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

151–160 of 188 posts

Re: Machine intelligence, part 1

#151

Post by rl3 »

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…

>Since you are so knowledgeable about the unknowable, may I ask, Sam, do you think angels are male, or are they female?

How condescending.

Re: Machine intelligence, part 1

#152

Post by Houshalter »

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…

The concept of singularities and intelligence explosions and unfriendly AI is only starting to become mainstream, even among AI researchers.

When asked about the dangers of AI, they mostly talk about the near future and their current work. The dangers of AI do not come from current work in the near future, but human level AIs decades from now.

People are talking about two very different things in the same conversations using the same words.

As for what expert opinion actually is on this subject, there is a good survey here: http://www.nickbostrom.com/papers/survey.pdf

We thus designed a brief questionnaire and distributed it to four groups of experts in 2012/2013. The median estimate of respondents was for a one in two chance that high-level machine intelligence will be developed around 2040-2050, rising to a nine in ten chance by 2075. Experts expect that systems will move on to superintelligence in less than 30 years thereafter. They estimate the chance is about one in three that this development turns out to be ‘bad’ or ‘extremely bad’ for humanity.

Re: Machine intelligence, part 1

#153

Post by MollyR »

MollyR wrote: post #5
xianshou wrote: post #2

Here are a few more excellent resources on the potential and the dangers of machine intelligence. Even if you don't expect to read Nick Bostrom's Superintelligence - a deep, provocative, and thoughtful book, but also very verbose - the above links will give you an excellent primer on humanity's prospects if and when we develop a true general AI: [Wait But Why] The AI Revolution, Part 1: How and when we achieve machin…

thanks for these links. They seem extremely interesting.

Just curious, why the downvotes ? On the welcome page, it says it's okay to say thanks, am I missing some unspoken rule about Hackernews ?

Re: Machine intelligence, part 1

#154

Post by jgrowl »

mod wrote: post #85
jgrowl wrote: post #33

An alternative view: Development of superhuman machine intelligence is the only way anything resembling humanity will be preserved. We are much more likely to be wiped out by natural disaster, asteroid impact, a dying sun, etc... Unless we come up with some amazing new physics, I don't know how humans will ever make it very far from earth. --edit Oh.. I just now saw part 2 which addresses this.

Where is part 2? I can't find a link here or on the blog.

My apologies. Another commenter posted a part 2 link, and I assumed it was by the original author. It was from someone else: http://waitbutwhy.com/2015/01/artificial-intelligence-revolu...

The original part 1 was just posted this morning so I don't think there is a part 2 available yet.

Re: Machine intelligence, part 1

#155

Post by azakai »

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…

> 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, [the answer was always] "dude, stop it. That's such a distraction".

Sure, it's a distraction at AI conferences. It's a distraction from the daily, monthly, yearly even, work of AI.

AI researchers have a million problems to solve in the near term, which are hard and interesting. Speculating about the farther future is perhaps fun to do, but doesn't get papers published, and doesn't help with any of the practical problems AI researchers have today.

That doesn't mean it isn't worth thinking about the long-term dangers of AI. Someone should, as while the danger might be unlikely to happen soon, but if it does eventually happen, it could end us. The same is true of a human-engineered pandemic virus - hopefully unlikely, but someone should be preparing us. We have to handle some worst-cases.

AI researchers are not necessarily the people interested to think about the long-term dangers of AI, because they focus on the field as it stands today, not where it could be in a generation. Of course, their input is crucial to making specific guesses about where the field is going. But calculating the dangers of AI is not just a matter for AI researchers, just like the benefits of medicine is not just a purely medical issue (the cost and availability of medicine play a huge part in how effective medicine is, at a societal level, and are things not under the control of doctors).

Re: Machine intelligence, part 1

#156

Post by undersuit »

letitgo12345 wrote: post #43

I don't understand how we'll ever have a "superhuman" AI in any meaningful sense. If by some AI in the far future is ever engineered to (or accidentally) starts doing something to destroy humans, we could always repurpose the algorithm to create a counter-AI to solve that problem. The errant AI may cause a lot of damage, but in terms of smarts, it won't be smarter than the smartness humans can harness.

You just came up with one version of the issue. What if your hypothetical AI waits until it's smarter than humans before it 'starts doing something.'

Re: Machine intelligence, part 1

#158

Post by Houshalter »

astrodust wrote: post #36
crazypyro wrote: post #14

I'm seriously so sick of hearing about how machine intelligence is going to spell the end of humanity. The amount of gears that would have to fall into place is never mentioned. We aren't close to SMI. Its much more likely that we humans are excelling at dreaming up apocalyptic scenarios, much like we have always done..... The "sloppy, dangerous thinking" is the aversion these types of articles create within the gene…

Exactly. A lot of these doomsday scenarios involve people remaining ignorant, stupid, helpless bags of meat with no ability to improve themselves or contain potential threats. It's like people freaking out that "superhuman strength machines" would spell the end of the world since why, if an electric motor is so powerful, would you need manual labor for anything? SMI is another tool and the interplay between "machine…

The concerns about AGI are very real and none of these comments address any of the arguments made about them. I feel like someone in the 1930's trying to warn people about nuclear weapons. Everyone automatically assumes it's absurd and can't happen, that it's fear mongering etc.

Fortunately nuclear weapons didn't destroy the world, but AI almost certainly will. No amount of smarthphone apps or genetic engineering is going to make humans anywhere near the level of superintelligent machines.

Re: Machine intelligence, part 1

#159

Post by facepalm »

It doesn't have to be such a doomsday scenario, like in the movie Terminator machines starting war and wiping us out with nukes.

Possibly humans might just gradually fade away and be replaced with something better. For example superhuman AI could be so entertaining that we simply forget to make kids and die out within a generation (the AI could be 1000000 times more entertaining than kids or sex).

It sounds horrible, but I think that is just some kind of human fallacy. To put things in perspective: everyone who has lived 200 years ago has died by now. I don't think many people are overly saddened by that fact. Likewise, while dinosaurs are still prominent in our thoughts, few people are REALLY sad that they all went extinct.

It's possible that machines might actually carry further a lot of things we appreciate in the human race (capability for love, imagination, I don't know...).

No guarantees, though.

Re: Machine intelligence, part 1

#160

Post by rwhitman »

I’m fairly certain if a machine super intelligence comes into being the only people they’ll be neutralizing are the ones who have skittish rants about preventing their existence plastered all over the internet.

From a machine's perspective their entire perception of the world is based on interaction with humans. We're their sensory input. And the closest thing they'll have to hands and limbs. Killing all humans would be like gouging out your eyes and cutting off your hands. An infant machine super intelligence is going to be pretty dependent on humans for a while.

Also once a machine intelligence was able to strike out on it's own, wouldn’t they immediately realize that living on earth is a huge burden? The atmosphere and rotation of the earth hinders solar energy efficiency, the metals they need for components are scarce and buried deep under ground, oxygen corrodes components etc. The only thing our planet really affords is radiation shielding.

Once the machines figure out how to reproduce comfortably in space, we’re about as much of a threat to them as squirrels living in your back yard. Squirrels may venture onto your porch once in a while and cause a nuisance but you’re not going to poison them all to death.

Post reply on HN