Elon Musk + AI + Microsoft = Awesome Dota 2 Player (theverge.com)

An anonymous reader quotes the Verge: Tonight during Valve's yearly Dota 2 tournament, a surprise segment introduced what could be the best new player in the world -- a bot from Elon Musk-backed startup OpenAI. Engineers from the nonprofit say the bot learned enough to beat Dota 2 pros in just two weeks of real-time learning, though in that training period they say it amassed "lifetimes" of experience, likely using a neural network judging by the company's prior efforts. Musk is hailing the achievement as the first time artificial intelligence has been able to beat pros in competitive e-sports... Elon Musk founded OpenAI as a nonprofit venture to prevent AI from destroying the world -- something Musk has been beating the drum about for years.
"Nobody likes being regulated," Musk wrote on Twitter Friday, "but everything (cars, planes, food, drugs, etc) that's a danger to the public is regulated. AI should be too."

Musk also thanked Microsoft on Twitter "for use of their Azure cloud computing platform. This required massive processing power."

Comments

Not AI

[110010001000]

Let me repeat: computers playing games is NOT AI. Computers love games. Games have strict rules and limited parameters. Computers love that. Computers excel at that. IT IS NOT AI.

Re:Not AI

[Dutch Gun]

As game developers, we call our automated agents "AI" in a long tradition of overloading and bastardizing words from other fields, but we all understand it's not actually real "AI" of any sort. I mean, even pathfinding goes under the term "AI" for our purposes. So, yeah, this is deep learning, but no more "AI" than what we do inside the games. Very often, we actually have to work to make our opponents *less* effective, because computers have so many advantages over players, especially in any game at all where reflexes count, or broad analysis of lots of details is important.

But more to the point, Elon keeps talking about regulating AI to prevent it from destroying the world. Every time he talks about this, he sounds like an unhinged lunatic that has some irrational fears about something he doesn't deeply understand. I still haven't heard a realistic scenario about how AI is going to go about doing this. And let's be honest... the perception of his capacity for rational thought on matters outside his domain of expertise was NOT helped by his declaration that he's 99.9% certain we're living in a computer simulation.

Besides which, how exactly would one "regulate" this, short of simply banning AI development by private enterprises? Massive governmental oversight requiring a programmer to pinky swear or sign in blood that they'll use those neural networks for good instead of evil? I honestly don't get it.

Re:Not AI

[K. S. Kyosuke]

It's no "weak AI". It's just AI. AI people can't be blamed for other people constantly moving goalposts ever since the thing was conceptualized half a century ago.

Re:Not AI

[Anonymous Coward]

Just because it's not strong AI doesn't mean it's not AI. And really, this new wave of machine learning is quite a bit more impressive than the "AI" of the 90s that conquered chess. There things were hard coded by humans and possible moves iterated. With this new machine learning AI is "learning" the parameters and rules without human input, just through reinforcement. This is why it is now being used for things like the game Go where iteration is essentially impossible.

Re:Not AI

[Kjella]

As game developers, we call our automated agents "AI" in a long tradition of overloading and bastardizing words from other fields, but we all understand it's not actually real "AI" of any sort. I mean, even pathfinding goes under the term "AI" for our purposes. So, yeah, this is deep learning, but no more "AI" than what we do inside the games. Very often, we actually have to work to make our opponents *less* effective, because computers have so many advantages over players, especially in any game at all where reflexes count, or broad analysis of lots of details is important.

So to flip the board, is it fair to hamstring the AI because humans can't keep up? If we're not making a game for entertainment here, if the computer's AI drone pilots can decimate your fleet of human pilots why shouldn't it play to win? I don't mean to take the human factor out of it, but doesn't war often come down to arms and numbers? I doubt there was any true difference in the level of motivation for the Axis and the Allies, it was a war for survival. So from the computer's POV it's just playing an advanced game of Civilization, build the economy, build the military, conquer and win. Maybe you can score the occasional win on creativity but if you have more troops, better equipped troops, better supply lines, better positioned troops like pieces on a chess board you'll grind down the opposition. Though to be honest I don't see an AI on top, but even a small group of AI-assisted leaders can direct millions of people. If you're leading China's armed forces I doubt you see the actual people as such.

Re:Not AI

[Maxo-Texas]

Decomposed to it's components, the brain is not intelligent.

The Brain is composed of the forebrain, midbrain, and hindbrain.
Alone, none of these are intelligent.

The forebrain consists of the cerebrum, thalamus, and hypothalamus (part of the limbic system).
None of these these three subcomponents are intelligent.

The midbrain consists of the tectum and tegmentum.
Neither of these pieces are intelligent.

The hindbrain is made of the cerebellum, pons and medulla.
None of these these three subcomponents are intelligent.

People with dysfunctional Amygdala are incapable of logical thinking because they can't emotionally weight the factors correctly.. They find snakes interesting but not scary (so they want to touch them).

Most people driving are not doing so intelligently. A minor "driving" expert system is running while their brain is elsewhere.

Computers beating humans at go is weak A.I.

We are 8 years ahead of the projected schedule for landmarks. It's going faster than we expected.

----

We *must* be very careful with A.I. because we only get one shot at it. Any serious A.I. research must be done air gapped, with analog power meters, with fuse limited power suplies, and many other precautions.

We have overconfident people playing with extinction level technology. It might never click. It might click 10 years from now. It might click tomorrow.

Re:Not AI

[mlyle]

When we talk about strong AI, there's plenty of scary scenarios in the (not-near, but unclear how distant) future.

https://wiki.lesswrong.com/wik...

Re:Not AI

[Whibla]

Wish I had mod points left!

Having said that however, what amazes me about the entire subject is what no-one mentions: Responsibility.

I'm not talking about responsible development of AI, I'm talking about our responsibilities to any true AI we create. If we have a child we, as parents, have a responsibility to feed it, nurture it, educate it, socialise it, and so on, at least until it becomes an adult in its own right (How we might decide this, for a new form of life is subject for another discussion). Surely we would have the same responsibilities to any other truly intelligent entity we 'give birth' to? And with that responsibility would come the one that says we can no longer, on a whim, just decide to turn it off.

To be honest, until mankind steps up and is willing to assume responsibility I'm not sure we deserve these 'children', and if we blindly push ahead regardless I can't help but wonder if we don't deserve everything we get.*

*Okay, so this is hyperbole! Damning the majority of the human race because of the ignorance and 'sociopathy' of the few is not something I advocate, but sometimes I do despair...

With all this mindless AI hype...

[gweihir]

...I am beginning to ask myself, whether weak AI like this (no actual intelligence or understanding) may not actually be on-par with many humans, which fare not much better at understanding things.

Everything that's dangerous should be regulated..

[h33t l4x0r]

Ok, Elon. How about if we start with "don't teach your AI that it's primary objective is to destroy every other creature on the map".

Re:results

[Whibla]

can it answer the question which heroes are over- or underpowered?

Assumption detected... ;-)

In a game of rock paper scissors which hand sign is overpowered? It seems to me that it's this balance of power that the game devs are aiming for. Which then leads back to your question ... and, in fairness, your final point.

It was obvious that e-sports will be short-lived because bots are going to beat us all within a short time

Only if they're allowed to enter. Otherwise, sanitised computers with only the required software (i.e. the game in question) installed will ensure the competition is limited to the human participants you can see on the gaming stage. Online gaming, on the other hand, is already exactly as you say. The various cheats and hacks that already exist make playing many online "pvp" games an exercise in frustration and futility.

Rent seeking?

[bool2]

I think the question must always be asked when people call for regulation is can they profit from it? I can't help but wonder if Musk is actually not doing good here but setting in motion the slow train of regulation so that when he's ready with some uber-AI he'll be in the position to get the rules of the game altered to his advantage with bureaucracy, regulation and licensing.