Darius Kazemi

Oren Etzioni from Allen Institute for AI talking about negative public discourse around AI. Taking a few (deserved) potshots at Nick Bostrom

I find AI optimists to be about as frustrating as AI pessimists. My position: AI is weird! Really fucking weird!!

Hmm, Etzioni seems to assuming that the ultimate goal of AI is to emulate the human mind, which is soooo uninteresting to me

Now Etzioni is talking about a moral imperative to deploy AI tech in medicine, transport, etc bc "without it, people are dying".

But also, the more we automate systems, the more attack surface we expose. You make shit smart, you also make it hackable.

Would asking questions of the CEO of the Allen Institute for Artificial Intelligence constitute "punching down"? asking for a friend

Now: the panel on designing accountability and fairness into AI, interpreting mechanisms and outputs, biases and discrimination

We jump straight to Skynet when talk AI ethics. It sucks all the oxygen out of the room. We need more granular discussions. -@katecrawford

Transparency in algorithmic determinations doesn't necessarily show you what the biases of an algorithm are over time -@katecrawford

Judges can't ask Qs of expert systems like an expert witness. they'll assume system output is science instead of expert opinion -@jackbalkin

If you're going to build AI systems to assist in legal cases, you need to account for the culture of the legal system -@jackbalkin

If AI systems are going to assist in a judicial setting, you will need trained computer scientists on hand (court employees?) -@jackbalkin

Now @jackbalkin compares the shift to AI expert systems as similar to culture shift in creation of the administrative state in early 20th C.

The administrative state agents now need to be part policy experts, part programmers! -@jackbalkin

And @katecrawford says we need to start teaching "data ethics" -- I'm wondering how effective ethics courses in bio curricula have been?

Need to deliver AI systems along with new professional norms. They go hand in hand. It's what's called a reprofessionalization -@jackbalkin

OMG someone in Q&A says learning algorithms can't have racial bias just because the outputs are different for different races ok bro

Ah good, the moderator chastised him for not asking a question and instead opining for a solid three minutes

I am glad I can wince at @thricedotted who is sitting next to me

Technically speaking, computers aren't capable of racial bias. Technically, I am just a bag of neurons that fire in response to stimulus.

Heyyy, @gleemie shoutout at this panel for her labor analyses of digital menial laborers