“Do Artificial Reinforcement-Learning Agents Matter Morally?” Yes, says Brian Tomasik, even present-day ones (by a very small but nonzero amount). He foresees their ethical significance increasing in the near future, and he isn’t talking about strong AI, but an increase in the ordinary applications of reinforcement learning to our technology.
The argument is, briefly: for various claims about what consciousness physically is, RL programs display these features to some extent as well. Therefore they have a nonzero degree of consciousness, and so a nonzero degree of moral standing. Enough that we should be thinking now about guidelines for the ethical creation of such software.
He suggests that, paralleling guidelines for the use of animals in research, RL algorithms should be replaced by others whenever possible, or if they must be used, reduced in number, and driven through rewards, not punishments.
He considers the idea of an organisation of People for the Ethical Treatment of Reinforcement Learners, and the embedding of RL algorithms in humanoid bodies and videogame characters as ways of persuading the public to the idea that they have moral significance.
“Do Artificial Reinforcement-Learning Agents Matter Morally?” Yes, says Brian Tomasik, even present-day ones (by a very small but nonzero amount). He foresees their ethical significance increasing in the near future, and he isn’t talking about strong AI, but an increase in the ordinary applications of reinforcement learning to our technology.
The argument is, briefly: for various claims about what consciousness physically is, RL programs display these features to some extent as well. Therefore they have a nonzero degree of consciousness, and so a nonzero degree of moral standing. Enough that we should be thinking now about guidelines for the ethical creation of such software.
He suggests that, paralleling guidelines for the use of animals in research, RL algorithms should be replaced by others whenever possible, or if they must be used, reduced in number, and driven through rewards, not punishments.
He considers the idea of an organisation of People for the Ethical Treatment of Reinforcement Learners, and the embedding of RL algorithms in humanoid bodies and videogame characters as ways of persuading the public to the idea that they have moral significance.
I would be much more morally concerned about reinforcement learning agents if this were a functional distinction.
He discusses that point in the paper.