Desert Island Discs – Demis Hassabis, CEO of DeepMind [audio] • Hacker News Board

Re: Desert Island Discs – Demis Hassabis, CEO of DeepMind [audio]

#41

Post by a_lieb » Sat, Jun 10, 2017, 7:18 PM UTC

MrQuincle wrote: post #9

nopinsight wrote: post #3

I enjoyed listening to the short segment on how and why he gave up chess at 11 despite being ranked 2nd in the world for that age group. Remarkable reflectiveness for an 11-year-old. Check out this down the page: "Demis Hassabis: ‘I thought we were wasting our minds’ AI expert Demis Hassabis on giving up chess tournaments aged 11." It makes me wonder how many other brilliant minds are wasted on trivial pursuits and h…

I'm not that brilliant, but I gave up on chess myself at around 18 when I crossed the threshold of playing blind and not being able to fall asleep at night because I was replaying games in my head. I think there are plenty of indicators that a more brilliant mind would take as clues of navigating to some local intellectual optimum. :-) Only now I'm twice that age I follow a bit of the tournaments again. Watching a ba…

Is there anything you can recall doing that would encourage or discourage chess stuff from popping into your head like that? Do you think it came out of intentionally replaying games on your head and then became more automatic? Or did it just randomly start to happen on its own?

This kind of automatic mental practice seems like an awesome skill to develop for any field of study. It's a way to practice and learn without using up any willpower.

I play guitar and sometimes I do find myself doing solfege exercises and visualizing fretboard patterns when I'm falling asleep/sitting in traffic/whatever. I'd love to be able to do that more.

Re: Desert Island Discs – Demis Hassabis, CEO of DeepMind [audio]

#42

Post by imranq » Mon, Jun 12, 2017, 2:43 PM UTC

nopinsight wrote: post #8

imranq wrote: post #5

Earlier quoted context omitted.

I mean you can argue that as nice as AlphaGo is, it has not accomplished any thing real outside of being better at some games.

For a start: DeepMind AI Reduces Google Data Centre Cooling Bill by 40% https://deepmind.com/blog/deepmind-ai-reduces-google-data-ce...

But how much of that saving is because AlphaGo is so great vs a baseline machine learning algorithm?

Re: Desert Island Discs – Demis Hassabis, CEO of DeepMind [audio]

#43

Post by imranq » Mon, Jun 12, 2017, 5:45 PM UTC

coldtea wrote: post #32

atmosx wrote: post #29

Earlier quoted context omitted.

So, all of you AI guys, started from chess? I see a pattern emerging...

You got that ("all of you guys", "pattern emerging") from just reading about 2 guys?

The cover for the seminal book on AI (norvig) has a chessboard and games are a fundamental training ground for AI so I'd say there's some truth there.

Re: Desert Island Discs – Demis Hassabis, CEO of DeepMind [audio]

#44

Post by Radim » Tue, Jun 13, 2017, 8:31 AM UTC

taneq wrote: post #33

Radim wrote: post #22

Earlier quoted context omitted.

Actually, you're not far off! Where else would the next major advance come from, if not from raising our collective consciousness through games (cheap focused simulations) and articulate story-telling (social alignment and morality correction)? We're not robots, to somehow perceive objective reality independently and logically. I certainly believe story-tellers and players are absolutely critical if humanity is to pu…

> Where else would the next major advance come from? Probably from sustained and detailed study of the universe around us and the way it works. That's how we figure out most things. Not that (board) games can't be useful as training but IMO they're a form of navel-gazing. (Games based on high fidelity simulations are definitely useful - look at the AI research using GTA5 as a simulator for self-driving cars, for inst…

Again, agreed for the most part. I just think you're underestimating the power of (board or otherwise) games, how they capture and reflect patterns of reality that are not immediately apparent or not directly useful (or even verbalizable).

There's this deep connection between curiosity, idle fun and problem solving.

The idea of driving people to "sustained and detailed scientific studies" to tackle "existential threats"... that is just not how the world works. Innovation doesn't happen by committee. It happens by making weird connections in unexpected places.

To be honest, I don't think you're giving enough credit to the evolutionary process that gave us our propensity to admire "games and stories". Dismissing it as "just navel-gazing", or just philosophy, in favour of rigorous science, sounds extremely perilous -- as if killing something fundamental to humanity.