The Centaur Model?
26 March 2020 19:30 - 21:00
Bot Club investigates the possibilities of Garry Kasparov's 'centaur model' as a metaphor for healthy working relationships between humans and artificial intelligences. Human-AI relations are complicated and often downright disfunctional. They are understood through many incompatible mythologies, each formed around their own concerns and backed by their own facts. At this event, the aim is to bring the myths, technologies and politics of human-AI relationships closer together - from a position of mutual respect, yet with an awareness of each other's shortcomings.
There are reasonable but also superstitious fears for human obsolence in certain domains of society. There are justified and wildly erratic hopes for the precise observations that AIs can make. AIs can't be trusted: they amplify human biases and mistakes that are present in the data sets that are used for their training. AIs can be relied upon. They can be trained to be inhumanly good at precise cognitive tasks like winning a game of go, or recognising early signs of tumours on x-rays. AIs can be brilliant artists, but sadly cannot have have artistic experiences. AIs are alien minds that should be able to see in ways that are unhampered by human-centric categorisations. AIs are also unbelievably stupid. They can't tell the difference between a truck and a piece of sky, they see snow mobiles in snowy landscapes with only trees and rocks, and regularly mistake a cat for a human.
A promising early proposal for a kind of normalised human-AI relationship came from chess champion Garry Kasparov after he lost the game against the AI Deep Blue in 1997- the first time a machine intelligence publicly beat a human intelligence in a field considered a high mark of human capacity.
Following his defeat, Kasparov discovered that a human chess grandmaster supported by an AI always beats an AI or a human on their own, and he developed 'centaur chess'. This is always played between two teams, each consisting of a human player and an AI assistant.
Basically, the AI here works as a recommender system just like in Spotify, Netflix or Amazon. The centaur is a metaphor for a human-AI relationship, with the AI doing the legwork of working out all the possible moves and preselecting potentially useful ones, and the human as the brain making the ultimate decisions.
Could this be a model for functional human-AI relations? What would be needed for humans and AIs be able to trust each other in such relationships? And is the centaur the best metaphor for this?
Before the Thursday Night you can enjoy a bite to eat with the speakers and staff of Het Nieuwe Instituut. At 18:00 Het Nieuwe Café will serve a light vegetarian meal. Dinner vouchers are available for ¬ 7.70 up to a day before the event via the Tickets link.
Become a Member
Become a Member of Het Nieuwe Instituut and you'll support our mission to navigate the vast and evolving field of design. You'll also be inspired by our special program of Members' events, meeting up with other like-minded people as we invite you to reflect with us on design's changing role in technology, economics, culture and society.