Humans + AI: Do we want to be Centaurs or Minotaurs?


After chess grandmaster Garry Kasparov was first beaten by a computer he introduced what is often called “centaur chess“, in which humans and machines collaborate, historically often beating either humans or computers on their own.

A centaur is half-human, half-horse, but importantly the head and torso are human and the body is horse, giving agency to the human part of the combination.

In contrast, a minotaur is half-human, half-bull, but the bull is the head and the human is the body that enacts its intentions.

The metaphor of minotaurs for human-AI collaboration is now being used more frequently.

The Flux Collective newsletter asks about Large Language Models: “does a given tool increase our agency or reduce it?”

An agency-taking generative tool might initially come across as a helpful partner. It will do chores for you, allowing you to ignore the minutiae. But in doing so, it diminishes your agency. Decisions are made for you — you no longer have the agency to make them. Given our limited time, some abdication of agency can be a good thing. But, at the extreme, agency-taking tools are minotaurs (creatures with a bull’s head and a human’s body) that run people’s lives. 

An agency-giving generative AI tool is more like a centaur: a creature with a human’s head and a horse’s body. It doesn’t do stuff for you. It enables you to do it better, more effectively, and with more power. Agency-giving tools augment and amplify our capabilities, rather than replacing them. 

Source: Flux Review Ep 101

A recent article in a US military journal, Minotaurs, Not Centaurs: The Future of Manned-Unmanned Teaming, says:

A centaur is a mythical creature with the head and upper body of a man and the lower body of a horse. When used to describe manned-unmanned teams, the image of the centaur promotes the idea that human beings will lead the team. We outline an alternative vision of the nature of manned-unmanned teams, which is more likely to be realized in key domains of warfighting in future wars. Rather than human beings directing multiple robots, we suspect artificial intelligences (AI) will direct the activities of multiple human beings. The cyborg soldier of the future is more likely to be a minotaur—a mythical creature with the body of a man and the head of a bull—than a centaur: they will have a monstrous head rather than a monstrous body.

Neither of these frames are enticing. We clearly want greater agency, with AI doing our bidding rather than the reverse, and we don’t want soldiers who are commanded by monstrous, unfeeling AI.

Yet it is worth recalling what Thomas Malone, Director of the MIT Center for Collective Intelligence, proposes in his book Superminds: AI can act relative to humans as a tool, an assistant, a peer… or as a manager. If machines can allocate tasks to humans so they can do what they’re best at and create a more effective system, that makes sense.

This also recalls Vitalik Buterin’s original framing of Distributed Autonomous Organizations (DAOs) as, ideally, automation at the center and humans at the edges. If the systems and protocols can allocate human tasks well, this can generate better synergies than humans attempting to allocate resources. 

Source: Ethereum Foundation

The first step is to recognize the metaphors of centaurs and minotaurs and their implications. 

There will be many increasingly complex configurations of humans and AI coming, we need to familiarize ourselves with the basic archetypes.