How Do Machines Learn?

I rewatch your videos all the time and this is one of my favorites. I usually watch your videos while I do 8+ hour puzzles so I’ve seen them a few (viz. very many) times each and they never get boring. Thank you for your content!

Be Owusu

Final cut pro, he talks about it on the podcast

Liam Brennan

smart

Ariana

ASMR Art

Keep it up ;) Merry Christmas to all ^~^

ZULEYKA GAMES

resigned!grey is the best Grey. I'm off to tweet/share/like/etc!

Catherine Germann

@godOf I thought as much. @kimberly That's a good point. @Myrddin Sounds like an interesting book. Proving consciousness in an algorithm would be very tough indeed.

Kairon R Woulfgang

Its a program but a neural net, which takes sums and products of connected elements that have variable values 0-1 to set values for other nodes. The machine is made of lots of elements that are indistinguishable and inscrutable individually.

we have yet to make any sort of progress toward true AI, don't believe the hype.

{ TRIGGERED }

Unanimous D

Well done. You stayed well clear from the words like 'simulate' and 'emulate'.

Ammar Mousali

You could technically see it working, in the literal, but you couldn't make sense of it's method.

In theory, yes. You could pause execution of the program and take a look under the hood. However, the chances of making any sense of what's under the hood is slim to none. It might be possible to understand the code that's being run, but not the gigantic set of data which actually powers it.

Kaleb Elwert

here is what I don't understand: it's a computer, running a program. can't you, at any point, pause the execution of the program, and stack trace your way through the data it's consuming and the algorithms it's running that through? Like, are you just saying that in kind of a colloquial sense the bot's function is unknowable, even though in reality it's entirely knowable and can be observed at any time? I mean, I get that a lay person might not be able to parse what they're seeing if you split open a bot and look at what makes it tick, but - that doesn't mean that the information isn't there, or that no one could possibly understand it, right?

matt

Yeah but that big blank whiteboard would be a great place for stashing short cuts :-D

z0mb1e564

I almost feel bad for YouTube. .....almost.

Unanimous D

OMG I want a 'ELI0' onesie (you know, for a baby under 1 year old)

Jon Levell

I'm more a fan of the frame at 8:20... :)

Jennifer Rittmann

Yep! (I couldn't remember who made the racist bot). Or when IBM's Watson had to be reset because it read Urban Dictionary. The big danger is that people like to say "It's a computer so it can't be biased" and then use that to justify racist policy.

Matthew Davidson

I don't suppose you'd make the frame at 8:50 into a wallpaper :-D

z0mb1e564

Like the instance where they did face tracking software and only trained using white people so in the end it couldn't track black people. Or the time Microsoft released a Twitter AI and it became racist within 24 hours.

Lu Peters

Hi Bots! This vid was great I watched the whole thing and even giggled at the end part. Well done creator CGP Grey!

Grant

Great video! Could you do a followup talking about how ML is only as good as the data it gets and the dangers of assuming that algorithmic results are unbiased (e.g. feeding racist crime data results in racist policing suggestions) and how it can end up going off in a direction you didn't intend (like that chat bot that became a racist). Thanks!

Matthew Davidson

How can you not smile at that music?

Tom

I would argue that Artificial Intelligence already has been created. What is debatable now is the •level• of intelligence. Most people, when they talk of 'AI' really mean 'AS' ... artificial sentience.

Kimberly Green

It depends on what you define as 'AI'. Unfortunately many smart AI scientists still don't really agree on exactly the definition of AI. I'm a big fan of a very good (and surprisingly, still relevant) book called Godel, Escher, Bach: An Eternal Golden Braid. It won a pulitzer, and it might be the funniest and weirdest book to discuss AI ever written, especially since it doesn't even tell you that it's talking about AI and the nature of consciousness until 3/4 of the way through the book! It spends the first 3/4 of the book teaching you the concepts you need to understand in order for the last 1/4 to make sense. Douglas Hofstadter makes the assertion that consciousness arises from symbolic self-representation, and further asserts that, due to Godel's incompleteness theorum, it's literally impossible to symbolically prove consciousness. It's a nifty idea, and it still holds up today, in my opinion.

Myrddin Emrys

I know it's unlikely and still fictional but is it possible to accidentally create AI using this method?

Kairon R Woulfgang

What's the background music?

RTT12

As always a very good video

Jonathan Morton