AI Is Learning How To Create Itself

Humans will want to do also. But, of course, they already can. It's just that, as they have learned the hard way via raising children, replicating yourself does not usually turn out exactly as planned. JL

Will Heaven reports in MIT Technology Review:

For decades artificial intelligence researchers have tried to build algorithms to mimic human intelligence, but the real breakthrough may come from building algorithms that attempt to mimic evolution’s open problem solving and sitting back and watching what emerges. Intelligence is the current climax of a continuous and open process. In this sense, evolution is quite different from algorithms in the way that people often think of them, as means to an end.

A little stick figure with a wedge-shaped head shuffles across the screen. It moves in a half crouch, dragging one knee along the ground. It’s walking! Er, sort of. 

Yet Rui Wang is delighted. “Every day I walk into my office and open my computer, and I don’t know what to expect,” he says. 

An artificial-intelligence researcher at Uber, Wang likes to leave the Paired Open-Ended Trailblazer, a piece of software he helped develop, running on his laptop overnight. POET is a kind of training dojo for virtual bots. So far, they aren’t learning to do much at all. These AI agents are not playing Go, spotting signs of cancer, or folding proteins—they’re trying to navigate a crude cartoon landscape of fences and ravines without falling over.

But it’s not what the bots are learning that’s exciting—it’s how they’re learning. POET generates the obstacle courses, assesses the bots’ abilities, and assigns their next challenge, all without human involvement. Step by faltering step, the bots improve via trial and error.

But there is another crucial observation here. Intelligence was never an end point for evolution, something to aspire to. Instead, it emerged in many different forms, from countless small solutions to challenges that allowed living things to survive and take on future challenges. Intelligence is the current climax of a continuous and open process. In this sense, evolution is quite different from algorithms in the way that people often think of them, as means to an end.

It’s this open ending, glimpsed in the seemingly aimless sequence of POET-generated challenges, that Clune and others believe could lead to new kinds of AI. For decades artificial intelligence researchers have tried to build algorithms to mimic human intelligence, but the real breakthrough may come from building algorithms that attempt to mimic evolution’s open problem solving and sitting back and watching what emerges.

Researchers are already using machine learning itself, empowering it to find solutions to some of the toughest problems in the field, such as how to make machines that can learn more than one task at a time or cope with situations they haven’t encountered before. . Some now think that taking this approach and executing it could be the best path to general artificial intelligence. “We could start an algorithm that initially doesn’t have a lot of intelligence inside of it, and see how it starts itself up to potentially AGI,” says Clune.

The truth is, for now, AGI is still a fantasy. But that’s largely because no one knows how to do it. Advances in artificial intelligence are partial and are being made by humans, and progress generally involves adjustments to existing techniques or algorithms, resulting in incremental leaps in performance or accuracy. Clune characterizes these efforts as attempts to discover the building blocks of artificial intelligence without knowing what he is looking for or how many blocks he will need. And that is just the beginning. “At some point, we have to take on the herculean task of putting them all together,” he says.

Asking AI to find and assemble those building blocks for us is a paradigm shift. You are saying that we want to create a smart machine, but we don’t care what it will look like, just give us what works.

Even if AGI is never achieved, the self-learning approach can still change the types of AI that are created. The world needs more than a good Go player, says Clune. For him, creating a super smart machine means building a system that invents its own challenges, solves them, and then invents new ones. POET is a small glimpse of this in action. Clune imagines a machine that teaches a robot to walk, then to play hopscotch, and then perhaps to play Go. “Then maybe he’ll learn math puzzles and start inventing his own challenges,” he says. “The system continually innovates and the sky is the limit in terms of where it could go.”