Rebuttal to Reviews for 2005 AAAI Symposium on Machine Ethics

My paper The Ethics and Politics of Super-Intelligent Machines was rejected by 2005 AAAI Fall Symposium on Machine Ethics based on a poor reviewing process (however, the organizers did invite me to present a poster and short abstract). Here are:

The version of the paper as submitted to the symposium.
The first review and my rebuttal (the organizers of this symposium on ethics have made a mistake by choosing a reviewer who does not understand Hobbes' social contract).
The second review (select View/Comments in Word to see the reviewer's comments) and my rebuttal.


---------------------------------------

The Technical Report from the 2005 AAAI Fall Symposium on Machine Ethics reveals some serious omissions in content. All papers that discuss implementation of machine ethics do so in terms of logic, except for Marcello Guarini's paper which describes supervised learning of ethics via artificial neural networks.

But reinforcement learning is the essence of intelligence and a symposium on machine ethics should at least consider implementing ethics at the level of reinforcement values. Just as symbols must be grounded in sensory information, symbolic logic must be grounded in learning. In a truly intelligent mind, able to adapt to general situations, logic develops from learning as a way to solve the credit assignment problem. That is, logic and the symbols manipulated by logic are part of a simulation model of the world that the mind learns and uses to trace cause and effect relations between behaviors and the rewards that reinforce those behaviors. Ethics expressed directly in terms of logic are liable to be overridden by the simulation model that an intelligent mind must learn through experience.

A second deficiency of the symposium is the lack of serious discussion of the practical ethical challenges that intelligent machines will pose for human society, such as the prospect for nearly 100% unemployment when machines are able to do all jobs better and more cheaply than people. Selmer Bringsjord's paper discusses these issues as raised in Bill Joy's 2000 Wired article, Why the Future Doesn't Need Us, but only to rebut Joy's argument (Bringsjord is very skeptical about the possibility of intelligent machines). I know of at least two papers, my own and another by a well-known writer on AI ethics, that were rejected for this symposium and that did consider these practical ethical issues.