I highly recommend this. It features all of the worlds foremost experts on AI. Pretty cool, but also very scary stuff!
Outstanding Documentary on Artificial Intelligence
- Thread starter Scalphunter
- Start date
That was worth watching, but I felt they spent the last half conflating topics.
Manipulating the suggestibility of people by carefully targeting what you present them is a scary/powerful tool. But that stuff is driven by humans, not really AI in my view.
And the part about Microsoft's chat bot turning into a troll made it seem like the thing turned hateful just being exposed to the public at large, but I'd guarantee that was the consequence of Anons targeting it to shape it. Still spooky, but the threat vector isn't really the bot, it's folks behind it manipulating it.
The scariest stuff is what Musk was talking about, where the bot has control, and a 'mission', and then executes that mission irrespective of things we'd consider important, but forgot to program into the bot.
It's still really hard to see the point where the bot 'learns' enough on its own to program its own mission and limits (or ethics, or whatever you want to call them). But I guess that'd be when we're the ants.
Manipulating the suggestibility of people by carefully targeting what you present them is a scary/powerful tool. But that stuff is driven by humans, not really AI in my view.
And the part about Microsoft's chat bot turning into a troll made it seem like the thing turned hateful just being exposed to the public at large, but I'd guarantee that was the consequence of Anons targeting it to shape it. Still spooky, but the threat vector isn't really the bot, it's folks behind it manipulating it.
The scariest stuff is what Musk was talking about, where the bot has control, and a 'mission', and then executes that mission irrespective of things we'd consider important, but forgot to program into the bot.
It's still really hard to see the point where the bot 'learns' enough on its own to program its own mission and limits (or ethics, or whatever you want to call them). But I guess that'd be when we're the ants.
Heard a great podcast a few weeks ago about something similar to this. Where due to a blindspot of the AI developers the AI doesn't account for certain factors -- in the case of the podcast it was an AI that made criminal justice decisions -- as such it was treating certain groups more severely based on race. Once discovered they stopped using the AI but when used at scale an AI is tougher to interrogate or unwind than a person making decisions themselves. Basically AI won't be some perfect, unquestionable tool -- it will carry some of the flaws or blindspots of the people engineering it.
Brave new world but it's happening. Maybe one day Jimbo can use an AI for his playcalling and will reduce his delay of game average into the single digits.
