. We will be socializing with robots instead of our regular tired human friends. We will keep robots as pets to keep us company as we grow old. But, the very next day we are reading about an AI singularity. Some self-aware general intelligence that lacks any trace of ethics. This synthetic monster will devour all life (or at least all human such) in its pursuit of fulfilling some obscure goal.
We watch YouTube videos about the need to raise our machines as we do our children for them not to grow up thinking about us as a virus of this world.

Some of us have had the pleasure of speaking to Siri, Alexa or any of the other robots. We know that no-one is going to put down their pet dog anytime soon. At least not to replace it with one of these robots. To us, it is evident that the reporting of AI is a bit over the top — a tad on the dramatic side.

Gartner’s Hype Cycle

Having media coverage being a bit too enthusiastic should be expected when the hype itself sits at the zenith of the hype cycle. Gartner, the company behind the Hype cycle, places Machine Learning and Deep Learning at the very peak of its hype. At what they call the Peak of Inflated Expectations. Inflated expectations — it sure sounds like what we are seeing. Gartner describes this phase of the cycle as where “a few success stories take up all the attention and inflate expectation only to be followed by a score of failures”.


Going over the edge

Next up, after sunbathing at the top of the hype cycle, the situation often becomes less fun. Signs of cracks start to emerge in the facade of the hype. As patience is not a key trait of investors, they are going to want to see return on investment sooner rather than later. Some of the more delirious ventures will start to fail and when this starts to happen the glow will begin to fade. The first ones to go under will be one of two types of ventures. First, the ones with incredible ideas that would never be deemed viable if considered soberly. These are ideas that are shot down immediately by others in the know when their stories hit Hacker News or Reddit. Ideas that only ever fool those not familiar with technology. The other kind of ventures that will fail early are those with unsound organizations. Single entrepreneurial engineers without any business know-how or entrepreneurs without any technical skills. The type of businesses that investors usually stay clear of no matter how excellent their pitch.

Investors will start to think twice about putting dollars into AI once the first start-ups implode. They will begin to redirect their capital elsewhere to the next big thing (did someone say 4d printing or quantum computing?). Entry-bars for investments will be higher. At that point interest in AI will plunge.

Consider Blockchain. A technology that leads the hype of Artificial intelligence by a year or two. Blockchain has come a bit further down the hype cycle. It is still up there in the inflated expectations zone, but we are now seeing it creeping over that edge. Investment capital is backing away. Some Blockchain initiatives have failed. Where people previously talked about them solving all our problems, the expectations are now a bit more modest.

Think about 3d printing and Virtual reality. Both of them have come even further through the hype cycle. Both are now emerging out of the Trough of Disillusionment with renewed vigor. Most of us have forgotten the time when they hit peak hype a couple of years ago. Virtual reality at its peak was promoted with applications that were not ready. At the time, we thought it was all sounding very cool. We marveled at the possibilities but had a hard time understanding how to apply it in our everyday life. Today, virtual reality is on the other side of the curve. There is now both real and useful applications. Adoption is growing. Virtual Reality is enjoying a renaissance at the other end of the hype cycle. 3d printing has had a similar journey. We are now seeing it having a real impact, and the usage will likely only increase from here. These technologies have matured. Adoption is growing and both of these technologies today have real-world implications.

Leaving the very peak of the hype cycle is not a bad thing. In fact, most of the applications of new technology emerge at the latter parts of the hype cycle. At Gartner’s so-called Slope of Enlightenment where the technology’s applications start to crystallize.


Time to solve real problems

The big crazy ideas are getting all the attention and all the money. The risk of having this hype-bubble burst is scaring away too much of investments for too long. It might mean that it takes longer to get through to the other end. As it looks right now, the fall into The Trough of Disillusionment could become so deep that viable applications have a hard time finding funding and fail to get off the ground. Smart and talented people that could push AI forward might choose other paths. So, what can we do about it? Short-circuiting the process is possible. Let’s pretend we are already at the end of the Hype Cycle. Let’s focus on actual applications. Let’s take what we know and start applying it wherever it makes sense right at this moment. Let’s not restrict ourselves to the very grandest of ideas. Ideas that will move the bedrock of society. Let us find smaller, more accessible, quicker wins. Those that create value right here and now. Let’s move some of the spotlight away from the ideas that tickle our imagination and feels like science fiction. If something feels too much like science fiction chances are that it is. Give more room to more mundane and “boring” applications. That way we manage expectations and can move more quickly into the productive side of the Hype Cycle.

Why are IT technicians on expensive on-call contracts still being called up in the middle of the night? Waking up only to follow some predefined procedure for fixing some IT-breakdown. In many of the cases where intelligence is in fact needed, the decisions are often simple. Same goes for back-office and operations organizations in banks. I have spent the last ten years on the technology side of Capital Markets. An industry that has been throwing big bucks after technology. Looking to streamline business processes and complying with regulations. Nowadays, real humans only ever step in in cases of anomalies. That is at least the idea. In most of these edge cases, for which the systems cannot decide what to do, a statistical model or an AI could surely do a good enough job.

I have friends researching in medicine, finance, and HR. The same kind of thing can be seen there. None of them are leveraging Artificial intelligence. They have of course all heard about AI. Some of them have even thought about ways that AI could improve their work. However, that is how far they have come.

One of these friends explained to me that they are manually classifying images of HIV-infected blood-cells. These are tech-savvy people who routinely use Google and Apple to classify all kinds of things in their personal photos. But when they are at work, they do the same thing manually without even thinking about it. What has caused this disconnect?

The problem lies in how Artificial intelligence is portrayed. It is not clear to anyone outside of the inner circle how to start applying the technology. They do not realize that the entry barriers are in fact quite low. We need to help these persons recognize that AI, Machine Learning, and Deep Learning are more than just buzz words. They are not obscure and hard to grasp black arts of science fiction. Most of Artificial intelligence is not only entirely comprehensible but often also simple to implement. In most cases, there is no need for massive data-sets. Many problems are solvable by using pre-trained models that can be downloaded from the internet.
It is time to spend less energy on the exploration of possibilities and to focus on delivering solutions to actual problems.