10 years ago I bought an audiobook on AI which I never got round to listening to. It’s still in my online library but I seriously doubt I’ll ever listen to it now. The AI and ML field is just too fast moving. Thanks to the topic’s current popularity, there is a glut of bang-up-to-date audio and video material out there. Much of it is likely to be on the far side of crap, but there’s still much more quality content than I could ever hope to get through.
So, when I came across Stuart Russell’s 2021 series of 4 AI focussed Reith lectures nestling in my ridiculously over-sized podcast feed my first impulse was to lump them in with my poor old audiobook and swipe them away into oblivion. All that stayed my finger was that I’d recently heard Russell described as a world-renowned expert and I was on the hunt for something to make philosophical sense of everything that’s happened following the release of Chat GPT3 last year. Since most experts tend to lecture as if presenting a sleep meditation with the speed turned down, I made a cup of strong coffee and settled back to have a listen.
The first lecture was designed to provide a bit of history and context to the AI debate. It is easy, swept up in the excitement of something shiny and new, to overlook the fact that we don’t fully understand the basics. For instance, before we can really understand intelligent AI we need to understand what we mean by intelligence. What makes us intelligent? What would make an AI intelligent? Are the two things comparable? What is the importance of consciousness: is it crucial to the debate, or, as Russell thinks, is it something of a red herring? These are just some of the questions that we need to keep in mind when considering current deep learning models and future emergent AI systems.
In lecture 2 Russell moves on to a discussion of that old science fiction staple, the killer robot, and the more general use of AI in warfare. The popular debate has been framed by science fiction inventions such as Arnold Schwarzenegger’s Terminator and the malevolent SkyNet, but Russell explains how the reality is quite different. What we should fear are autonomous weapon systems, the kind that play chess with human lives. They have no malevolent intent, but that doesn’t make anything better for their victims. What’s more, such systems are already well within our grasp and we have been lackadaisical at best in setting limits on what they can and should be allowed to do.
In lecture 3 Russell addresses the economic implications of advanced AI systems. This has been covered in the media a great deal in recent months following the release of Chat GPT3 and GPT4, however Russell’s commentary remains highly relevant. We are at the start of an enormous revolution that, if managed well, could enhance human existence beyond our wildest dreams. The alternative is that it could enhance the futures of the fortunate few while leaving the majority behind, or destroy the future for us all. Now is the time to act if we want to make AI work for everyone. It is unlikely that our current economic model will suffice and if humanity is to benefit from AI then we will need to altar many things about today’s world.
The final lecture in the series draws together the threads of the preceding lectures and discusses the possible ways in which humanity can create AI that works for, rather than against us. Russell thinks that we can design AI systems that we will be able to control, but the way to do this is by changing the underlying models upon which they are based. Science fiction has fooled most of us into thinking that the way to deal with AI is through tricking or outsmarting it with human guile. Sadly this is almost certainly the exclusive province of fiction. AI will be smarter than us and will be able to read us better than we can read ourselves. If we want it to work then we need to get it right from the start.
The BBC’s Reith lectures are almost always excellent value, and Russell’s contribution was no exception. He had plenty to say, and said it with a gentle authority that I found particularly engaging. The content was fascinating and the delivery was excellent. I’m happy to say that the strong coffee I’d made at the start remained untouched at my elbow.
The AI revolution remains in its infancy, but there is no doubt that our world is on the brink of substantial change. If we are not careful, the coming AI revolution could spell the end of our species as we know it. Fortunately this is not a foregone conclusion. Thanks to Russell, I was reassured that there are plenty of things we can do to mitigate the negative effects and build a world that fosters human prosperity for all. If you are interested in helping towards that lofty goal then listening to these excellent lectures would be a good place to start. They can be found via the BBC or wherever you get your podcasts.