Artificial intelligence is without a doubt one of the most hyped buzzwords of 2017. Every week brings another exciting announcement on the progress of artificial intelligence. Terms like ‘machine learning, deep learning, neural network and reinforcement learning’ have become an everyday part of our vernacular. According to KMPG in 2017, VC investment in artificial intelligence nearly doubled, climbing up to 12B dollar in investments in 2017. Nonetheless, as psychologist and economist Dan Ariely noted:
‘Big data is like teenage sex: everyone talks about it, nobody really knows how to do it, everyone thinks everyone else is doing it, so everyone claims they are doing it’.
I believe we are currently in an era where AI is both overhyped and underrated.
There are 4 reasons why AI is suddenly drawing so much attention:
1. Computers became much more powerful over the last few decades.
For the last couple of years especially, we have seen the computing power increase exponentially. This rise has mostly to do with bundling computer power in cloud services and focusing specifically on AI.
The biggest cloud players in the world – Amazon (AWS), Google (Cloud), Microsoft (Azure) and IBM (Watson) – now deliver intelligence as a service. Raw computer power as a commodity, just like electricity. This is a recent phenomenon and is likely to quickly advance further in the upcoming years.
2. Governments and companies are afraid of missing out, and they are investing heavily.
McKinsey estimated the total investment in 2016 somewhere between 26B and 39B dollars, a 300% increase over the year before. More data can be found in Stanford’s AI Index report. Big companies like Google and Facebook are going through organisational changes to optimize themselves for AI.
3. Extensive research and media hype has made AI look further than it actually is.
There has been a lot of focus on research and the long-term projections on the AI field in general. This has made it look like AI is further away than it actually is.
The fact that both Elon Musk and Stephen Hawking are warning about the risks of superintelligence has drawn a lot of attention to the AI field. The promise of a super intelligent computer triggers our wildest imaginations.
Over the last couple of years, companies like Amazon, Google, Nvidia and IBM have published research papers and proof of concepts on AI. The focus has largely been on how AI is the next logical step from the companies’ point of view. They are publishing
what they have found not only for marketing purposes but also to attract new (data science) talent to their companies.
Still, it does give a skewed perspective of the field at large. And it creates quite a bit of media hype.
4. AI breakthroughs
Finally, there have been a number of breakthroughs in AI. Renewed focus by leading businesses like Google has led to multiple breakthroughs on approaches to AI, like deep learning with convolutional neural networks in 2012. These new approaches had a positive impact on the field at large and sparked a wide range of new enthusiasm and possibilities for the field.
Why AI is overhyped
Marketing around AI is ramping up. AI sells. A lot of companies claim to have ‘AI inside’. While often technically true, it is a little bit like the ‘renewed recipe’ sticker on a fruit salad. In the strictest sense, it's true, but it doesn’t tell you anything.
If we classify AI as ‘all non-human intelligence’, then that Casio calculator from the 1970s possessed AI capabilities as well. However, so far, the AI solutions are not so significant that they shape whole industries, yet.
They are currently mostly focused on the areas of productivity, accuracy increases and costsavings. Can you name a couple of successful pure AI companies? Neither can I, and I’m quite close to AI.
AI is not a magic wand which can solve problems we barely understand. At least not yet, and it’s my assumption that it will take a while before we can just ask raw intelligence to solve our problems for us. And even long-time AI practitioners like Yann LeCun (Facebook) and Marc Hinton (Google) warn against inflating expectations about deep learning too much:
‘If intelligence was a cake, unsupervised learning would be the cake, supervised learning
would be the icing on the cake, and reinforcement learning would be the cherry on the cake. We know how to make the icing and the cherry, but we don’t know how to make the cake.’ – Yann LeCun
AI can be very successful when it is fully embedded in your corporate structure. Nonetheless, we got spoiled by software products which require almost no work to implement. With AI this is different, at least for now.
If you look at the biggest successes of AI, you can see how it still requires a lot of hard work from scientists and engineers. Problems which we understand well, and which we are able to define in strict rules, have seen many breakthroughs in the last couple of years. Current techniques lend themselves well to areas where there is an absolute right (or wrong).
In those areas, AI usually exceeds the expectations. But, it is not for nothing that a lot of the recent breakthroughs involve applications with a clear right outcome, like games.
Meanwhile, completely unsupervised learning is currently best suited to industries where we can accept very high error-rates. Experimentation with recommendations on Amazon is probably OK because even a 50% hitrate would already be great. (You just ignore half of the recommendations). But you cannot apply the same techniques to life and death situations.
Why AI is underrated
Despite the overblown promises, there are some really promising technologies and techniques which are currently finding their way to business applications.
As said above, you don’t have to invent new algorithms to make use of the possibilities of AI. You can you just buy computational power or even AI as a service directly.
More importantly, there is a lot which can be achieved even with comparably low complexity AI. With relatively crude systems you can already outperform many currently manual equivalents as long as you implement machine learning systematically. Areas where AI is
currently showing most of its potential are chatbots, predictive analytics, and machine vision. And other areas will follow soon.
It’s misleading to extrapolate the recent rapid succession of breakthroughs as the state of AI. The successful cases can be matched with an equal number of failures.
However, to dismiss AI as just a buzzword is equally foolish. Yes, there is a lot of inflated hype. And clever marketers will try to sell you snake oil. However, even relatively simple implementation of AI can harvest great results.
In a next blog, I will lay out a number of things which I think you should do in 2018 to take advantage of AI and prepare for the future.