DARPA wants AI to explain itself

Artificial Intelligence Exploration programme will make AI justify its actions

Artificial intelligence is an area of huge technological change, but despite the high levels of investment AI is still...kind of dumb. Many implementations act as simple processing: put numbers in one end and get analytical data out the other, with no insight as to what goes on in the middle.

The USA's Defense Advanced Research Projects Agency (DARPA) wants to change that, launching a new Artificial Intelligence Exploration (AIE) programme that will explore ‘third-wave' AI, in which machines can contextually adapt to changing situations while explaining how they arrived at certain conclusions.

DARPA says that artificial intelligence that can show its working is crucial for turning AI-enabled machines into ‘trusted, collaborative partners in solving problems of importance to national security'.

An example might be an AI that spots a man carrying a gun, which it identified by seeing the barrel, trigger and stock all placed together in a familiar way.

One of DARPA's other aims with the programme is to speed up the development of AI projects. AIE will use a series of funding opportunities to reach a start date three months after an opportunity is announced, with feasibility established by researchers within 18 months.

DARPA Deputy Director Peter Highnam said: "DARPA has established a streamlined process to push the state of the art in AI through regular and relatively short-term technology development projects. The intent is to get researchers on contract quickly to test the value and feasibility of innovative concepts. Where we're successful, individual projects could lead to larger research and development programs spurring major AI breakthroughs."

Although the technology industry is by no means bound to DARPA's schedule, it will be hard to ignore a major government body throwing its weight (and investment) behind AI in this way.