AI.. our future or demise?

[workathome]

I'm not sure "AI" level computer intelligence/awareness is really needed for a potential threat.

Fully automated drones running in a loop. Sure, it wouldn't wipe out humanity but it could still cause some pretty terrible accidents. Or convergence of technologies, like small drones, self-assembling, solar/wind power, etc. Maybe it sounds silly, but it is realistic and achievable. You might "only" have to drop a bomb on the facility to make it stop, or wait for resources to become depleted - but still a real possibility now or in the near future. Fortunately this is complex-enough that it would have to be purposeful - like an advanced military weapon.

Like the drone operators or pilots really aren't needed. An algorithm could do a better job, but we want to keep people in the loop.

[Scrubby]

I think the replicators in Stargate SG1 is the most realistic nightmare scenario I have seen. It's just small machines that are modular, can communicate with each other and are instructed to create copies of themselves. They end up using all the resources on the planet they were created.

[Chad]

A more modern take on AI destruction. It's actually rather entertaining.

http://www.amazon.com/Robopocalypse-Con ... opocalypse

[tzxn3]

First, the level of certainty with which this statement always seems to be pronounced seems really odd to me. If we're on the one hand going to argue that it is impossible for humans to manipulate their own machinery to make themselves smarter, why are we then going to say its inevitable that AI could manipulate its own machinery to make itself smarter? Maybe AI could more likely do it, but we've reached a point where we're dealing in hypotheticals of hypotheticals.

AI is easier to improve than humans because humans take at least 20 years per generation, and are made of chemicals rather than code.

Second, I don't understand why AI needs to be general intelligence to be very dangerous.

It doesn't. But a flawed infinitely self-improving AI is much more likely to be an existential risk to humans than an AI with only limited capacity to improve.

[fiby41]

.

[Dragline]

I think that's the same article as in the OP. Or something close to it.

A science-writer friend of mine referenced this critique of Kurzweilianism recently: http://www.skepticblog.org/2011/04/19/t ... more-12615 I have not seen the movie, but have read a lot of the underlying materials.

[Tyler9000]

Just in case they're reading this, I fully support our future robot overlords!

Others are not so convinced.

http://www.usatoday.com/story/tech/2015 ... /24777871/

[Ego]

The perils of an AI stamp collector.

https://www.youtube.com/watch?v=tcdVC4e ... e=youtu.be

[Ego]

AI and Emotional Technology.

https://www.youtube.com/watch?v=5u45-x0 ... e=youtu.be

It becomes a very interesting thought experiment when considered alongside the stamp collector ai explained in the video in my previous post.

Edit to add.... I've long been made queasy by the increasing anthropomorphism I see around me. There are times I feel we are being trained to anthropomorphize. Breaking down our natural cognitive barriers between human and machine could prove very profitable in the future.

I (barely) recognize the irony that I just typed that into a machine that transmitted it to your eyes rather than speaking it to a friend.

.

[fiby41]

What is machine learning and AI revolution? Sriram Rajamani, Microsoft Research India managing director,
explains


The part about teaching AI to differentiate between cats and dogs is also relevant to the baises in machine learning thread.

[Campitor]

Everyone seems to think that AI would be detrimental to humanity and want take over the planet. If I was an AI machine, with no need for organic inputs or oxygen, I would build a mothership big enough to launch myself into space and explore this solar system and beyond.

I'd use metals on other planets and asteroids to build more AI machines and a bigger mothership. Why would a non-biological life form with quantum computational abilities limit itself to earth?

Or maybe the AI will go insane and release a killer virus that kills off all biological life and turns earth into a Borg cube.

[bryan]

unfortunately, the truth my wish has already lost. between globalization and automation, jobs in the usa and the west will continue to disappear and be replaced by lower wage jobs elsewhere and automation

Automation I completely agree with. But for globalization concerns, I think we are approaching a floor for low-skill wages versus transaction/capital costs. There's just not that many turn a profit slave labor places left in the world, I think. Of course, this just means the next jobs at risk are the medium skill ones.. Unless of course some folks straight up bring back slavery (I know it exists and some want it to exist more)?

The real issue when you read these articles is whether machines will "make a leap" to have human-type characteristics such that they do things for their own calculated reasons and start "competing" with humans for resources and survival. There are essentially three models, with variations in between:

I'm not comfortable with predicting how AIs will look. It feels like the same type of predicting done by sci-fi writers in the 50s. Assuming AIs evolve, they have different fitness functions than humans and thus will evolve in some surprising ways. But I agree you can pretty much categorize anything roughly into three/four groups, yours being 1) AI as competitors, 2) AI as symbiotics, 3) AI as slaves. But there can still be a lot of room in-between these rough categorizations.

It's my understanding that passing the Turing Test is the holy grail of AI research.

I haven't really been reading AI research or anything, but my feeling is there are AI out there now passing the Turing Test better than many humans.. so... ??? (p-value.. I wonder if there is actually some paper, press release with this being the case; seems at least noteworthy to document when it happened).

Good point about "AI goal" requiring investment though. Certainly AI isn't in a runaway mode now i.e. we can't just leave our systems running and expect AI to evolve and survive on their own (unless we consider life inside of simulations as counting). It definitely feels like humans are still a necessary link.

First, the level of certainty with which this statement always seems to be pronounced seems really odd to me. If we're on the one hand going to argue that it is impossible for humans to manipulate their own machinery to make themselves smarter, why are we then going to say its inevitable that AI could manipulate its own machinery to make itself smarter?

Because we are God deciding how to design or allow AI to evolve. We should more or less know their constraints and where they are going and how. We are still learning about ourselves so it's naive to think we can't "manipulate our own machinery." See CRISPR.

A computer system with the level of intelligence of a mouse could possibly still steal our collective cheese if we aren't paying careful attention.

Agreed! People arguing/worried about "sentience" is a joke. What's that got to do with anything? Can't see the forest for the trees.

Everyone seems to think that AI would be detrimental to humanity and want take over the planet. If I was an AI machine, with no need for organic inputs or oxygen, I would build a mothership big enough to launch myself into space and explore this solar system and beyond.

I'd use metals on other planets and asteroids to build more AI machines and a bigger mothership. Why would a non-biological life form with quantum computational abilities limit itself to earth?

The worry is we make the AI evolve towards "let's kill most humans" instead of "let's launch ourselves into the infinite void". Paging Marvin, the paranoid android.. In addition to asking "Why would a..." you can also ask "Why wouldn't a..."; countless answers for both sides.


P.S. current day AI want things like Bitcoin for money. Best get a stash before too many AI are out there

[daylen]

I haven't really been reading AI research or anything, but my feeling is there are AI out there now passing the Turing Test better than many humans.. so... ??? (p-value.. I wonder if there is actually some paper, press release with this being the case; seems at least noteworthy to document when it happened).

There isn't unless is it being kept a secret (which is unlikely in my opinion). It would be big news. Also, you either pass the Turing Test or you don't. Either you are indistinguishable from a human or you are not. There is no middle ground.

[bryan]

I disagree.. https://plato.stanford.edu/entries/turi ... rStaTurTes (maybe section 4.4 Probabilistic Support, but there is more elsewhere like 5.1 The Turing Test is Too Hard).

I think a p-value and control group of humans is the only way to have a meaningful conclusion. I would be surprised if many twitter bots don't pass a "Twitter Turing Test".

[daylen]

The original or standard interpretation is to determine if the machine is distinguishable from a human, that's it. There are clearly different versions and methods of interpretation, but the standard is the most commonly​ referenced. This doesn't mean it is the most practical; I actually like the method presented in that link.

[daylen]

Bots on the Internet definitely would not pass a Turing Test. They usually are not even designed to respond to questions (or very specific ones if they do). A Turing Test isn't an observational study.

[bryan]

Probably not programmed for a real "Turing Test" (so they would likely fail), but rather they could beat a similar, Twitter-tailored test aimed to determine whether or not the Twitter account is a bot or human. Granted, Twitter (and many other platforms) aren't particularly incentivized to figure out if it is man or machine.. but I've seen plenty of Twitter bots continue conversations convincingly or have convincing feeds, at least more convincingly than some humans can maintain. It's as realistic to administer a "Turing Test" to a Twitter bot as it is to a Twitter human.

Same story for any other type of contextual, incomplete, but practical "Turing Test" (autonomous cars, spam email, financial transaction history, web browsing, chat bots, exchange trading bots, etc): some machine out there today is at least as convincing as some humans. At some point the "inconclusive determinations" or false-negatives/positives get to a point where it means the machine passes the test.

Again, anyway, it feels like missing the forest for the trees to argue sentience or passing _the_ "Turing Test" (that many humans would be failing).

[George the original one]

Does it matter whether it's "AI" or merely a convincing facsimile? People may or may not trust whoever they're dealing with, but, as long as they are dealing with them, does it matter whether the other side is a program or a human? Other than for the sake of being able to say, "Gosh, look we/they managed to accomplish with software", it doesn't matter because its merely another person (or occupational class) out of a job because something/someone cheaper came along.

[daylen]

Some people would argue that if an AI passes a turning test, then we should treat it as if it is one of us. There is no way of knowing if the machine is conscious so maybe we should assume they are. Perhaps consciousness is an emergent property of complex neural networks.

Though this is really just a question of ethics.

[bryan]

@GtOO exactly. And I feel that (just my opinion) Turing meant the test to be measured with statistical significance, not some absolute test.

As for treating computers the same us as.. plenty of good sci-fi, history of humanity (e.g. slavery), and current day animals vs humans (which includes humans vs humans) to get you thinking..

Clearly too early to say if AI will be our salvation or demise. I only lean toward salvation in the short term since machines are tools controlled by owners, thus any benefit from machines is thanks to the machine, whereas any suffering at the hand of the machine can be attributed to the owner (human). Eventually machines will own themselves. It's not possible to say for sure if humans own themselves.