Learning machine learning
In August 2001, I was a telecoms analyst visiting investors in Tokyo. In one of these meetings, a portfolio manager at a Very Large Fund asked me what would happen now that GPRS meant that all mobile voice calls would be packet-switched and that therefore mobile operators’ voice revenue would disappear within the next 18 months or so.
This was a surprisingly hard question to answer well. It was nonsense, but to explain why it was nonsense you had to work out quite which things the person asking it didn’t know, and what completely incorrect narrative he’d arrived at to think that this was going to happen. He’d heard ‘packet’ and 'mobile' and added 2+2 to get 22.
I’m reminded of this now, after not thinking of it in 15 years, because as AI explodes in some senses I risk being exactly that guy.
As has happened with many technologies before, AI is bursting out of universities and research labs and turning into product, often led by those researchers as they turn entrepreneur and create companies. Lots of things started working, the two most obvious illustrations being the progress for ImageNet and of course AlphaGo. And in parallel, many of these capabilities are being abstracted - they’re being turned into open source frameworks that people can pick up (almost) off the shelf. So, one could argue that AI is undergoing a take-off in practicality and scale that’s going to transform tech just as, in different ways, packets, mobile, or open source did.
This also means, though, that there’s a sort of tech Tourettes' around - people shout ‘AI!’ or ‘MACHINE LEARNING!’ where people once shouted ‘OPEN!’ or ‘PACKETS!’. This stuff is changing the world, yes, but we need context and understanding. 'AI', really, is lots of different things, at lots of different stages. Have you built HAL 9000 or have you written a thousand IF statements?
Back in 2000 and 2001 (and ever since) I spent a lot of my time reading PDFs about mobile - specifications and engineers' conference presentations and technical papers - around all the layers of UMTS, WCDMA, J2ME, MEXE, WML, iAppli, cHTML, FeliCa, ISDB-T and many other things besides, some of which ended up mattering and some of which didn’t. (My long-dormant del.icio.us account has plenty of examples of both). I didn’t spend very much time learning about processor roadmaps, which in hindsight was only partially the right decision - if I had, I might have been more aware of what might start becoming possible around 2007. But none the less, I got to a pretty good understanding of what was going on. More precisely, I got to the point that I knew what I was talking about when I talked about the things I was interested in, and also knew what things I really didn’t know anything about. I knew what I didn’t know.
The same process will happen now with AI within a lot of the tech industry, and indeed all the broader industries that are affected by it. AI brings a blizzard of highly specialist terms and ideas, layered upon each other, that previously only really mattered to people in the field (mostly, in universities and research labs) and people who took a personal interest, and now, suddenly, this starts affecting everyone in technology. So, everyone who hasn't been following AI for the last decade has to catch up.
What strikes me here particularly about AI as opposed to mobile is both that it is actually many different ideas, concepts and capabilities, often quite different, and also it has many different applications. For almost as long as there’s been science fiction people have talked about machines that were as intelligent as people (indeed that’s what the layman would probably understand by ‘artificial intelligence’) and in some senses we would see that with, say, fully autonomous vehicles. But there is also the old line that ‘a computer should never ask something that it should be able to work out’ - a lot of AI will mean that a computer is less stupid, in invisible and inhuman ways. That makes it very broad indeed - over time you will encounter ‘AI’ in some senses in a huge range of things that you can’t talk to or put on YouTube.