DARPA’s new initiative, known as the Communicating with Computers (CwC) program, aims to improve the ability of machines to communicate effectively with their human counterparts. The agency has two initial experiments planned, focusing on the somewhat differing fields of improved conversational skills and better cancer detection.
It’s difficult to miss DARPA’s intent to create more technologically advanced war machines, with initiatives such as the Ground X-Vehicles Technologies program aiming to make smarter, more agile vehicles. However, the agency is also aware of the importance of making machines communicate better with their human overlords, and that’s where the CwC program steps in.
Two-way communication with machines is a significantly more difficult proposition than it might first seem. A simple conversation between two people involves constant assimilation and contextual understanding of information – a process that’s second nature to humans, but represents a huge challenge for machines.
DARPA program manager Paul Cohen commented on this, stating, "Human communication feels so natural that we don’t notice how much mental work it requires. But try to communicate while you’re doing something else – the high account rate among people who text while driving says it all – and you’ll quickly realize how demanding it is."
The goal of the CwC program is to develop computers that think more like people, and are therefore better able to communicate as people do. The team will work to develop a system that’s capable of completing tasks that require effective communication, the first of which will be collaborative story-telling.
For the experiment, the two parties (one human, one machine) will be required to complete subsequent sentences to complete a story. This will require the machines to keep track of the ideas presented by its human counterparts, before creating their own ideas based upon established data – similar to a normal human conversation.
The second initial CwC task approaches the same problem from an altogether different direction, building computer-based models of the molecular processes that cause cells to become cancerous. While machines are better at reading large quantities of data, their ability to autonomously process said information falls short. The project will tackle this, aiming to develop a system that’s better able to judge the biological plausibility of proposed molecular models.