RE: AI is not dangerous because of the reason most people think

You are viewing a single comment's thread from:

AI is not dangerous because of the reason most people think

in technology •  7 years ago 

We are already having this problem over at Google. The algorithm that recommends you Youtube videos based on all available data on previous preferences is in the form of a neural network which has performed so many internal modifications that the programmers and engineers there don't know how it runs the algorithm anymore. They are still fully capable of controlling it, but learning the particulars of the algorithm exacts such a high price in terms of manpower dedicated to it that even Google doesn't have enough brains to fully vet it.

Now, I wonder if that'll happen to DTube....

Authors get paid when people like you upvote their post.
If you enjoyed what you read here, create your account today and start earning FREE STEEM!
Sort Order:  

Yep, that's a great point and a great example. It's not about having an algorithm that itself is overly complex, it's just that it changes so fast and makes so many decisions that at some point it becomes practically impossible to keep track of. And things are just getting started.

I wonder if self-driving cars would at some point realize that drifting allows them to corner at higher speeds and at some point decide to start fast-and-furious-ing passengers that are in a hurry because they are learning driving from GTA (which some are) and because one way to teach algorithms about stuff is to make them watch YouTube videos :P

I doubt DTube would ever evolve into this as its goal is to be open and there is nothing less open then an incomprehensible black-box algorithms.