Report by Anthony Lewis, Chair of Windsor Humanists
In February, Windsor Humanists had an excellent talk from John Barron, a retired computer scientist with a lifetime interest in AI. He summarised the development of AI since the 1950s and he gave a brief overview of the key differences between AI programming and the more linear, algorithmic programming of 'normal' computers. He did this without losing most of the audience whilst keeping the more expert listeners interested and engaged as well!
He highlighted the importance of the year 2016 when the Deepmind AlphaZero AI program, now owned by Google but originally developed in the UK, was able to beat human grandmasters at the games of Go and Chess. The program was able to do this by running the games millions of times by itself to learn the best strategies to beat any opponent. No human could ever match this experience of playing so many games in a lifetime. Grandmasters have described the strategies deployed by AlphaZero as 'chess from the gods'. It is now being used to help biochemists map the morphology of all proteins, a development which will lead, inevitably, to important medical advances over the coming decades.
John then ran a highly interactive and enjoyable session using the generative AI web-based program called NightCafe to produce a series of unique images in different artistic styles on a range of topics suggested by the audience. Many of them were rubbish! For example, the Mondrian image was disappointing, but the AI did create some remarkable images. Here are two of them, both on the subject of childbirth. The first image, above, is in the style of a Japanese anime cartoon. The second image, below, was the best image created during the session and is in the style of Dali the Spanish artist. It was clear that the art produced needed human intervention to decide what was good or not but, even so, the AI produced some stunning images with little effort.
A week after the Windsor Humanist session, The Times reported that Kellin Pelrine, an American player, was able consistently to defeat KataGo, an AI program, at the game of GO. Pelrine's victory was made possible by a research firm called FAR AI, which developed a program to probe KataGo for weaknesses. It appears that even 'invincible' AIs like KataGo can be defeated by using other AIs. Perhaps the dystopian AI arms race and AI Armageddon is closer than some fear?