Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

How Robots Can Acquire New Skills from Their Shared Experience

research.googleblog.com

Post reply on HN
Print view View original

31–40 of 48 posts

Re: How Robots Can Acquire New Skills from Their Shared Experience

#31

Post by eli_gottlieb »

Retric wrote: post #29
Udo wrote: post #11

Earlier quoted context omitted.

The problem is at today's state of AI and robotics, we're giving a couple of exceedingly high-powered tools (both physical and virtual) to entities who possess less general intelligence than a cockroach. This is what gives AI doomsayers ammunition, because it enables them to extrapolate towards a future where AIs incapable of ethics or deeper insights are put into positions where they can execute catastrophic global…

I think the fear from AI is driven in part by the realization that their is nothing unique about humanity. Being subjected to how humans treat apes, or even other humans is scary. Intelligence makes things more dangerous without necessarily making them kinder.

>Intelligence makes things more dangerous without necessarily making them kinder.

Then you'd think someone would bother to study how kindness works.

Re: How Robots Can Acquire New Skills from Their Shared Experience

#32

Post by squeaky-clean »

slowmovintarget wrote: post #13
bollu wrote: post #8

Earlier quoted context omitted.

Where is this from? Seems like an intriguing read

Me... about an hour ago. :) Maybe this ought to be my Nanowrimo entry this year.

I Googled the quotes looking for a source to this too (of course before scrolling slightly and seeing this comment). I wish I could read more!

Re: How Robots Can Acquire New Skills from Their Shared Experience

#33

Post by saint_fiasco »

eli_gottlieb wrote: post #31
Retric wrote: post #29

Earlier quoted context omitted.

I think the fear from AI is driven in part by the realization that their is nothing unique about humanity. Being subjected to how humans treat apes, or even other humans is scary. Intelligence makes things more dangerous without necessarily making them kinder.

>Intelligence makes things more dangerous without necessarily making them kinder. Then you'd think someone would bother to study how kindness works.

The people who are worried about AI risk do study that stuff. In the context of AI research, the property of having an overall good effect on humanity is called "friendliness".

https://en.wikipedia.org/wiki/Friendly_artificial_intelligen...

Re: How Robots Can Acquire New Skills from Their Shared Experience

#34

Post by yummyfajitas »

Udo wrote: post #11
slowmovintarget wrote: post #6

"We had him pinned down and we didn't dare let the little guy out of that Farraday cage. We had to chase him down for two hours until we got off a lucky shot and gutted his battery." "Why did you have to isolate him?" "He'd begun 'solving' his challenge tasks by killing the kittens instead of feeding them. Him sending a solution pulse to the rest of his generation would've ruined the lot." "You're just going to delet…

The problem is at today's state of AI and robotics, we're giving a couple of exceedingly high-powered tools (both physical and virtual) to entities who possess less general intelligence than a cockroach. This is what gives AI doomsayers ammunition, because it enables them to extrapolate towards a future where AIs incapable of ethics or deeper insights are put into positions where they can execute catastrophic global…

The intelligent doomsayers say something fundamentally different.

They say that the space of possible objective functions (read: goals) for AI is huge, and the space of methods for achieving those goals is also huge. In contrast, the space of goals/methods that humans find acceptable is much smaller.

I.e., 99.99% of AI designs will be dystopian.

I'm not sure why you consider minds hardwired to enjoy serving us to be unethical. Is it also unethical for a human to enjoy serving others (some do)?

Re: How Robots Can Acquire New Skills from Their Shared Experience

#35

Post by infinite8s »

dharma1 wrote: post #24

Makes sense to train in parallel and sync the trained models frequently. Still think a lot of the training could be done much faster in a physically accurate simulated environment prior to real world training. Or is real world physics too different from simulations?

I think the difficulty in using virtual environments for training purposes isn't in simulating the environment, but accurately simulating the physical responses/limitations of the robotic hardware in that environment in a way that would reflect real engineered hardware (ie motor responses, signal latency, etc).

Re: How Robots Can Acquire New Skills from Their Shared Experience

#37

Post by Udo »

yummyfajitas wrote: post #34
Udo wrote: post #11

Earlier quoted context omitted.

The problem is at today's state of AI and robotics, we're giving a couple of exceedingly high-powered tools (both physical and virtual) to entities who possess less general intelligence than a cockroach. This is what gives AI doomsayers ammunition, because it enables them to extrapolate towards a future where AIs incapable of ethics or deeper insights are put into positions where they can execute catastrophic global…

The intelligent doomsayers say something fundamentally different. They say that the space of possible objective functions (read: goals) for AI is huge, and the space of methods for achieving those goals is also huge. In contrast, the space of goals/methods that humans find acceptable is much smaller. I.e., 99.99% of AI designs will be dystopian. I'm not sure why you consider minds hardwired to enjoy serving us to be…

> I.e., 99.99% of AI designs will be dystopian.

Not sure if I consider this a valid premise, much less an intelligent one, but luckily I'm not the arbiter of such things. If this becomes a "number of possible goals times the means to carry them out" argument, you could easily construct the same one against human progress as a whole.

> I'm not sure why you consider minds hardwired to enjoy serving us to be unethical.

Holy shit, yes it's unethical.

If I asked you "would it be unethical to genetically engineer a new race of humans who enjoy being slaves to the upper class?" I hope your answer would be yes. The same rules apply to a non-human species as well.

Re: How Robots Can Acquire New Skills from Their Shared Experience

#38

Post by yummyfajitas »

Udo wrote: post #37
yummyfajitas wrote: post #34

Earlier quoted context omitted.

The intelligent doomsayers say something fundamentally different. They say that the space of possible objective functions (read: goals) for AI is huge, and the space of methods for achieving those goals is also huge. In contrast, the space of goals/methods that humans find acceptable is much smaller. I.e., 99.99% of AI designs will be dystopian. I'm not sure why you consider minds hardwired to enjoy serving us to be…

> I.e., 99.99% of AI designs will be dystopian. Not sure if I consider this a valid premise, much less an intelligent one, but luckily I'm not the arbiter of such things. If this becomes a "number of possible goals times the means to carry them out" argument, you could easily construct the same one against human progress as a whole. > I'm not sure why you consider minds hardwired to enjoy serving us to be unethical.…

Here's a way to recognize that it's almost certainly a true premise. Out of all the animal species on earth, which ones would (if they took over) create a world that is as pleasant for humans to live in as the current one?

If I asked you "would it be unethical to genetically engineer a new race of humans who enjoy being slaves to the upper class?" I hope your answer would be yes. The same rules apply to a non-human species as well.

We've already done this to several non-human species. Do you consider owning dogs unethical?

I don't think it's such a simple question.

Re: How Robots Can Acquire New Skills from Their Shared Experience

#39

Post by uremog »

slowmovintarget wrote: post #30
Udo wrote: post #11

Earlier quoted context omitted.

The problem is at today's state of AI and robotics, we're giving a couple of exceedingly high-powered tools (both physical and virtual) to entities who possess less general intelligence than a cockroach. This is what gives AI doomsayers ammunition, because it enables them to extrapolate towards a future where AIs incapable of ethics or deeper insights are put into positions where they can execute catastrophic global…

What if one of the parties in that dialogue was an AI?

And the subject is a human

Re: How Robots Can Acquire New Skills from Their Shared Experience

#40

Post by jonnycowboy »

Very interesting, thanks for the share.

One point I noticed not mentioned yet is that the images used for training are only 64x64! In the original google "grasping" research, the images were 472 × 472, 54 times bigger! I think they are looking for "minimum visual information" required to trigger the required learning. This will help in mobile applications (ie: robotics, smartphones, etc) where processing power is severely limited.

Post reply on HN