AI software is figuring out how to best humans at designing new AI software

Who programs the programmers? Soon enough, it might not be people behind the development of advanced machine learning and artificial intelligence tech, but other AI. MIT looks at the most recent work done by a range of different organizations, including Google Brain, who are working on AI that can develop machine learning software – and finds that in many cases, the results that come from machines coding other machines match or even exceed equivalent work done by humans.

Does that mean even machine learning programmers are facing employment extinction? Not exactly, and not yet – efforts to create machine learning programs that best their human-designed equivalent require a lot of computing firepower thrown at the problem; Google Brain's person-besting experiment in building image recognition systems via AI development used 800 ugh-powered graphics processors working together, which is a costly endeavor to be sure.

But the advantages are clear, and there's a path towards lessening the resource burden in creating these systems, too. Offloading machine learning development to machines would help address the serious shortfall in human expertise in the area, for instance, which is currently leading to a land grab for startups and academic talent with any kind of AI chops. It could also free up human researchers to work on more important problems, rather than spending their time on more rote or mundane training of AI systems using massive data sets.

AI tuning AI has another potential benefit – improving the learning curve for AI systems so that the volume of data required to produce meaningful results is also cut down. This would help greatly with endeavours like self-driving automobile systems, for which even millions of miles driven is little more than a good start in terms of delivering real-world usable results.

MIT Media Lab is open-sourcing its own efforts to create learning software from other machine learning programs, and this should help with industry-wide efforts to make this a practical way to create new software.

AI industry experts are quick to point out that developing machine learning requires an immense human effort at the outset, but offloading some of that work to other machine learning systems could drastically reduce the human input required at the beginning of the process and throughout. That'll mean much-improved go-to-market times or products that require AI, including self-driving cars – but it definitely won't help alleviate concerns that machine systems will replace an increasing number of human jobs in a growing number of fields.