AI Safety Gym - Computerphile

Rob Miles discusses the idea of a gym for training AI algorithms.


This video was filmed and edited by Sean Riley.

[AcornElectron]

Awesome. I’ve watched literally everything Rob has recorded on AI. He’s very relatable, knowledgeable and informative.

[DrumsKylePlays]

Miles is an excellent teacher. Always does a great job fielding questions from a layperson.

[RasperHelpdesk]

"A ship in harbor is safe — but that is not what ships are built for."

[danielm]

I used to know this guy. Glad he’s still at it. Easily one of the smartest dudes I’ve met in person

[LeoStaley]

Rob's video on the 3 laws of robotics is what really demonstrated to me how serious ai safety really is.

[ragnkja]

The sponsor intro is too loud.
Edit: as is the sponsor segment at the end.

[hattrickster]

Looks like they're taking security very seriously. This guy is always kept inside a prison to avoid his rogue AI pets from escaping.

[DIECARS]

never knew notts uni had a prison to film in

[joshie]

I'm a man of simple tastes - I see Rob Miles, I press the like button.

[TheStarBlack]

7:58 that artificial camera movement is both trippy and impressive!

[esquilax]

I like the fact that young Robert uses the same Simpsons references I remember from 20-odd years ago

[letsgobrandon]

I really love listening to Rob's explanations.

[locarno]

Completely agree. Big safety failures - in organisation structure, or real world industry, or whatever - usually occur because of either unknown elements in the environment or unexpected interaction by known elements. Because - at stupidly obvious level - if you could predict it you would (you'd hope) have done something about it.

Thanks for the the description on the constraint learning. Keeping constraints and goals kept as modular elements is one of those things that makes obvious sense *once* someone explains it to me.

[-ek]

That is an absolutely miserable classroom!

[doodlebobascending]

I initially read this as "AI Sentry Gun" and thought Rob was having a crisis.

[moon_bandage]

He never ended up explaining what this "gym" thing is :(

[abcdemnopq]

Fab and super interesting video, also v. much appreciated your [Rob's] EA talk yesterday - will definitely be checking out the AI Safety field in more depth.

[elephantwalkersmith]

Non linear optimization methods like sqp often include constraints. This is very common in fields other than machine learning. The problem with constraints is their formulation is actually very difficult, and infeasible path optimization is necessary to solve the learning problem.

[ashurean]

5:19 Would it be possible to mix VR and test simulations to have real humans interact with the simulated machine? Just have it open to the public and you have all the "real human reactions" you'll ever need.

[SirWilliamKidney]

+ 100 points for the THHGTTG reference!