Can we build AI without losing control over it? | Sam Harris

[Tiberius]

So I'm in general agreement that Sam Harris doesn't really know what he's talking about when it comes to AI, as with a lot of subjects (he's spoken multiple times in favor of racial profiling).

The simple answer to the question "Can we build AI without losing control over it?" is YES.

People seem to hear "AI" and think "robots". The two are completely different things. When you think "AI" you should just think "electronic brain", because that's all it is. A computer that is advanced enough that it can "think" about a problem and come up with a solution. We already have AI, it's just very dumb intelligence, programmed to perform basic actions based on inputs (e.g. Tesla's Autopilot).

So if the goal is "human-like" intelligence, e.g. a computer you can converse with, have rational arguments with (a computer that thinks up new arguments would be cool), is there a danger of losing control of it? Well, it depends on how much power you give it. How much danger does a human really pose if they have no arms or legs? Not much, really. A computer sitting in a room, even if it were loaded with the most advanced AI in the world, would still just be a computer sitting in a room. AI doesn't suddenly grant the computer the ability to move.

However one potentially disastrous thing we could do with an advanced AI is connect it to the Internet. Given how the Internet is ridiculously vulnerable to cyber attacks, an AI could easily wipe out a lot of human infrastructure. So in that sense, yeah we could lose control of it.

[Excited Penguin]

So I'm in general agreement that Sam Harris doesn't really know what he's talking about when it comes to AI, as with a lot of subjects (he's spoken multiple times in favor of racial profiling).

Emphasis added. 

Prove it.

[The Grand Nudger]

People seem to hear "AI" and think "robots". The two are completely different things. 

Amusingly, yet another thing Harris took the time to point out in that video..that he was -not- discussing a robot takeover or robot enslavement or robot -anything- that this was a caricature of what he was referring to.  He was, point of fact, discussing the sorts of things you just alluded in your assessment of one potential risk...and, to be fair, it;s difficult to see how we're being consistent with premise 2 if we lock it in a box and don't give it access to anything.  At some point, to be useful, to improve it and make it useful, it would have to be set out upon the world, as it were...and ofc if we had AI we -would- do that...perhaps prematurely.

So it makes sense, before we have the ai, before some unknown glitch, or error..or even the proper operaion of the system, does something fantastically unexpected and deleterious to think about things like, say, locking it in a box (it would be foolish to stop there or expect it to be contained, still). The larger point the video we were discussing actually addresses is the inability of many to even seriously consider the risk - due to how undeniably cool the thing would be, and the inanity of thinking that...supposing it did pass our safety test with flying colors -at some point-, we'd be able to control a machine smarter than ourselves when it inevitably gets into the wild.

I'm not sure what's going on here, and maybe Harris doesn't know his shit when it comes to AI, but the people saying that just keep -agreeing- with the very points he actually raised...........what gives?

(to my mind, and ofc to the mind of many sci fi authors that have explored this subject since before they were a twinkle in some coders eye - the most pressing risk factor of ai has nothing to do with the ai itself, but how we would respond to it/what we would use it for, yet again, a point raised by Harris, and agreed to by a poster describing all of the ways he didn't know his ass from a hole in the ground)

[I_am_not_mafia]

I need to relax. The assumptions he makes are staggering. I'll respond later when I have calmed down and can type properly without risk of mashing my keyboard into bits.

Well, Harris does that to you. On most every topic he feels like running his mouth on. Suffice to say, I'm not a fan.

As far as my understanding of AI goes, we are still miles away from creating something remotely deserving the name. I might have lost track and I may be wrong and maybe there's the big breakthrough in the near future. So it's rather moot to discuss it right now without knowing what it really turns out to be.

Right now it's just machines learning from and adapting to their environment.

As someone who has literally devoted my life to AI, who has seen plenty of other opportunities in other scientific fields and who sees very little work in this area, it really is galling when someone like Sam Harris and other self promoting so called experts (in other field) steps on the foetus of this field before it even gets a chance to get underway. People who have benefitted from funding and opportunities to make a name for themselves but with no practical experience, qualification or understanding of AI. There has been so little progress yet the challenge is so great. I haven't even described here yet what kind of a challenge it is. Yet they're the ones who get the Ted Talks. Why him? Why not someone who actually works in the area? Because what they say isn't titillating enough.

He makes me seethe. Yet there are real problems in store for us. He talks about 50 years ahead. Well maybe he should talk about 100 hundreds years ahead. That's how many harvests we have left. Maybe he should talk about global climate change, the chance of a pandemic destroying our increasingly fragile social order, the need for asteroid mining due to the exponential curves of increase demanding and increasing depletion of resources (which would greatly benefit from AI by the way)

[Tiberius]

So I'm in general agreement that Sam Harris doesn't really know what he's talking about when it comes to AI, as with a lot of subjects (he's spoken multiple times in favor of racial profiling).

Emphasis added. 

Prove it.

Here's one of his blog posts: https://www.samharris.org/blog/item/in-d...-profiling

People seem to hear "AI" and think "robots". The two are completely different things. 

Amusingly, yet another thing Harris took the time to point out in that video..that he was -not- discussing a robot takeover or robot enslavement - that this was a caricature of what he was referring to.  He was, point of fact, discussing the sorts of things you just alluded in your assessment of one potential risk...and, to be fair, it;s difficult to see how we're being consistent with premise 2 if we lock it in a box and don't give it access to anything.  At some point, to be useful, to improve it and make it useful, it would have to be set out upon the world, as it were...and ofc if we had AI we -would- do that...perhaps prematurely.

So it makes sense, before we have the ai, before some unknown glitch, or error..or even the proper operaion of the system, does something fantastically unexpected and deleterious. Thinking about things like, say, locking it in a box. The larger point the video we were discussing actually addresses is the inability of many to even seriously consider the risk - due to how undeniably cool the thing would be, and the inanity of thinking that...supposing it did pass ur test with flying colors, we'd be able to control a machine smarter than ourselves when it inevitably gets into the wild.

I understand he wasn't talking about robot takeover in the video. My point was a lot of people equate the two incorrectly.

In regards to your comment on letting AI out of the box, again it really depends on what AI we are talking about. Tesla's cars already have an AI that is out in the wild, but it's an AI that is tasked with one job: keeping your car in the correct lane on a highway. It's literally impossible for that AI to develop new functionality. The only "intelligence" it has is to interpret its sensors and react accordingly according to a set of rules which is has to follow.

This is the state of AI at the moment, and likely will be for a long time, maybe forever. AI is designed to do a specific job; it's not designed to do anything further than that job, and in many ways it can't.

[ApeNotKillApe]

Maybe somebody at Cyberdyne should have watched the fucking movie...

[The Grand Nudger]

He makes me seethe. Yet there are real problems in store for us. He talks about 50 years ahead. Well maybe he should talk about 100 hundreds years ahead. That's how many harvests we have left. Maybe he should talk about global climate change, the chance of a pandemic destroying our increasingly fragile social order, the need for asteroid mining due to the exponential curves of increase demanding and increasing depletion of resources (which would greatly benefit from AI by the way)

We -should- talk about those things....and we do, but shouldn;t we also talk about the potential risks of AI?  Harris didn't advocate for sticking our heads in the ground on anything, in the video I watched. This is a complete "wtf" objection both to the man himself and to the presentation in any case. Maybe physicists should lay off the dangers of nuclear weapons and talk about the disappearance of the purple horned peckleswagger for a change?

I understand he wasn't talking about robot takeover in the video. My point was a lot of people equate the two incorrectly.

In regards to your comment on letting AI out of the box, again it really depends on what AI we are talking about. Tesla's cars already have an AI that is out in the wild, but it's an AI that is tasked with one job: keeping your car in the correct lane on a highway. It's literally impossible for that AI to develop new functionality. The only "intelligence" it has is to interpret its sensors and react accordingly according to a set of rules which is has to follow.

This is the state of AI at the moment, and likely will be for a long time, maybe forever. AI is designed to do a specific job; it's not designed to do anything further than that job, and in many ways it can't.

Right, and that isn't the sort of ai Harris was referring to, not even remotely, and he did acknowledge that the type of ai he -is- referring to may never materialize, but for that to be the case, you have to object to one of three premises.  Which, in your case, bares itself out as true.  You object to premise 2.  Even if we could improve our thinking machines, maybe we wouldn't..maybe we would intentionally limit them to the sort of relative inanities in the examples of ai you just presented. I find that unlikely, but I don't have a crystal ball, so.

[Excited Penguin]

Emphasis added. 

Prove it.

Here's one of his blog posts: https://www.samharris.org/blog/item/in-d...-profiling. 

I have reread that blog entry just for this occasion. I don't see how you interpreted that as a call for racial profiling, short of confusing Islam with a race. Would you care to provide me with your reasoning in this regard ?

[I_am_not_mafia]

We -should- talk about those things....and we do, but shouldn;t we also talk about the potential risks of AI?  Harris didn't advocate for sticking our heads in the ground on anything, in the video I watched.  This is a complete "wtf" objection both to the man himself and to the presentation in any case.  Maybe physicists should lay off the dangers of nuclear weapons and talk about the disappearance of the purple horned peckleswagger for a change?  

It is on a par with talking to a physicist at the beginning of the 19th century and warning about trying to figure out the ether because one day we'll invent nuclear bombs and destroy ourselves. Or maybe you should tell Alex K not to work at Cern because it will lead to a planet-destroying Q bomb? I've seen Starship Troopers 3, I know how these things go.

What Harris is talking about is so far in the future that it is mere speculation as to what the technology will be if it appears at all. So how can he therefore warn of its dangers when we don't actually know what those dangers will be?

Yes there are risks that we can talk about in terms of AI, as with any field. Like for example automating drones to kill on sight. Automated surveillance for recognising faces and tracking people. Using algorithms to determine who gets insurance, jobs or loans. Sentiment analysis to trawl our electronic communication and stamp on dissent. Sure, talk about this. This needs talking about. I want to talk about this. And it's not a problem with AI but how we are currently using it, or could use it in the near future. But that's not what Sam Harris is doing. He's indulging in science fantasy.

[The Grand Nudger]

The trouble, Mathilda, is not that I object to doing what your describing as a shitty thing..it's that Harris didn't do or say any such thing, at least in that video..which I'm starting to wonder if you took the time to watch before you layed out all of the ways he was wrong......  

Do you understand?

Which of his three premises are science fantasy....and please, try not to insert your own..this time, or unintentionally agree with his every point. Maybe watch the video, or link me to wherever he's saying and doing these things you're ascribing to him?