DARPA Wants to Make a Computer Program that Evolves for 100 Years

​The most fearsome specter brandished by the "artificial intelligence will ​enslave us all" crowd is the computer program that can adapt. Not only would such software's core function evolve and change, but it could survive both new and crumbling physical infrastructure.

This week, the Defense Advanced Research Agency (DARPA) published a request for proposals on how to design software than can do exactly this. Not just for 10 or 20 years, when much of today's technology will surely be obsolete, but for a century. The program is known as BRASS—Building Resource Adaptive Software Systems.

When Florida is underwater and California is a desert dystopia, BRASS programs will live on. When the personal computer and the smartphone are distant, archaic memories; when our bodies interface with machines; when our thoughts become data inputs; BRASS programs will adapt with us.

"100 years from now, I wouldn't dare to guess where technology will be at that point," said Steen Rasmussen, head of the University of Denmark's Center for Fundamental Living Technology. "At that point, technology today will likely have melded with biology. This would be something that has all the advantages of both of those regimes."

According to DARPA's proposal solicitation, the work will be split into the discovery and transformation abilities of a BRASS program. The discovery function would allow a program to scan its ecosystem—the hardware it interfaces with, the resources it relies on, the protocols it and other programs use to communicate, as well as the software it works with—and detect changes. The transformation component would allow it to morph itself and adapt.​

The agency is also looking for people or organizations who can work on platforms for a BRASS program. Some suggestions for BRASS platforms include autonomous robots that can adapt to a new environment or situation, Internet of Things applications that can morph to accommodate new devices and protocols, and mobile platforms.

The proposal period will close on May 22nd, and DARPA expects to dole out numerous grants of unspecified amounts to the groups that can achieve these goals.

Much work has been done to date on code that can reproduce itself, and even adapt to some degree. Rasmussen himself led a team that designed binary code that can copy itself like DNA. Artificial neural networks—programs that simulate the neuronal connections in the human brain—have also been programmed to evolve and change based on the tasks they need to perform. But these advances are rudimentary compared to the scope of the BRASS program.

"The question is: how can you make programs that are able to keep innovating?" Rasmussen said. "And we don't know the answer to that. That's one of the deep, big scientific questions: can we make code that is able to evolve open-endedly? The short answer is no. Do we know what the necessary conditions are? No."

A BRASS program in full swing is the classic scenario for a Skynet-style techno-dystopia. A computer program decides it needs resources to power the hardware it governs—in The Matrix, the logical solution was to turn humans into batteries (however illogical that is in real life). But should we worry that, if DARPA's plans pan out, we'll be facing a new, digital overlord? Rasmussen doesn't believe this is the case.

"It would be wonderful if we had really robust information technology," said Rasmussen. "That means that you wouldn't have any problems when an earlier version of your phone is broken, and you've lost all the information on it. Even if your phone is still alive, it can be difficult to transfer from one generation of phones to the next. It would be nice to have code where you wouldn't need to worry when you move from one platform to the next."

Since DARPA is a military agency, we should consider the applications of an adapting program in that light. Malware that can infect and spread in a foreign country's networks, adapting to every new change and scooping up data, undetected for an entire century—as governments rise and fall, as revolutions take place—would no doubt be a boon for generations of spies.

Security is also a concern. A program that can scan and automatically adapt to its virtual and physical environment would be continually updating its security measures to address every new threat. "Shapeshifting" software that can detect vulnerabilities and divert resources to protect them against attack has also been developed in the past. A BRASS program would be like one of these on steroids.

The BRASS program will require some serious breakthroughs in computing, and it will surely take years of work. But government funding provides a fairly steady pot of resources to drive innovation forward. And sometimes, this is exactly what a field needs.

"I do think that it is possible," Rasmussen said. "It's a wonderful challenge, and maybe this will put some new energy into the question."