10 Sep

Artifictional Intelligence

Posted By Harry Collins


The trouble with artificial intelligence research is that its aim is reproduce the brain not the person. But people aren’t brains. Some critics of artificial intelligence have argued that the body is crucial but this is too narrow a view. To reproduce human intelligence one must understand where human intelligence comes from and that is not the brain but society. Therefore, to reproduce human intelligence we will have to make machines that fit into societies in the way that humans fit. The trouble is we have not noticed that this is where the problem lies and, if we did, we have no idea how to fix it. The fascinating thing about the latest and best approach to AI – deep learning – is that it gets a little closer to that aim and that is why it has been so successful. But it isn’t there yet nor will it be in the foreseeable future.

One can see how the neglect of the social plays out in AI in lots of ways. One is the crazy panic about ‘the singularity’ – the time when super-intelligent computers will take over the world. But the ambition to take over the world comes out of James Bond movies and there is no reason to think that super-intelligent computers will share that socially created ambition; after all, the sociologist, Max Weber, spent some time trying to explain why humans wanted far more than they could consume and put it down to religious belief. Computers just don’t consume the same stuff as humans and they won’t share the religion unless we put it and the associated ambitions into them. Language is another way in which computers can be seen to fit into society. Fluency in language is a social accomplishment since we take what counts as fluency from society in the way a thermometer takes its temperature from its surroundings. With deep learning, computers are getting better at language but there are simple tests that show how far they have to go. The book lists a number of these tests that the reader can try on their computer or mobile phone. Nowadays, with the cloud, we are all in touch with the frontiers of AI so it is easy to test those frontiers with our own little Turing Tests and these will find computers wanting.

Harry Collins is a Fellow of the British Academy, and Distinguished Research Professor in the School of Social Sciences at Cardiff University. His latest book, Artifictional Intelligence: Against Humanity’s Surrender to Computers , is now available from Polity. He is also author of Are We All Scientific Experts Now? (Polity, 2014) and co-author of Why Democracies Need Science (Polity, 2017).