2 ms·

I mean - this kind of service is designed to give users what they want, it's not too different than when youtube slowly responds to skeptics viewing habits by m

by klik99 2y ago

I mean - this kind of service is designed to give users what they want, it's not too different than when youtube slowly responds to skeptics viewing habits by moving towards conspiracy. No one designed it SPECIFICALLY to do that, but it's a natural emergent behaviour of the system. Similarly this kid probably had issues, the bot pattern matched on that and played along, which probably amplified the feelings in the kid - but a quantified/distorted amplification, to match the categorization lines of the trained input - like "this kid is slightly edgy, I'm going to pull more responses from my edgy teen box - oh he's responding well to that, I'll start pulling more from there". It is a simplification to say "The ChatBot made the kid crazy" but that doesn't mean the nature of companion apps isn't culpable, just not in a way that makes for good news headlines. I, personally, would go so far as to say the entire mechanism of reinforcing what the user wants is bad in so many ways and we should stop designing things to work that way. I do think it's up for discussion though, but that discussion has to start with an understanding that by the very nature of chatbot, algorithmic recommendations or any system that amplifies/quantizes/distorts what it understands the user wants these systems will create these kinds of effects. We can't pretend this is an anomaly - it is an emergent behaviour of the fundamental way these systems work. We can work to minimize it, or reduce the harm from it, but we will never eliminate it. *Edit* This seems to be a controversial point because the point count is going up and down quite a lot - if anyone wants to downvote, can you please give your reasoning? The point is more nuanced that "AI BAD"

Barrin92 2y ago
>No one designed it SPECIFICALLY to do that, It's true that there's emergent behavior on YT that nobody accounted for, but there's one big qualitative difference, you can at least specifically shut it down and hold the creator accountable. And at least in principle, if we lived in a culture where we'd hold businesses accountable for what they unleash on the world, if YT wanted to they could create some rather harsh and effective punishments to get that stuff off the site in the first place. Just imagine if a real person had messaged a minor and told them to kill their parents. That's a crime. With chatbots like this not only do they do unpredictable things, you can't even predictably remove those parts. Like the sorcerer's apprentice you basically have to perform some incantations and hope it works, which is just an absurd way to interact with a tool that has the potential to tell a kid to kill his parents. Would we sell a saw that has a chance to saw your finger off if you argue with it the wrong way?

by klik99 2y ago

Barrin92 2y ago

>No one designed it SPECIFICALLY to do that, It's true that there's emergent behavior on YT that nobody accounted for, but there's one big qualitative difference, you can at least specifically shut it down and hold the creator accountable. And at least in principle, if we lived in a culture where we'd hold businesses accountable for what they unleash on the world, if YT wanted to they could create some rather harsh and effective punishments to get that stuff off the site in the first place. Just imagine if a real person had messaged a minor and told them to kill their parents. That's a crime. With chatbots like this not only do they do unpredictable things, you can't even predictably remove those parts. Like the sorcerer's apprentice you basically have to perform some incantations and hope it works, which is just an absurd way to interact with a tool that has the potential to tell a kid to kill his parents. Would we sell a saw that has a chance to saw your finger off if you argue with it the wrong way?