Will DARPA’s Third Wave Get AI to Think Like a Human?

AI Quantum Computing

In the quest to keep the upper hand in the burgeoning field of artificial intelligence (AI), the Pentagon’s top research arm just put $2 billion in chips on the table to spur development of the third wave of AI technologies.

In announcing the multi-year program last week, the Defense Advanced Research Projects Agency (DARPA) said the program will pump funding into new and existing projects to turn current AI tools into “partners in problem-solving.” It also will incorporate another recently announced DARPA program, called Artificial Intelligence Exploration (AIE), which is fast-tracking third-wave projects through the agency’s “Disruptioneering” contracting process to get ideas from the drawing board to proof-of-concept in 18 months.

It’s all part of a concerted effort by the Department of Defense (DoD)–and the U.S. government in general—to support AI development in light of substantial investments by other countries into AI technologies. China has been straightforward about its plans to dominate, recently announcing efforts such as a $5 billion development fund, a $2.1 billion technology park, and $4.5 billion in private investments. China wants to build its AI industry into a $1 trillion market by 2030. Other countries also are pouring money into AI research, if not quite on China’s scale. Russia, whose president has famously claimed that the leader in AI would be “ruler of the world,” is building an “innovation technolpolis” of its own.

The United States, meanwhile, is working with industry–where major companies (Google, IBM, Microsoft) and an array of startups are advancing to find faster paths to unlocking AI’s capabilities. The DoD, for instance, earlier this year established its Joint Artificial Intelligence Center, which for starters brings some 600 AI projects under one roof, to get AI tools to warfighters more quickly. And the White House in May staged an AI Summit with leaders in business and academia to promote coordinated AI development, although several industry groups felt the summit avoided too many key issues. DARPA now is adding $2 billion to the pot.

With its pursuit of third-wave technologies, DARPA isn’t necessarily looking for sentient robots that would fulfill sci-fi fears, but the next phase in AI’s development. Machines that act as “partners in problem-solving” actually has been the goal of DoD’s AI research for a while, and getting to that point would represent quite a leap in the technology.

For all of AI’s impressive feats in recent years, the majority of AI technologies are rule-based programs that are good at one thing, and require hundreds of thousands of hours of training to get to that point. An AI program can beat a world champion at chess, but the same machine couldn’t figure out how to play checkers if it hadn’t been programmed for it. That’s the first wave of AI. The second wave, such as neural networks, have made progress in filtering knowledge through computational layers and has led to significant advances in fields such as natural language processing, face recognition, and navigating driverless cars.

The third wave would push AI toward the human-machine teaming that the DoD envisions for the technology. This will depend on machines being able to think and communicate more like humans, and thus be able work side-by-side with them. Part of the partnership will depend on trust, which DARPA is looking to develop through programs such as Explainable Artificial Intelligence, which seeks to get machines to explain in human terms how and why they reached a conclusion, which AI can’t do now.

Enabling AI agents to be more human in how they function would be more cost-effective as well as practical. “Today, machines lack contextual reasoning capabilities, and their training must cover every eventuality, which is not only costly, but ultimately impossible,” Steven Walker, director of DARPA, said in announcing AI Next.

The program will be applied to the roughly 60 AI projects DARPA has underway, as well as future efforts. The agency said it expects to issue a series of Broad Agency Announcements over the next 12 months in search of new programs worth pursuing.

Recent