Can AI Games Create Super-Intelligent Humans?

destinyland writes "A technology CEO sees game artificial intelligence as the key to a revolution in education, predicting a synergy where games create smarter humans who then create smarter games. Citing lessons drawn from Neal Stephenson's The Diamond Age, Alex Peake, founder of Primer Labs, sees the possibility of a self-fueling feedback loop which creates 'a Moore's law for artificial intelligence,' with accelerating returns ultimately generating the best possible education outcomes. 'What the computer taught me was that there was real muggle magic ...' writes Peake, adding 'Once we begin relying on AI mentors for our children and we get those mentors increasing in sophistication at an exponential rate, we're dipping our toe into symbiosis between humans and the AI that shape them.'"

Comments

Have you not seen

[Glarimore]

Terminator? Or the Matrix?

Re:Have you not seen

[bky1701]

You do realize those aren't documentaries, right? Sometimes I wonder if slashdot forgets that.

Re:Have you not seen

[wagnerrp]

You have to remember two things:

1. 1. Of all the colleges at a university, the teaching college will generally have the lowest, or near the lowest, admissions requirements. Low pay just doesn't draw the high quality talent. Now sure, you'll find some absolutely stellar teachers, ones that actually care about their students, and spend lots of time outside of school researching the stuff they're teaching, building lesson plans, projects, field trips. You'll find a lot more who are just teaching straight out of the text book. I could outwit at least half my grade school teachers.
2. 2. We are in school of some form or another for a good chunk of our lives. A couple years of daycare. Another decade of elementary and high school. From there, a few years of vocational, or several years of college, or up to another decade for higher level degrees. For 20 years of care, we only get another 40-50 of functional lifetime out of a person. We simply can't afford as a society to have a low student/teacher count. AI could fill the gaps for the less demanding tasks. An AI could guide individual students through directed self study, and aid them in homework, allowing a teacher to assign more work and still expect it be accomplished. An AI could handle larger lectures, allowing teachers to focus one-on-one, or with small groups.

AI in schools would allow the teachers we had to operate more efficiently and more effectively. That in turn means fewer teachers per student, increasing individual teacher pay, and drawing in a better quality of teacher. Think of it as the same thing that has happened in manufacturing for the last 200 years. Machines don't replace humans all together. They simply fulfill the more repetitive tasks.

Re:Have you not seen

[__aayuzx6098]

Intelligent tutoring systems in education is my field, so I say with some confidence that so-called AI won't replace human tutors anytime soon. Online workbooks and computer-aided learning are a wonderful adjunct to classroom instruction, but cannot replace a live teacher. About 30% of instruction can be reasonably handled remotely (software- or video-based instruction), but the other 70% of the task of educating and motivating learners is non-trivial. File the OP under jet-cars of the future.

Re:Have you not seen

[mcgrew]

Before we create real intelligence we're going to have to understand what sentience is and how it works. People seem to forget the second part of science fiction is fiction. It's not only possible to write a program that will fool people into thinking it really can think, I've done it myself. What's more it was back in 1983 on a TS-1000 -- Z80 processor with 16K memory and no other storage (program loaded from tape).

The irony is I wrote the thing to demonstrate that machines can't think, and nobody believed me when I told them it was simple trickery like a stage magician uses. The magician doesn't really saw the woman in half, and the computer doesn't really think. It's a trick, nothing more.

The only science fiction I've read that got this right was Dune; the thinking machines that enslaved humans were controlled by other humans. Herbert didn't actually come out and say the machines didn't really think, but the implication was there.

If you don't know how a NAND gate works, you're not really qualified to even discuss it. We will likely (IMO almost inevitably) have replicants that actually think, but replicants are biological.

How many beads do I have to string on an abacus before it becomes sentient? A computer is simply an abacus with billions of beads. The danger to this "thinking machine" nonsense is that your grandkids' generation will have PETA-type nuts lobbying for machine rights. Anthropomorphism can be a dangerous thing.

Re:Have you not seen

[somersault]

If you assume that "intelligence" means "thinks just like a human" then sure.

There's lots of stuff "like" AI. In fact there's plenty of actual AI out there that works well in the domain that it was designed for.

Projects like Watson are really cool though, and heading in the right direction for building machines that can process a wide type of information in an intelligent manner, and respond to questions about that information and the links between it. Watson isn't really designed to teach (that I know of), but I would think of it as a form of intelligent encyclopaedia, which would be useful as a teaching aid.

If the guys at IBM improved Watson to be able to create lectures and its own quizzes (guided by a syllabus with the main points that students should be learning), it would be well on the way to being a personal tutor.

If they further improved Watson to be able to ask its own questions, or at least take in new information from sources outside of the original quiz show database (and not just blindly accept all information as "truth" of course, there would have to be heuristics to see how well the info fits in with what Watson already "believes", or at least some way of separating out facts from fictional ideas, if it doesn't already do that), it could actually be fun, and perhaps even insightful to talk to. Just don't let it read any YouTube comments.

No

[WrongSizeGlass]

Can AI Games Create Super-Intelligent Humans?

If all the universities, colleges, think tanks, etc can't produce super-intellegent humans then what makes them think we'll be able to produce AI that can?

Re:No

[Thing 1]

"X(s) can't produce Y, and someone else thinks Z can produce Y?" You fail logic.

Re:No

[WrongSizeGlass]

"X(s) can't produce Y, and someone else thinks Z can produce Y?" You fail logic.

Um, no. Basically, he's talking about a 'perpetual intelligence machine' (which I'm sure violates one of the laws of thermodynamics) fueled by the educational system (which is running out of money). This is the same system that is demonizing teachers as greedy, unqualified babysitters. As we chase the good teachers out of the education system we're going to try to use AI to create 'super-intellegent humans'? We're going to be lucky if the next generation of children learn anything not on a standardized test.

The current generation of school kids is going to be the ones going on to college to create this super-duper AI? With the price of higher education going through the roof, and the interest rates on student loans giving the loan shark on the corner a run for his money, we're going to see a drop in college graduates. I suppose anything is possible, but I think we have a better chance of seeing affordable college education before we see AI creating 'super-intellegent humans'. At the rate we're going I think we should be striving for 'educated humans'.

Re:No

[ShakaUVM]

>>The Creationists, ID-ists and the slew of others nutjobs all having their pound of flesh taught in the US school system seems to show that it certainly isn't simply a matter of getting the right teaching methods.

Yes, like in Creationist Texas that just voted 8 to 0 to reject Evolution! Oh, wait. It was 8 to 0 to support Evolution and reject ID.

Your paranoid hysteria is a bit overblown if IDers can't even get one vote in *Texas*. You're probably one of those folks that confused the proposals for changes to the history standards with actual changes.

While I'd agree that a slew of nujobs have their say in education, it's more the people who invent new teaching methodologies every year, and then force them on teachers, not your fantasy about the all-powerful Koch brothers rewriting textbooks.

Education is screwed up for a lot of reasons, but that's not one of them.

Re:No

[LXPK]

I'm Alex Peake the author of the article and your post is unintentionally inspiring, I look forward to the day when I can send my kids to Logical Preschool, although transitive properties are usually something teens learn in high school.

Re:No

[__aayuzx6098]

Regards your .sig: All due respect, but science does *not* encompass the mystical ("Wovon man nicht sprechen kann, darüber muss man schweigen." -- L. Wittgenstein); rather the converse. Science and empirical method represents only a very tiny, self-referential fraction of what is intuited about the universe. Objectivity is more of a myth than Flying Spaghetti monsters (see Critical Theory; Post-modernism).

Re:No

[LXPK]

Thank you! I've been lurking on Slashdot all these many years and never acquired much karma. In the 90s I began the original Primer codebase as a branch of Slashcode and learned Perl as my first web language because I wanted to build on the early innovations in metamoderation that were so revolutionary at the time. It is an honor to be on Slashdot now, and an honor to be awarded one internet.

Re:No

[silentcoder]

You'd think geeks would understand basic physics better than this. It was okay when Asimov got thermodynamics just plain wrong - because it was 60 years ago and everybody had it wrong. Even Roger Penrose still had it wrong in the 70's but the whole "universe increases in entropy so why are there constellations and life" paradox doesn't exist.

Real scientists figured that out a long, long time ago. The longer version is: thermodynamics is a model of the behavior of gasses in a closed system which makes a lot of assumptions (for starters: it treats all gas molecules as solid, perfect spheres that bounce off once another with no loss of velocity - e.g. energy transfer without friction). These assumptions are okay to make in thermodynamics because the energy "lost" to friction doesn't get "lost" from the system - it becomes heat energy and it all averages out. But they do mean something. Thermodynamics tells you if you dump a bunch of gas in a tank it will end up evenly spread rather than clumped in the corner (that is the ENTIRETY of the entropy law - and information entropy has NOTHING to do with it and the "relationship" is not even conjecture - it's BAD science fiction - for starters it goes in the OPPOSITE DIRECTION and the symbols in the equations have different meanings [c is a distance in geometry and a velocity in physics for example]). That is the purpose of thermodynamics, it's a model of physics that explains the behavior of gasses (and with minor adaptation liquids and solids) from a certain point of view, but it's a model of reality that ignores a bunch of inconvenient things because within the scale of the model those things have no measureable impact. Scientists call it "coarse-graining" - the details we don't see in thermodynamics because they are too small do matter on the scale of a star system.
The most important detail that thermodynamics completely ignores is the force of gravity. There is no thermodynamic law that considers the impact of gravity. In a tank containing only one gass - it actually means that on average the gas will be slightly more concentrated on the bottom of the tank than the top - not enough to matter, so the model does it's job - but thats because individual gas molecules have very little mass, on astronomic scales the whole tank has almost no mass.
But star systems and planets have huge amounts of mass - and suddenly a force which on the nuclear level is incredibly weak (a small speaker magnet has more than enough force to lift a nail against the gravity of the entire planet) starts to matter: a lot.
Gravity on the universal force acts to collect things together, and the more they collect, the more gravity they have - it is a trend in the opposite direction of entropy.
On small scales (what thermodynamics is for) entropy wins. On the scale of galaxies, entropy is (much) weaker than gravity and the entropic "heat death of the universe" is unscientific hogwash.

In the real universe the battle between entropy and gravity is pretty close. That's why we don't have a "smoothly spread" universe now. If gravity was too strong - we'd have the opposite: the entire universe collapsed into a single black-hole. The ultimate singularity.

In the end, one or the other may yet win out - but the odds of a black-hole ending is at least as good as a for an entropic heat death and either way for either to actually win when the difference is so small - would take a lot longer than the lifetime of the universe so far still - meaning we got at least a few hundred billion years left.

Re:No

[LordLimecat]

fueled by the educational system (which is running out of money).

If you look at nea.org info (nea.org PDF), you can see a number of interesting things.

First, that many "states" that rank quite high on "expenditure per pupil" (page 55)-- DC for example, which is #1-- do not coorelate to better education. In fact, DC is the top spender, and you will find MANY lamenting how bad schools are there.

Second, the total revenue of schools (page 68) has RISEN significantly over the last 10 years. Crying about constantly running out of money as you get more and more each year is perhaps an indicator that the money needs to be used more widely.

Third, if you look at "performance per dollar spent per pupil" (here) (2006), you can see that there isnt much of a correlation. NJ, NY, and DC are the top spenders, and are at the absolute bottom of the barrel in student SAT scores. Conversely, the top 15 or so scoring states all spend roughly 50% of what NJ, NY, and DC spend.

If you need another example, perhaps this article would be enlightening.

This is the same system that is demonizing teachers as greedy, unqualified babysitters

Perhaps the problem IS those teachers which are greedy, and unqualified, and have some ridiculous politically driven objection to merit-based pay (ie, if your 9th grade students consistently get awful grades in later years or dont go to college or get rejected from college, perhaps you suck at preparing them for later years), as well as ridiculous objections to vouchers (not really clear what possible grounds there are for objecting to them).

If your teachers are requesting ever more money, and yet we can see that there isnt really a correlation between "more money" and "better results"; and they also refuse to be evaluated in any meaningful way on how well they do their job-- and it absolutely CAN be done without people teaching to the test (which, dont get me wrong, I hate as much as the next person)-- forgive me if I think that the problem MIGHT not be the funding.

aperture education

[Anonymous Coward]

"children. Keep calm and continue testing."
"At the end there will be cake."

Well that's a new record

[JoshuaZ]

This is one of the silliest versions of a Singularity I've seen yet, and there are already a lot of contenders. This has a lot of the common buzzwords and patterns (like a weakly substantiated claim of exponential growth). It is interesting in that this does superficially share some similarity with how we might improve our intelligence in the future. The issue of recursive self-improvement where each improvement leads to more improvement is not by itself ridiculous. Thus, for example humans might genetically engineer smarter humans who then engineer smarter humans and so on A more worrisome possibility is that an AI that doesn't share goals with humans might bootstrap itself by steadily improving itself to the point where it can easily out-think us. This scenario seems unlikely, but there are some very smart people who take that situation seriously.

The idea contained in this post is however irrecoverably ridiculous. The games which succeed aren't the games that make people smarter and challenge us more. They are the games that most efficiently exploit human reward and mechanisms and associated social feelings. Games that succeed are games like World of Warcraft and Farmville not games that involve human intelligence in any substantial fashion. The only games that do that are games that teach little kids to add or multiply or factor, and they never succeed well because kids quickly grow bored of them. The games of the future will not be games that make us smarter. The games of the future will be the games which get us to compulsively click more.

Re:Well that's a new record

[Bing Tsher E]

This is one of the silliest versions of a Singularity I've seen yet, and there are already a lot of contenders.

I just had to stare at the original post in wonderment.
The whole 'self-fueling feedback loop which creates 'a Moore's law for artificial intelligence,' with accelerating returns'

Da-woop-dee-woop-de-woo.

An AI generated that big clump of meaningless drivel and buzzwords, didn't it?

Or has Minsky broken into the liquor cabinet again?

Minsky!!!??!!?

Citing lessons drawn from Neal Stephenson's The Di

[SpectreBlofeld]

I don't think citing a work of fiction to support your thesis about video games will get you taken very seriously,

Re:Citing lessons drawn from Neal Stephenson's The

[WrongSizeGlass]

I don't think citing a work of fiction to support your thesis about video games will get you taken very seriously,

Not mention his reference to 'muggle magic'.

Artificial Intelligence 2012

[bitbucketeer]

The gold farming bot that can pay off a $14.8 trillion debt has my vote!

Re:Why do we need AI

[blue trane]

The answer is not to throttle technology; the answer is to understand that money creation is a technology in itself, and should be democratically controlled instead of the exclusive right of private individuals. The recent story about the Fed creating $16 trillion shows that govt could easily create enough money to provide a basic income to everyone, so that we can each explore the natural wonder and creativity that we are born with, using tools such as AI to expand knowledge ever-greater bounds...

I Assume He Means Serious Games

[mentil]

I think he's referring to 'serious games', not standard entertainment-focused video games. Imagine a simulation where you interact with an AI in different scenarios. The AI's actions and responses to the user can be standardized and tweaked to ensure that the child playing the game learns the intended lesson/skill. This could be especially useful in teaching children social interactions, where how another human responds is unpredictable, even if they've been trained beforehand.

The 800 pound gorilla is that we're going to live in a Star Trek future with strong AI and a pure robot economy before parents leave child-rearing to AI simulations, so the 'exponential increase of intelligence' isn't going to come from this; genetic engineering or self-designing AIs are much more plausible for a trigger of a singularity.

Re:I Assume He Means Serious Games

[bky1701]

"The 800 pound gorilla is that we're going to live in a Star Trek future with strong AI and a pure robot economy before parents leave child-rearing to AI simulations."

You can't be serious. Do you have any idea how many parents use video games as their babysitter? There is no "before" here, it is already here. I'm not so sure it is a bad thing on the whole, either.

Re:Citing lessons drawn from Neal Stephenson's The

[Bieeanda]

He writes tortured metaphors about katana-wielding Mafia pizza delivery men, and pulls endings out of his ass. Referencing mathematicians and writing novels that appeal to backpatting nerds doesn't make him a genius, it just makes him aware of his audience.

Hmmm...

[modecx]

I think this is more an example of Lawnmower Man.

Those singularity guy always miss the obvious

[aepervius]

Growth curve are almost never infinite in real life. They almost always slow the growth before reaching a limit, then become semi flat never reaching the limit.

Smart as we are already

[Vecanti]

No. Kids playing with AI that is as smart as a humans will not make them smarter than if they were kids playing with people that "are" actually human. We've been doing that for a while now. ;)

Re:Intelligent Tutoring Systems

[TaoPhoenix]

"Sufficiently advanced educational processes verge on terrorism". (Hi Mods! Note the quotes which means it's rhetorical!)
We already have this game.

A "bunch of script kiddies", er, Students, have been beating various professional IT departments at the game called "Cyber Security". Since two years ago we would have called anyone who said they could bust federal contractors a "tin foil hat", they took some bits as prisoners to prove it. This then caused Memos to be Issued to block those security holes. The Students then observed the results, and then took NATO for a ride in Round 2. This caused more Memos to be Issued by the "AI". (Insert rest of article here.)

Oh wait, you're saying that's not a game? Games are supposed to be cute little self contained exercises that *don't matter* right?

Right. Gotcha. Uh huh.

'a Moore's law for artificial intelligence'

[makubesu]

Pardon me for a second.
AHAHAHAHAHAHAHAHAHAHA
Thanks. I needed that. What a ridiculous statement. AI is a hard problem. Just look at the history of the field. People once were optimistic about it, they solved the toy problems, and thought that skynet was on its way. But when you start to expand the scope of the problems, all your traditional techniques fall apart. To get to where we are today has been a long grind, with increasingly sophisticated mathematics being used to make any advances. Moore's law for processing power has been the opposite. Yes people have had to work hard to make it happen, but it was a manageable problem. They comparison is ridiculous.