Content-type: text/html Downes.ca ~ Stephen's Web ~ Cheap Talk, Reinforcement Learning and the Emergence of Cooperation

Stephen Downes

Knowledge, Learning, Community

One way of supporting cooperation is to increase the cost of communication, making dishonesty more risky. This paper discusses an alternative whereby we can decrease the cost of communication by making dishonesty less profitable. That, at least, is my take on what I'm reading here. There's a lot of jargon in this paper (14 page PDF) from the world of games and decision theory. Essentially, communication becomes less risky. In more structured systems, "signals must be costly in order to ensure that they are reliable, or honest, for otherwise such signals could be easily forged by opportunistic agents." By contrast, 'cheap talk', combined with a 'discounting function', allows agents to get past an early period of betrayal and mistrust to emerge into a cooperative association. This is important because it allows us an alternative to 'expensive talk' systems like blockchain that are specifically designed to prevent dishonesty, and it suggests that open cooperative systems are less risky than supposed.

Today: 0 Total: 1399 [Direct link] [Share]


Stephen Downes Stephen Downes, Casselman, Canada
stephen@downes.ca

Copyright 2024
Last Updated: Apr 29, 2024 07:55 a.m.

Canadian Flag Creative Commons License.

Force:yes