In the 1980s, back when I was a PhD student, big advances were made in AI. But they were nothing like the stunning leaps forward that have been made in the last five years. A series of problems previously thought almost impossible to solve have fallen, one after another – the era of deep learning and AI is truly upon us. But in 2019, we will realise that much of our excitement about AI is overblown.
Every pitch I come across these days has AI in it. Products in almost every area are being “reinvented” with AI at their core. But the current level of investor interest in AI hides the fact that, although AI techniques are useful in solving certain problems, they are not yet applicable in every case. Adding an off-the-shelf algorithm to old software is not necessarily going to teach it new tricks.
Often these companies get funding. However, in reality, the vast majority of them have little AI capability. They are using AI to market existing approaches, or even just assuming that they will be able to hire someone who will know how to insert AI into their product – AI as a commodity skill, like Java programming.
Investments in these companies will crash and burn because AI is still very difficult to really make work. In demos, AI can be shown to do a great job, making investors desperate to part with their cash. But the real world – handling unpredictable conditions, managing those difficult things called “customers”, dealing with all of the exceptions that bedevil our everyday lives – is much more of a challenge.
Driverless cars are a good example. In theory, traffic on our roads should already be flowing freely with autonomous vehicles, communicating with each other in real time. In reality, we are more than a few trials in a garden city away from automotive utopia. I was recently in Amalfi in Italy, on one of those impossibly narrow stretches of road that had obviously been designed for a horse and cart, where I watched two coaches try to pass one another. Somehow, with much gesticulating and uttering of medieval Italian swear words, the drivers managed to temporarily suspend the laws of physics and pass without falling into the sea.
This is exactly the sort of exceptional circumstance that we need to train AIs on. But recreate this scene in snow or in a hailstorm, or with an impatient driver trying to overtake, and you see just how hard it is to turn a theoretically viable product into a safe and commercially successful one.
The reality of the current AI hype is that many of the companies who succeed in securing funding show a woeful lack of understanding of the difficult nature of their problem and of how to test their AI for robustness in the real world. And, while a few will be successful; many others will be unable to navigate their equivalent of that road in Amalfi so smoothly.
Investor enthusiasm will wane with the first big failures, but this won’t be the end of AI. There will be a reckoning while the industry regroups and redefines the problems it is trying to solve. And then there will be a resurgence of interest as those that have understood the question use AI’s powerful tools to create robust solutions.
But the hype will have passed. Investors will no longer trust lab demos but will insist on seeing products live. And after that essential period of creative destruction, the AI revolution will continue to march forward and change our lives forever.
Mike Lynch is founder of Invoke Capital and co-founder of Autonomy Corporation
– Meet the companies fixing depression by stimulating neurons
– An e-bike revolution is about to upend urban transport
– How companies will use AI tackle workplace harassment
– The blockchain needs protecting from quantum hackers
Get the best of WIRED in your inbox every Saturday with the WIRED Weekender newsletter
This article was originally published by WIRED UK
