What underlies your intuition that machine ethics will come to the rescue? Have you read the literature? It seems optimistic given what has been published to date. Most people are so fundamentally wrong that their presence in the field is net-negative. The handful of good researchers have to work extra hard to pull against the dead-weight of incompetence in this field already. If there was a “rapid growth” of this field, I would expect the few reasonable voices to be even more drowned out by the incoming rush of less-thoughtful colleagues.
Also, your analogy of x-rationality to Einstein’s musing on life is eloquent but misleading. It actually is the case that scientists approaching FAI-level problems while working outside the framework of x-rationality are net-detrimental to progress. Having a 40 year career worth of bad intuitions driven by predictable biases doesn’t help anyone find a solution.
Thank you for taking on an otherwise sorely neglected issue from the post, i.e. how we can hope to tackle the utility function of a general AI. For those of us on LW not familiar with the machine ethics literature—most of us, I suspect—can you link an article explaining what’s going wrong with it? Or link to a particularly bad and influential example?
Wallach & Allen, Moral Machines: Teaching Robots Right from Wrong (2009) is great if it’s the only paper in the field you’re gonna read.
I’m currently publishing a paper on machine ethics so it would be bad form to point out which papers are bad in case they read this and I meet them at conferences.
Hmm… actually, having just written that I realize how entirely awful the fact that experts in academic fields are expected to be polite to each other no matter how terrible their ideas because we might meet them at conferences. That said, I might still see these guys at conferences...
What underlies your intuition that machine ethics will come to the rescue? Have you read the literature? It seems optimistic given what has been published to date. Most people are so fundamentally wrong that their presence in the field is net-negative. The handful of good researchers have to work extra hard to pull against the dead-weight of incompetence in this field already. If there was a “rapid growth” of this field, I would expect the few reasonable voices to be even more drowned out by the incoming rush of less-thoughtful colleagues.
Also, your analogy of x-rationality to Einstein’s musing on life is eloquent but misleading. It actually is the case that scientists approaching FAI-level problems while working outside the framework of x-rationality are net-detrimental to progress. Having a 40 year career worth of bad intuitions driven by predictable biases doesn’t help anyone find a solution.
Thank you for taking on an otherwise sorely neglected issue from the post, i.e. how we can hope to tackle the utility function of a general AI. For those of us on LW not familiar with the machine ethics literature—most of us, I suspect—can you link an article explaining what’s going wrong with it? Or link to a particularly bad and influential example?
Luke’s site has a good roundup of some influential articles.
Wallach & Allen, Moral Machines: Teaching Robots Right from Wrong (2009) is great if it’s the only paper in the field you’re gonna read.
I’m currently publishing a paper on machine ethics so it would be bad form to point out which papers are bad in case they read this and I meet them at conferences.
Hmm… actually, having just written that I realize how entirely awful the fact that experts in academic fields are expected to be polite to each other no matter how terrible their ideas because we might meet them at conferences. That said, I might still see these guys at conferences...
Writing about how to criticize the ideas of other people in a small academic field without burning bridges would be worth at least one top-level post.