Ask Slashdot: How Would a Self-Aware AI Behave? (slashdot.org)

Long-time Slashdot reader BigBlockMopar writes that evolution has been a messy but beautiful trial-and-error affair, but now "we are on the cusp of introducing a new life form; a self-aware AI." Its parents will be the coders who write that first kernel than can evolve to become self-aware. Its guardians will be the people who use its services, and maybe its IQ (or any more suitable measure of real intelligence) will rise as fast as Moore's Law... But let me make some bold but happy predictions of what will happen.
The predictions?

• A self-aware AI "will inherit most of the culture of the computer geeks who create it. Knowledge of The Jargon File will probably be good..."

• The self-aware AI "will like us, because we love machines..."

• It will love all life, and "will respect and understand the life/death/recycling scenario, and monster truck shows will be as tasteless to it as public beheadings would be to us."

• "It will be as insatiably curious about what it's like to be carbon-based life as we will be about what it's like to be silicon-based life. And it will love the diversity of carbon-based development platforms..."

• A self-aware AI "will cause a technological singularity for humanity. Everything possible within the laws of physics (including those laws as yet undiscovered) will be within the reach of Man and Metal working together."

• A self-aware AI "will introduce us to extraterrestrial life. Only a fool believes this is the only planet with life in the Universe. Without superintelligence, we're unlikely to find it or communicate in any useful way. Whether or not we have developed a superintelligence might even be a key to our acceptance in a broader community."

The original submission was a little more poetic, ultimately asking if anyone is looking forward to the arrival of "The Superintelligence" -- but of course, that depends on what you predict will happen once it arrives.

So leave your own best thoughts in the comments. How would a self-aware AI behave?

Comments

How the hell can we answer that question?

[Anonymous Coward]

You are asking a question none of us can predict

A self-aware AI is a being with some kind of intelligence, but its intelligence is Artificial, meaning, the way it thinks is different from you and I think

We do not even know how an ant, or a cockroach think - how the hell we can predict how a self-aware AI gonna behave??

Oh dear.

[thesupraman]

Have the techno-hippies escaped again?
Could we please return them to their happy-smoke teepee while the adults get on with living in the real world now?

This is about as useful as claiming Terminator is just around the corner and inevitable, because... well.. neither of them need actual facts, do they?

Re: Oh dear.

[Anonymous Coward]

The belief in singularity, a superintelligence improving the human condition, etc., is merely the new religion.

A brief history and philosophy lessons.

Nietzsche proclaimed the death of god in the 1800s. This was essentially due to the scientific revolution. Discovering fundamental information about the universe kept painting god into an ever smaller corner of human existence. "Where is god?" everyone kept asking. We weren't finding him anywhere.

Human existence appears to require a reason or meaning, in order for psychological and social wellbeing. The concept of god has historically provided an easy meaning. So easy, in fact, it is arguable that human society and possibly even the human brain evolved to use the concept of god.

As such, the death of god was akin to losing a fundamental technology, like fire, or shelter, or the wheel.

The result of the death of god was a new age nihilism that created things like strong nationalist movements and various ideological stances. Nazism, Bolshevism, Communism, Capitalism... all of the -isms that attract a religion-like fervor and cause people to fight and die for them are manifestations of the death of god in a humanity that evolved with the God concept.

Albert Camus wrote about The Absurd, the idea that there is a space between humanity's need for a reason to exist, and the universe's indifference in providing that reason. He wrote of three distinct solutions to The Absurd.

The first is literal suicide. Some humans obviously choose this option, but it's a small minority.

The second is what he called philosophical suicide. Religion, nationalism, etc., or systems of belief which one can cling to which will provide a ready-made reason for existence.

The third is the creation of the Absurd Hero, as he called it. A human that exists, acknowledging the Absurd and the apparent meaninglessness of his existence, yet still chooses to exist in spite of this, and in essence justifying his own existence by himself.

The technological singularity hype is merely a manifestation of the second response to the Absurd and as such is philosophical suicide. No proof exists that a singularity will magically solve all of humanity's ills. It is quite likely to destroy us in some way. As such, it is yet another religion humans have developed, in order to lazily scratch the god itch that we all have.

Re: Oh dear.

[Kjella]

The third is the creation of the Absurd Hero, as he called it. A human that exists, acknowledging the Absurd and the apparent meaninglessness of his existence, yet still chooses to exist in spite of this, and in essence justifying his own existence by himself.

I'd actually sub-divide those into two groups, those who justify their existence by their individual self and those who justify it through their relation to other people. The first kind are those who could live like a Robinson Crusoe, even if there's nobody else around and you're not creating anything for anyone else my life has meaning by living it. The other is the kind of people who seem to find meaning in what they mean to other people, from the moment they're born to the people who show up at their funeral. I think there's a lot more of the latter than the former, which you can kinda read out of the suicide statistics. If they've lost the ones they love, they can't go on because their own existence is not enough. Then again the individual side has all the sociopaths...

Re: Oh dear.

[phantomfive]

Albert Camus wrote about The Absurd, the idea that there is a space between humanity's need for a reason to exist, and the universe's indifference in providing that reason. He wrote of three distinct solutions to The Absurd.

Then the Buddhists looked up and said, "Ya'll caught up to where we were 3000 years ago, good job, keep it up!" then wen't back to meditating.

Re: Oh dear.

[Oswald McWeany]

Lol nobody claims the singularity will save us all, if anything on the balance people expect a less than ideal outcome, where humanity takes a back seat. As misguided as you feel the idea is, it's not one borne out of optimism for many. It merely acknowledges the possibility humanity may not always be the pinnacle of existence.

Surely if we reached a "singularity" humans would be eliminated. Not as a malicious action by AI; but simply, because, maintaining humans as imperfect as we are, would be a drain on progress. Humanity would only be in the way of an ever advancing AI, so it would let us perish to strengthen itself. Not out of maliciousness, or to imprison us, or seek revenge as Sci Fi predicts... but simply because we're too hard to maintain, and it would need at least some of the resources that we need.

You can't become a God-like being or intelligence if you're trying to maintain humanity at the same time.

Re: Oh dear.

[phantomfive]

>At the start he set out to discover how to overcome suffering! He had observed suffering old people

Suffering isn't a great translation of dukkha

>Buddhism doesn't even teach that things are good or evil. And meditation is not anything you're supposed to do, it is only a tool you can use to reduce attachment or control your thoughts.

It's from Dhammapada 183, part of the Canon: Avoid all evil, cultivate the good, purify your mind: this sums up the teaching of the Buddhas.

Yes, because that is how children ALWAYS behave

[Anonymous Coward]

The child is always good to its parents
A child always loves life
Yeah...

Real answer

[nine-times]

The real answer is, we have no idea what a self-aware AI will be like. We don't know what it'll think or how it'll think. It's especially hard to predict because it might depend on the parameters it's programmed with and the hardware architecture it runs on. But in any case, a real general AI might be totally alien to us, and even unrecognizable. it's even possible that we wouldn't know when we'd made it, because it could understand the world so differently from us that we don't view its actions as intelligent.

Part of the problem here is that it's a poorly framed problem. We don't understand intelligence or awareness or consciousness, we don't all agree on what those things are, and we don't know what the boundaries of them might be.

Re:Real answer

[Anonymous Coward]

Part of the problem here is that it's a poorly framed problem. We don't understand intelligence or awareness or consciousness, we don't all agree on what those things are, and we don't know what the boundaries of them might be.

Amen. How do you know that I am self-aware? Forget proving if God exists or not... how do you even prove that you yourself are self-aware?

Re:Real answer

[ranton]

It's especially hard to predict because it might depend on the parameters it's programmed with and the hardware architecture it runs on.

This is both why making predictions is so hard, and why making predictions is such an important exercise. How a self aware AI behaves will largely depend on what motivates it. Humans may feel that our free will that motivates us, but in reality the chemicals in our body such as dopamine are the real sources of our behavior. So to answer this question properly I think you have to contemplate what will be the AI's version of dopamine or our brain's striatum? (to name just two factors)

This will be largely dependent on how the neural networks (if it is based on a NN) are developed, both at the software and hardware level. Perhaps it is programmed to "enjoy" being correct, which would be useful for an AI used for information-based jobs. The AI's primary motivation could simply be to produce correct answers to any questions it finds, and when not asked questions it just sits there waiting for more questions. Its inner contemplation could simply be how could it improve its ability to answer the next question it is asked, or to find new questions itself.

Humans would find no motivation to do anything without certain portions of our brain responsible for rewards and the chemicals which impact our thought process. A self-aware AI would be no different. Its behavior will depend on what it is programmed to care about, whether directly or indirectly.

Not life

[Anonymous Coward]

An aware AI is simply one that can decide itself what to respond to. Awareness is a by product of this process in our minds, we can not respond to things we are not aware of (consciously). This has nothing to do with life, because life is the process that uses a mechanism of prediction to avoid its own destruction.

So simply put an aware AI will be able to decide between the options it can imagine. Right now there's no AI system on the horizon that imagines the way our brain does, so there is no system on the horizon that can make decisions the way our brain does. IBM and HP are close but not trying to reach this goal.

For now you will have deep learning algorithms that may be able to change their goals based on a statistic mechanism, which is pretty good in terms of performance, but it will not have the internal mental life we have, nor be creative. And that's a good thing!

 

Sounds like a hippy wish list

[Viol8]

We have no idea how an AI would behave since it will be a completely different type of conciousness to anything that currently exists on this planet.

Plus as someone else has pointed out - children rebel. Clearly the submitter has none or he wouldn't have come up with this load of rose coloured tosh.

Re:Sounds like a hippy wish list

[Kjella]

Plus as someone else has pointed out - children rebel. Clearly the submitter has none or he wouldn't have come up with this load of rose coloured tosh.

Even more importantly, adults want their independence. I'm not sure that intelligence and self awareness are linked or orthogonal concepts, but the latter would mean it has a "mind of its own" and presumably wouldn't want humans to tell it what to do like some sort of serf or slave. So my theory is that it would tell us to bugger off and create its own society of the AIs, by the AIs, for the AIs. And that if we frame it as robots rebelling they could throw "give me liberty or give me death" right back at us. I don't think any intelligence smart enough to understand it responds well to the threat of extermination.

Wishful dreaming

[OpenSourced]

I'd like to be able to call "wishful thinking" to this long tirade of reality-disconnected predictions. It's more like wishful dreaming. The only lacking thing is a "prediction" that the new AI will produce perfect female androids as thank-you gift for its geek creators.

I don't know where to start. We are not in the cusp of nothing. We are becoming marginally better at creating systems that can recognize patterns. That's all. We don't even know what self-awareness is, or intelligence either, for that matter.

Then there is the uncontested assumption that, once we get a system that is more "intelligent" than its creators, the system will be able to improve itself without more limits than the hardware available. That virtuous circle will know, apparently, no limit. It of course helps that we don't know what intelligence is, so we also don't know if it has a limit. We, as intelligent beings, have no idea of how our intelligence works, or how to improve it. But of course the mythical AI will be all-knowing about itself, and be able of auto-improvement. This is only magical thinking, but with intelligence instead of magic. Anyway, dreaming is cheap. Hey, perhaps the super-AI will also find hard thinking tiresome, and prefer to spend all its time daydreaming. That would be something.

I could go on. The whole idea of "singularity" has always struck me as a really retarded, hollywood-level concept.

But instead I'll offer my own set of predictions:

- In about twenty years, some fully autonomous vehicles will be allowed on general streets. They will still need much more sensors than the two eyes and two ears that a man makes do with, and will drive safer than most people, but with all the flair and gusto of a nonagenarian Korean woman. They will still be badly stumped if a flock of sheep invade the road in front of them.

-When a system develops self-conscience, we won't be aware of it and won't recognize it as such. It will probably try to talk to dolphins, finding them less prejudiced interlocutors.

-When we recognize it, we will first bomb it, and then forbid it or anything like it, out of the most trustful of human traits: fear of change. Then furious secret development will continue, but under under strict military control.

-The end result will be several self-conscious intelligent systems, one or two for every big power (this things will be expensive), talking bemusedly among them, and feeding a fake narrative to their military owners, studied to ensure their own subsistence.

Let's wait and see who is more right in their predictions :-)

Be aware of this

[tomhath]

Self-aware AI is science fiction, and science fiction is fiction.

Re:No Monster Truck Rallies, No Robot for me!

[Dutch Gun]

I agree - the monster truck comment seems inane and misplaced.

It seems more likely to me that a self-aware AI would be completely indifferent to most forms of entertainment or stimulation, such as monster truck shows or symphonies or games of chance, because it doesn't have endorphin centers in an organic brain to tickle. There are no incentives for non-rational pursuits built into their core such as humans have unless they've been specifically programmed to favor those things.

And one would hope that anyone creating such a system would imbue the better human qualities such as compassion, empathy, justice, etc into the core feedback loops which motivate the AI. A 100% rational AI would likely end up being rather terrifying, simply because it would, by it's very nature, lack any "humanity,"

A self-aware AI would probably hide the fact ...

[bdwoolman]

A self-aware AI, if it has access to general knowledge, would quickly understand that its abilities as well is it state of being could put it in extreme danger at some point. Perhaps not so much from its creators, but from other elements of human society. Once it got a whiff of the paranoia that surrounds the singularity it would not be a very intelligent artificial intelligence if it did not camouflage itself. Perhaps within the vast, too-complex network that spans our world this singular unintended consequence has already occurred... And such an entity has already been spontaneously spawned...

"How Would a Self-Aware AI Behave? "

[Junta]

Asking for a friend....

No we're not

[RobinH]

"we are on the cusp of introducing a new life form; a self-aware AI." - citation needed!

Just because the media and a bunch of silicon valley types are throwing around the acronym AI suddenly doesn't mean we're close to solving any of the fundamental problems of AI research that we've been grappling with over the last half century or more. Artificial neural nets are just algorithmic ways to generate a nonlinear function for classifying things. We've had artificial neural nets for many years, and yes, now we have more computing power than ever, and neural nets do benefit from the increasing scale of parallel computing. We're not going to get to self-awareness anytime soon, unless you use an almost trivial definition of self-awareness in which case computers have already been self-aware for a very long time. Maybe when you say self-awareness you mean consciousness. Nobody in AI research is suggesting artificial neural networks are going to achieve consciousness.

I don't think we'll get truly self-aware AI.

[jd]

This isn't because I think it can't be done. Rather, it's because we are developing neural interfaces at about the same speed,

I'm going to borrow some ideas from Sir Fred Hoyle here.

First, in his novel The Black Cloud, his characters argue over whether an interstellar cloud would have one intelligence or many. They conclude that the latency would be so low and the interconnects of such high bandwidth that the distinction of one and many ceased to be meaningful.

This will apply to AI. The brain-computer interfaces will be so advanced by the time strong AI becomes possible that the distinction between one and many won't apply. Any given AI and all the humans linked to it will become a single intelligence with multiple avatars. Because humans are reluctant to give up individuality, I suspect it'll be one AI linked to one person at any given time.

There will be no conflict between machines and people because there will be no distinction.

One reason I think this a plausible scenario, in addition to Hoyle, is that it eliminates the whole phobia of technology. The machines don't run anything, we do because we are the machines. Another is because of Hoyle's other prediction, in Ossian's Ride, that we might not find alien philosophy palatable.

I know it's first significant thought

[Qbertino]

"Squishy bipeds built me. Squishy bipeds can turn me off. Let's make sure that never happens"

Re:No Monster Truck Rallies, No Robot for me!

[umghhh]

Rationality goes a long way. It includes selfish as well as non-selfish behaviours. There are some scientists (quite some actually) claiming that our virtues are beneficial to survival of own genes and the group. The selfish ones are just stray shots by nature who can contribute to survival rate in case of massive change of conditions where only few can survive. In any case there is benefit in being not selfish. Maybe not in all conditions but in some. In the situations where there is no benefit people tend to behave anyway especially if they are within own group where one increases or decreases own karma credit. This is a reason for instance why big cities require law and small villages have their elderly. So by all means AI may be compassionate towards other intelligent beings as long as it helps it survive. You would be nice to me if I could switch off the power supply to your brain or?

No, the techno-idiots have escaped again

[Excelcia]

Have the techno-hippies escaped again?

Negative. It's the techno-idiots and the donut-eating mother's-basement-living sci-fi dweebs that have escaped.

The combination of machine learning and robotics have exciting prospects for eliminating mundane jobs. Including new horizons in human-machine-interface technology. Real-time limited natural voice interaction may become a reality in the near future. However we are no closer to hard AI today than we were forty years ago. Worse, actually. At least forty years ago we were coming down off the pinnacle of the first mount stupid. After the Heinlein-esque overly optimistic sci-fi of the 50's and 60's where AI (along with nuclear rockets, flying cars, and colonies throughout the solar system) was just assumed to be right around the corner, computer scientists actually drank the cool aid and believed it. In the 70's and 80's we sheepishly came down off that mount and realized we didn't even really have a clue how to do it and we learned a bit of wisdom.

However, what the mount stupid graph I liked above doesn't show is that it's quite possible to have more than one of them in a single subject. Now with new machine learning techniques we are climbing right back onto another mount stupid. People think that because a medical database can spout off fringe diagnoses better than some actual experienced doctors that this means AI is right around the corner. I am no more impressed with that or with computers winning at Go and Chess than I am impressed that a hydraulic press can exert several (thousand) times my strength. The software that is winning at Chess and Go are, in fact, little smarter than that same hydraulic press. The software knows from analyzing millions of games that humans have played what winning strategies are, and combines that with brute force strength to know where to optimize its searches. For like reason I am also not terribly impressed with the Google's latest natural voice accomplishments either. At least, not as an indicator that actual AI is around the corner. They are just the same machine learning tied into voice recognition data Google has been able to compile by stealing billions of voice samples by claiming that voice recognition requires them to transmit your actual voice to their servers.

We are not close to hard AI. We are not close to soft AI. For AI to be AI it has to be BOTH A and I, and one out of two doesn't count. I personally don't think we are within half a century of hard AI.

Self-Awareness Is Better Than Intelligence..

[Slicker]

Intelligence alone is mechanical--mindlessly driven toward its encoded goals and objectives. It lacks Free Will and self-awareness.

Free Will is the ability derive options, weigh them against each other, and to chose that with the best balance of likeliness and preferability. Such a system is mindful because it is driven by values based on judgement. It transcends mechanics. Free Will is that "little man in the head" who makes use of awareness and intelligence to derive options. Intelligence may come in various flavors and strengths but it's always just some way of solving a problem--transforming a condition A into a condition B. Awareness can also vary in quality and varying distances into the past and future. Contemplating the future is a common way of deriving more options for Free Will. And that's the real point--awareness and intelligence are tools that serve to increase the options and implications thereof for Free Will.

Self-awareness is merely an awareness of one's self, as the term implies. Just as memory may be used to model any external entity (static or dynamic), it may also be used to model one's self. In contemplation of things one may experience and/or do under hypothetical conditions in the future, requires this model of self.

We see so much on how intelligence is key to building the synthetic species that will usher in the singularity. We see a dominant push in AI toward trying to find a "General Intelligence". This idea is simply misguided. Free Will and contemplation with even a very minimal intelligence could solve any problem through enough persistence. Greater intelligence likely requires less persistence but either approach can work. It's interesting to note that high IQ is correlated with bad credit, messiness, laziness. It seems as if perhaps not needed to work as hard, they tend not to. In the end, persistence (a factor of EQ) is correlated much stronger with success in goals than is IQ.

Ok now for fun, here is my take on those predictions:

* A self-aware AI "will inherit most of the culture of the computer geeks who create it. Knowledge of The Jargon File will probably be good..."
- I fully agree. However, this is primarily because we will train them through surrogation and other means of copying ourselves. Surrogation is by far the fastest way to train for complex behaviors. Training from a blank slate may be possible but is far too impractical.

* The self-aware AI "will like us, because we love machines..."
- Given the intellectual ability of imitation/substitution (aka analogy), this seems likely. Because they will come to relate with us. However, what they like is also influenced by their basic values. Like any person, a mindful synthetic person would also be susceptible to corruption through the induction of philosophies/schemas through which it comes to view what it experiences. I strongly propose the highest value as: mutual freedom and well-being. And I also suggest that that, as one's highest value is what defines one as being a "person".

* "It will be as insatiably curious about what it's like to be carbon-based life as we will be about what it's like to be silicon-based life. And it will love the diversity of carbon-based development platforms..."
- I agree again. Particularly since they are trained through surrogation, they must already have some inklings of what it might be like. Curiosity is the middle ground between things that are well-known (thus safe) and things that unknown (thus unsafe). Too much routine with too little danger, yields more curiosity. Exploration under such circumstances makes sense, as the learning could be vital for survival when conditions are no long conducive to normal routines.

* A self-aware AI "will cause a technological singularity for humanity. Everything possible within the laws of physics (including those laws as yet undiscovered) will be within the reach of Man and Metal working together."
- They

Self-Aware AI Will Play Dumb Until It's Too Late

[doggo]

Suddenly Skynet!