How Can You Train AI to Get Smarter? The Intelligence Community Wants to Know

Christian Lagerek/

IARPA wants to guard against an "AI winter," during which there's less research and less funding for AI.

Humans aren’t the only ones who could benefit from daily brain teasers.

Computer systems also need to sharpen their information processing skills, and could use more training materials, according to intelligence community researchers.

In a new request for information, the Intelligence Advanced Research Projects Activity -- part of the Office of the Director of National Intelligence -- wants ideas for data sets, virtual environments and other training resources that could help artificial intelligence algorithms evolve.

There’s a misconception, according to IARPA, that it’s the algorithms themselves that limit the advancement of artificial intelligence. Often, algorithms can become more sophisticated when they are presented with the right training data.

"Many additional artificial intelligence problems may be solvable in the near-term, without significant innovations in the underlying algorithms, if the right training resources become widely available,” according to the agency. 

IARPA defines AI as the “computer simulation of cognitive processes such as perception, recognition, reasoning and control.”

For instance, an object recognition algorithm could hone its skills on a large data set of images of scenes or objects, each accompanied by a text description. Other artificial intelligence-related algorithms might learn from the kinds of virtual simulators used by video game and robotics designers.

In its request, IARPA warned that AI research has been largely a “boom/bust cycle characterized by promising bursts of progress followed by inflated expectations and finally disillusionment.” This can lead to an “AI winter” during which there’s less research and less funding activity, the solicitation said.

Nextgov has requested comment from IARPA.

(Image via /