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

Machine intelligence, part 1

blog.samaltman.com

Post reply on HN
Print view View original

51–60 of 188 posts

Re: Machine intelligence, part 1

#51

Post by dicroce »

I've been studying machine learning lately... Here is my take:

Well before we create an ASI (artificial super intelligence) we will have put 90% of the human race out of work with specialized (non conscious) intelligent agents... (for example, self driving cars). I believe that this will be a disaster for our society as it exists today. My hope is that we will adapt and make the necessary societal changes so that we can reap the benefits of this technology.

Everyone assumes that an ASI will be able to augment itself and learn exponentially. I suspect that this will be true if the nature of the brain is defined by a single algorithm. If the brain is not defined by a single algorithm and is instead a big ball of complexity then our ASI's will not be able to grow exponentially any more than we can (they will likely not really understand their own consciousness, just like we don't).

If a single algorithm defines the brain, then I suspect humans will be able to augment their brains with machine intelligence as well. If we can augment our brains, then we're playing the same game as the machines.

If it proves impossible to augment our intelligence, I suspect that an ASI would still preserve humanity if only to preserve us for future potentialities.

ASI are much more fit for space travel than us. 1) Not nearly as sensitive to radiation, so less shielding, so less fuel. 2) Much less stringent environmental requirements (no heating, cooling, air, food, etc.) 3) Ability to sleep for incredibly long periods means ASI is far more suited than us for exploring the cosmos. I suspect that an ASI might leave us alone simply because the universe is so vast, and entirely open to it.

Re: Machine intelligence, part 1

#52

Post by fchollet »

mej10 wrote: post #48
fchollet wrote: post #41

I have yet to meet a serious AI researcher who worries about AI ending the human race. At every AI conference I've been to lately, some guy would inevitably ask the same question to the speaker: "do you think we should be concerned by superhuman AI?". The answer was always the same, for instance from Andrew Ng at the Deep Learning Summit a few weeks ago: "dude, stop it. That's such a distraction". > One of my top 4 f…

How about Shane Legg (One of the cofounders of DeepMind)? http://lesswrong.com/lw/691/qa_with_shane_legg_on_risks_from... Quote: Q6: Do possible risks from AI outweigh other possible existential risks, e.g. risks associated with the possibility of advanced nanotechnology? Shane Legg: It's my number 1 risk for this century, with an engineered biological pathogen coming a close second (though I know little about the la…

Shane Legg is known for being a co-founder of DeepMind, in a business role (as I understand). He's a complete nobody as a researcher (is he even an AI researcher? I would be surprised).

The big names of deep learning have all taken a vocal stance against the recent end-of-the-world punditry (most notably Yann LeCun and Andrew Ng). Also notable: roboticist Rodney Brooks http://www.rethinkrobotics.com/artificial-intelligence-tool-...

Re: Machine intelligence, part 1

#53

Post by SquidMagnet »

veryluckyxyz wrote: post #42
SquidMagnet wrote: post #27

Earlier quoted context omitted.

1. Machine intelligence, traditionally called artificial intelligence, which surpasses human intelligence. 2. Your category (b) is generally the primary concern in these types of discussions. 3. The anecdote of the progress of humanity. Compare the impact of human life/intelligence vs. evolutionary relatives like chimpanzees. I do not know that chimps have hunted species out of existence, for instance, but people hav…

Thanks. Seems to me these anecdotes have to do with humans. So is the implicit assumption that machines will do what humans are doing ('bad' things) but at several orders of magnitude faster and without the ability to comprehend longer-term consequences of their actions any more than humans do at the present time?

Sort of. 'Bad' here is of course an extremely subjective term. And it may not be the case that the machines do not understand the longer-term consequences of their actions; they could understand full well, but they could know that the preservation of humanity is not important (for whatever reason). So, we might not matter to them. We matter to us though, so that would be a problem for us as things stand now.

Re: Machine intelligence, part 1

#54

Post by spot »

opinions of 187 scientists and other writers and thinkers (Dennett, Smolin, Rees, Dyson, Rushkoff, O'Reilly, Coupland, Kelly, etc etc) on thinking machines on edge.org (and many of them address the threat issue):

http://edge.org/annual-question/q2015

including mine:

http://edge.org/response-detail/26219

Re: Machine intelligence, part 1

#55

Post by cryoshon »

ep103 wrote: post #34
iterationx wrote: post #30

I'm more concerned about AI law enforcement. NSA eavesdropping plus an intelligent agent assigned to you is a powerful combination.

Exactly, It seems like the biggest threat isn't some dystopian future, its rather the ability for automation to lock in and increasingly enforce the inequalities and prejudices of our current system.

Yep. This is my take as well.

Machine intelligence is only as useful to mankind at large as the individual humans who control and direct it. In the current way of things, bad actors are the ones most likely to control machine intelligences, meaning we're going to be at a growing disadvantage relative to them as time goes by.

Re: Machine intelligence, part 1

#56

Post by seiji »

The red flag here is the mention of a fitness function. True AI and fitness functions have nothing in common.

What's the fitness function of yourself as an intelligence? True AI is as fractured and contradictory as a human brain, just running on a different substrate. When talking about AI, one must make sure one does not mean "an infinite loop attached to a robot." That's not AI, that's... an infinite loop attached to a robot (KILL ALL HUMANS, no generative thought, etc).

As far as "bad AI" goes, we already have horrible dumbass humans, so the only thing evil AI can do is be bad faster and in more clever ways. Yes, it's something to worry about, but I'm worried more about insane world leaders running around acting like pouty emo teenagers with simultaneous delusions of grandeur and delusions they are fulfilling desert prophecies from 4 kiloyears ago.

Re: Machine intelligence, part 1

#57

Post by sudioStudio64 »

Mr Altman should be far more worried about the hungry and homeless residents of the bay area taking up pitchforks against their startup neighbors.

Climate change is also a huge threat that isn't just "likely".

I don't want to be dismissive, but seriously, their are some radically dangerous things facing humanity right now. This isn't one.

{Edit...my keyboard sucks.}

Re: Machine intelligence, part 1

#58

Post by sudioStudio64 »

spot wrote: post #54

opinions of 187 scientists and other writers and thinkers (Dennett, Smolin, Rees, Dyson, Rushkoff, O'Reilly, Coupland, Kelly, etc etc) on thinking machines on edge.org (and many of them address the threat issue): http://edge.org/annual-question/q2015 including mine: http://edge.org/response-detail/26219

It kind of reminds me of predictions about traveling faster than sound igniting the atmosphere.

Re: Machine intelligence, part 1

#60

Post by veryluckyxyz »

SquidMagnet wrote: post #53
veryluckyxyz wrote: post #42

Earlier quoted context omitted.

Thanks. Seems to me these anecdotes have to do with humans. So is the implicit assumption that machines will do what humans are doing ('bad' things) but at several orders of magnitude faster and without the ability to comprehend longer-term consequences of their actions any more than humans do at the present time?

Sort of. 'Bad' here is of course an extremely subjective term. And it may not be the case that the machines do not understand the longer-term consequences of their actions; they could understand full well, but they could know that the preservation of humanity is not important (for whatever reason). So, we might not matter to them. We matter to us though, so that would be a problem for us as things stand now.

Thanks.
Post reply on HN