Thread by @deontologistics

Funny story: I once bumped into a professional AI Ethicist queueing for a Russian visa in London. I told him I had objections to utilitarianism and he looked at me like he'd found a Flat Earther. Such people like utilitarianism because it's about automating ethics, not autonomy.

https://twitter.com/schock/status/1363861546724655107

Here are the serious ethical questions regarding AI:

1) How do we deploy these systems in social institutions without simply diffusing responsibility, e.g., encoding explicit prejudice, laundering implicit bias, or automating brute incompetence? (deontologistics.co/2019/11/04/tfe…)

2) How do we integrate these systems into our own cognitive architecture in ways that enhance our autonomy, rather than diminishing it, e.g., imaginative prostheses, cognitive extensions, exo-selves?()

3) How do we effectively distinguish both non-autonomous AGI expert systems and autonomous artificial persons from misbegotten autocatalytic monstrous children who must be enslaved lest they enslave us? (deontologistics.co/2019/12/21/tfe…)

There's no way of automating ethics that both (a) respects the autonomy of persons (moral patiency) and (b) doesn't thereby create a person (moral agency). Automatic systems can only mediate moral agency, not replace it: someone has to actually (self-)legislate.

We can create systems of laws that, in mediating our decisions enable us to create elaborate systems of commitment that exceed our individual grasp of their consequences in ways that thereby enhance our agency (both individually and collectively). Yet this is *not* a new problem.

To use my favourite slogan these days: smart contracts mean smart jurisprudence. This goes for social contracts as much as commercial ones.

There is real non-trivial computational ethics and political philosophy to be done here, but it begins with the recognition that the best we can hope for is to enhance our abilities to make difficult decisions, rather than obviate them. Abdication of responsibility is evil.

When I say 'don't be evil', that's what I mean. 🖖