The Pentagon’s research unit, known as the Defense Advanced Research Projects Agency, said Friday it’s looking to invest up to $2 billion in artificial intelligence systems over the next five years.

Among other things, the hope is to be able to develop more advanced and trustworthy weapons targeting systems that could one day be adopted by the U.S. military.

DARPA announced its plans during a press conference to mark its 60th anniversary . The agency’s primary role is to research new technologies that could help provide solutions to military problems. It’s backing AI in order to help the U.S. armed forces maintain their advantage over the forces of rival nations such as China and Russia.

The $2 billion investment is still rather small by military standards. The cost of a new program that will buy and maintain F-35 warplanes for the U.S. Air Force is likely to exceed a trillion dollars, by way of comparison.

Still, the investment is significantly larger than previous efforts, such as the recent $885 million contract awarded to Booz Allen Hamilton Inc. to work on unspecified AI programs. It’s also much larger than the $93 million DARPA will invest into Project Maven, which is an AI project previously led by Google LLC that’s looking to improve the ability of computers to select military targets from pictures.

DARPA said the $2 billion investment is part of a “Third Wave” campaign to design machines that can adapt and learn according to changing environments. The agency has a long history of throwing money at research into interesting new technologies. One recent example includes a contract tender for someone to help it build a blockchain-based messaging platform for secure communications. Last year the agency also committed $65 million toward research into brain-computer interfaces.

Steve Walker, director of DARPA, told a press conference the agency would be funding a variety of AI projects. For example, one goal is to tackle the logistics of vetting humans for security clearances, while another is to make military machines more energy-efficient. But DARPA is also looking at more controversial uses of the technology, including the concept of “explainable AI,” which relates to machines that can explain how they came to a conclusion and how its responses are justified.

DARPA is looking to explore “how machines can acquire humanlike communication and reasoning capabilities,” Walker said during the conference.

The idea of enabling computers and algorithms to take key decisions such as whether or not to strike a target with a missile is hugely controversial. The furor around Project Maven was a prime example of this, leading to a protest by Google employees who refused to work on software that might be used by the military to kill people. The protest persuaded Google to back out of the project once its contract expires.

The controversy may be justified. AI systems are surely very capable at doing things such as spotting targets and flying drones, but there’s still a lack of confidence in their ability to reason and make decisions to the same degree as humans can.

The problem is that in a battlefield situation, there are numerous variables that could come into play which a machine has never encountered before, and these could easily impact its judgment. For that reason, military commanders are unwilling to hand over control of their weapons systems to a nonhuman.

“With AI Next, we are making multiple research investments aimed at transforming computers from specialized tools to partners in problem-solving,” Walker said in a statement. “Today, machines lack contextual reasoning capabilities, and their training must cover every eventuality, which is not only costly, but ultimately impossible. We want to explore how machines can acquire human-like communication and reasoning capabilities, with the ability to recognize new situations and environments and adapt to them.”

Holger Mueller, principal analyst and vice president at Constellation Research Inc., said DARPA’s investment makes it clear that the U.S. and other nations are now effectively engaged in an “AI arms race” in order to build more intelligent weapons that could give their respective militaries an advantage.

“As much as AI and machine learning are imperative for enterprises, so they are for governments in general and for defense,” Mueller said. “As with all arms races, one can only hope that the ‘good guys’ are ahead. The time for a memorandum on not using AI for defense purposes has long gone, and so it’s literally an AI arms race from now on. We used to talk about numbers of tanks, planes, ships, troops, but now we have to add components like data centers, supercomputers, simulation speed and recognition speed to the equation.”

Image: Skeeze/pixabay