Thread by @Rachel_Gutin

The seventh panel I attended at #Readercon was "Compassionate AI," with @WriteTeachPlay, @mattkressel, @eilatan, and Kestrell Verlager, moderated by @tithenai. This panel was at 10 AM on Saturday and worth waking up for. It was also on Shabbat, which has note-taking implications.

I'm a Jew who keeps Shabbat in a way that precludes writing. This means I had to hold all of the key points in my mind until after Shabbat, when I could write that down. For more info on how I do that, check out this thread:

https://twitter.com/Rachel_Gutin/status/1151650216841568256?s=20

The short, short version: I attach the notes to an image in my mind. This means I remember key points, but not the order they were said in, definitely not direct quotes, and usually not who said what either. With that in mind, here are my take-aways from this panel:

Something to consider: When we say "compassionate AI," compassionate to whom? (Our biases get included in programming.)

One panelist brought up values alignment: we want the AI's values to match our own.

The idea of compassionate AI is not a new one in SF. It's been around for a while.

Our current expectation is that AI will communicate with us like humans, in a natural-feeling interaction. This is what programs like Siri and Alexa aim for.

AI arrive at their intelligence/awareness via a different path than humans. For example, they might examine every possible path before figuring out what works.

A lot of times, how an AI or bot reaches its current response set is unknown. It's a black box, and that's disturbing.

Most programming is done in English, which comes with limitations, biases, and barriers to access.

AI can restrict human paths too, how humans interact with the world. For example, when you book a plane ticket, you have to choose a gender: male or female. There aren't any other options offered.

The panelists discussed the idea of "raising" an AI like we raise kids. Different people will have different parenting styles.

When raising AIs, we're raising very rich kids, because it's expensive.

In addition to viewing AIs as children to be raised, we can view them as our teachers, or as a godlike power watching over us.

Bonus content: On Saturday evening, I was chatting with @WriteTeachPlay, and he noted something that didn't get said during the panel: There are two possible funding sources for AI development: wealthy people/corporations, and universities. We should encourage the latter.