Elon Musk Open Sources New 'AI Gym' (csmonitor.com)

An anonymous reader writes: OpenAI, a billion-dollar research non-profit backed by Elon Musk and other Silicon Valley executives, just released a public beta of a new Open Source gym for computer programmers working on artificial intelligence. "Nothing beats a competitive environment to motivate developers," says Patrick Moorhead, an analyst at Moor Insights & Strategy. "It's like a monster truck rally for AI programmers."

The gym lets developers run tests in a standardized environment and share their results, and was built by OpenAI to develop algorithms for the non-profit's own research, according to the Christian Science Monitor. "The gym's exercises range from robot simulations to Atari games and are designed to develop reinforcement learning, the type of computer skills needed for motor control, and decision-making. 'Long-term, we want this curation to be a community effort rather than something owned by us,' Greg Brockman and John Schulman wrote in an OpenAI blog post. 'We'll necessarily have to figure out the details over time, and we'd would love your help in doing so.'"

Comments

Bollocks.

[Anonymous Coward]

Nothing beats a competitive environment to motivate developers.

Capitalism has become so ingrained like a religion in the psyche that we assume it reflects nature.

Humans are sophisticated social species. The norm is to get together and to cooperate - we fight as a last resort. Our current system has put us in a constant state of last-resort thinking. Sure, we're good at it, but only because that's the final option we have been evolved to deal with - everything beats a competitive environment.

Re:Bollocks.

[Anonymous Coward]

1) I didn't say that competition was always bad - I said it was a last resort;

2) Supermarkets tend only to bring prices down at the wholesale level, but in a mature market will match each other at retail. This couldn't be more obvious in the UK, where e.g. milk producers end up tied into contracts where they're selling milk at a loss a lot of the time;

3) I don't drive since it seems an extremely inefficient way of getting around. I know that the US is unlike Europe in that, again, people take the competitive approach of private transport rather than the cooperative approach of cheap, clean, reliable public transport.

In the long run, businesses consolidate and stagnate, i.e. you end up with a cooperative environment anyway, except that the cooperation is for profit. The only place where competition appears effective is in an immature market, where most of the players haven't sold out yet.

Re: Bollocks.

[Anonymous Coward]

IF only my fellow Americans would learn the poor aren't that dangerous. I would describe most of my interactions with them on public transportation as pleasantly amusing. Of course most of the time I'm minding my own business while retaining my sanity and saving hoards of cash by not piloting myself to my destination.

Re:Bollocks.

[shawn2772]

In the long run, businesses consolidate and stagnate, i.e. you end up with a cooperative environment anyway, except that the cooperation is for profit. The only place where competition appears effective is in an immature market, where most of the players haven't sold out yet.

You've made the same fundamental error here that Karl Marx did in his formulation of economic theory: that production systems do reach maturity, where change ceases and competition therefore no longer matters[1]. In fact, this theory of non-advancement actually does describe reality at times, but only when other forces have arranged to squash competition. As long as competition is present, markets do not stagnate because the competitors continue innovating, looking for an edge to make their products more valuable or their production costs lower -- resulting ultimately in better goods at lower prices as competition forces all participants to adopt each new improvement. I don't believe there is any good in the marketplace that has ever reached perfection in both design and production process, and I'm skeptical that any good ever could. (I'd be very interested in counterexamples if you can think of any.)

You also mentioned consolidation. Yes, over time there is a tendency towards consolidation due to economies of scale, but even very mature markets tend to consolidate down to two or three players, never only one. When you see a marketplace reduced to a single player, it's almost certain that there is some extra-market force involved, because without that extra-market force there's a powerful incentive for capital to create a competitor in order to exploit potential innovations that a monopoly doesn't have a reason to make.[2]

Yes, cooperation is a core human characteristic, but so is competition. All the way back to pre-historic tribalism, people banded together in cooperative groups, then competed with other groups. And even within tribes, there has always been competition for status, and that competition has always included the formation of smaller cooperative groups to improve their ability to compete. At root, human cooperation is nothing more than a (very effective) competitive strategy.

Further, looking beyond humanity and social structures competition is the fundamental mechanism by which all knowledge is created. More precisely, variation and selection is the basic mechanism of creation, whether you're looking at the evolution of organisms, the construction of social structures, even the development of ideas. Selection, of course, is inherently competitive.

Competition is deeply ingrained in humanity, because it's fundamental to all creative activities and processes... including the processes that produced humanity.

[1] Marx didn't state this assumption explicitly, but it underlies all his analyses of economic relationships, which he viewed as static and based solely on allocation of labor with little notion of the value of knowledge, and no consideration for how applied or created knowledge could change the structure.

[2] I'm skeptical that government anti-trust efforts are either necessary or even useful. If you look at the history of government anti-trust efforts what you find is a repeated pattern of government stepping in with relatively ineffective measures, followed by the destruction of the monopoly by an innovation-driven market reorganization that which would have happened regardless. I'm not saying long-term self-sustaining monopolies (other than governments themselves) couldn't exist, but I don't think we've ever seen one.

Re:Bollocks.

[kaiser423]

Nothing beats a competitive environment to motivate developers.

...The norm is to get together and to cooperate - we fight as a last resort. Our current system has put us in a constant state of last-resort thinking...

It's interesting that you think that competition is the same as fighting. It's not.

Re:Bollocks.

[werepants]

[2] I'm skeptical that government anti-trust efforts are either necessary or even useful. If you look at the history of government anti-trust efforts what you find is a repeated pattern of government stepping in with relatively ineffective measures, followed by the destruction of the monopoly by an innovation-driven market reorganization that which would have happened regardless.

Over and over this topic comes up, and over and over I wonder how in the hell people are so unaware of THE defining example of a government regulated monopoly. That being Ma Bell and Bell Labs, of course. The combo of immense corporate power, focus, and resources, combined with government scrutiny that required them to limit their profits meant that they dumped most excess cash into Bell Labs, which did the lion's share of the fundamental research work that created the information age. Seriously, it's a long list: the transistor, UNIX, cellular networks, information theory for starters... They had all that excess cash from being a government-granted monopoly, but also due to that reason they had to constantly demonstrate that the existence of their monopoly was good for the public. So they dumped the money into research, developed all this revolutionary technology and licensed that tech out cheaply. Bell Labs more than any other single entity deserves to be credited with developing the enabling fundamental technology that's changed the world in the past few decades.

So all this nonsense dogmatism that monopolies are inefficient or that government regulation strangles innovation needs to die - it's the exact opposite of reality. In honesty, we should look at a combination of corporate efficiency and government scrutiny as the model to strive for. This isn't to say that monopolies (or government regulation) are always helpful when they are allowed to exist unchecked, but to say that our single best example in recent history of a total monopoly is also on of our best examples of effective government regulation, and happens to be the best instance of innovation and technological development in modern history.

Re:Bollocks.

[shawn2772]

Bell Labs did some great work, sure. But there's no reason to expect that any other organization with tons of excess cash wouldn't do as well, or that any other monopoly-supported organization would have, or for that matter some directly government-funded research lab (because that's exactly what Bell Labs was, in effect, a research lab funded by "tax" revenues).

Bell Labs was one confluence of circumstances which happened to work out fairly well. We think. It's not evidence of anything except that sometimes when you throw lots of money at smart people you get good stuff out.

pick an analogy

[Anonymous Coward]

is it a gym, or a monster truck rally? nevermind I don't care.

Re:Christian Science Monitor

[Anonymous Coward]

Despite its name, Christian Science Monitor is actually one of the best news outfits around. Its quality is astoundingly good, its bias is minimal, and it puts most other mainstream news outlets to shame.

Re:So all the future "AI"

[LifesABeach]

I always thought Gyms were 3D in nature. Newtonian mechanics requires this. Maybe Elon can use something like Blender3D as a Gym?

Oh no

[m76]

Nothing like a competitive environment to do half assed uninspired work with major corner cuts.

Got the Singularity Badge

[ssssssssssssssssssss]

Now AIs up to level 50 will obey you.

Wait, what

[Actually, I do RTFA]

Nothing beats a competitive environment to motivate developers

Wait, really? Because I see when people try to motivate me with a "competitive environment" they're not motivating me with, say money. Which is fine for academics (the Underhanded C guy), but I wouldn't expect group with a $1 billion to do that.

Also, I don't really think I would send algorithms with high value, and low ability to detect theft (if even protectable) to an offsite location, or especially this offsite location

One More

[Anonymous Coward]

Things like this already exist, but these guys are pushing the social, sharing, and gamification of this much more than the others. The existing 'gyms' are more domain specific. Many of them are started by research students or professors and then eventually left to rot right when they're getting useful. I wonder how long this one will last.

What makes the environment competitive is the ranked 'score board', nothing more. No prizes, etc... I expect people will quickly try to start gaming the system. AI systems are full of trade offs. You can always take someone's code, make a minor tweak to a variable or even just use a different random seed and get a slightly better result.

Want to fund Ai?

[GoodNewsJimDotCom]

If you want to fund AI, just assemble a team that knows their stuff. AI isn't as mysterious as it sounds. You just need good vision recognition to know your environment and then do goal oriented tasks on obects in the enviornment. www.botcraft.biz

why?!!

[zlives]

arn't these the same asshats that said the Al bundy is going to destroy the world?

AI Gym Membership

[Digital Mage]

You'll know that the AI has surpassed humanity when its able to successfully escape from its monthly gym membership fees.