By Kim Nilsson, CEO
Three weeks ago I attended the inaugural O’Reilly conference on Artificial Intelligence in New York. Being a lifelong sceptic of all things “hyped” and “buzzwordy”, I was curious to know how much of this “new” industry was real, and how much was general excitement.
First a note on the conference itself. I also attended the Strata + Hadoop conference which followed immediately after, and the two events could not be much more dissimilar. Strata + Hadoop has grown to a beast of an event with 10’000 attendees registered. With fifteen tracks in parallel and an air craft hangar (essentially) to feed us all, it is clear that the data industry itself is maturing and becoming mainstream. Picking talks to go to was a tough choice, and the quality of talks high, but I couldn’t shake the feeling of being an ant in a gigantic digital haystack. The AI conference was so different, in that it was a much more intimate affair. I would guess the attendee number was less than 1’000 and there were “only” three tracks in parallel. This allowed more of a community feel to the conference, making it much easier to meet people and network.
I went to the conference unsure of whether AI really was a new thing, or just the latest name for statistics/machine-learning/data science. I left with greater understanding of the advances in this field, and how it could be classified as different from “traditional data science” (who knew we were already at that stage). My thoughts on the talks at the AI conference that they were circa 20% incredible, amazing, cutting-edge break-throughs, and 80% fluff. Which is actually ok. Every new field starts with a lot of buzz and excitement and talking, but as long as there is some substance to the talk, that’s fine.
Among the more spectacular advances in the last year or so are the enormous improvement in speech recognition (now doing better than humans on average!) and image recognition. On the latter, research has gone from identifying a dog or a cat, to being able to describe what is happening in the image. Microsoft’s project on letting blind people “see” is a great example. But there was also criticism. Gary Marcus questioned whether the field of AI is going in the right direction. In his words, “building a better ladder [e.g. a better deep learning network] will not necessarily get you to the moon”.
With a lot of talk lately on the technological singularity, the idea that computers will become smart enough to improve themselves faster than we can keep up, and Elon Musk wanting to go to Mars to avoid killer robots, the question is just how smart the computers are and how far away true AI is?
I was intrigued by a number of speakers mentioning times when they had to limit the capabilities of the computer due to the expectations of humans. Aparna Chennapragada spoke about product development at Google, and explained how they had tested a version of Android that could predict what app the user wanted when they opened their phone, and thus placed that app right at the top of that screen instantly. This feature was promptly rejected by the users, who much preferred to have each app in its determined place, because our brains follow certain learned patterns, including finding an app in its usual place. Brad Sarsfield from Microsoft Hololens project spoke (at Strata + Hadoop) on the opportunities and limitations of augmented reality. He mentioned how it is important to us that items in the augmented reality not disobey natural laws (such as hover in the air), and that items left in a certain place should remain in that place (based on our muscle memory). It is fascinating to realise that as AI allows advances in technology, we need to better understand human psychology in order to adapt the capabilities to what is actually wanted.
So is AI a threat to us? And is it hype or reality? Starting with the latter question, yes, it is hyped. But it is also real. As with any innovation, it needs to go through the hype cycle to reach maturity. The contrast between Strata + Hadoop and the AI conference clearly showed that data science has now emerged from hype to become much more mainstream. I was positively surprised by the amount of actual progress in the field of AI, and expect many advances in the near future. Should we be afraid? No, I don’t think so. We are still a long way away from truly sentient machines, and when they emerge, well, they will be the children of mankind after all and I see no reason for great concern. I, for one, will be proud of our AI offspring and will look forward to their advances.