Computers that write reports, automatically update websites and selectively record spoken information during meetings are being developed in the US as part of a five-year defence research programme.
The artificial intelligence (AI) machines would be used to carry out tasks normally completed by personal assistants and secretaries, to free personnel from laborious tasks.
The machines, being developed by SRI International as part of the US government’s Perceptive Assistant that Learns (PAL) programme, could receive instructions through text or speech via personal digital assistant, desktop computer, wireless-networked laptop or phone.
To develop the technology researchers will have to make advances in machine learning, natural language understanding, reasoning, and human computer interaction. Improvements in these areas could also enable the machine to carry out routine tasks such as sifting e-mails, allocating office space and even coping with unexpected events, said Dr William Mark, vice-president for information and computer science at SRI, which is subcontracting out much of its work to universities and specialist companies.
For computers to learn to cope with the unexpected they will need to analyse how humans deal with surprises by applying learnt knowledge, said Mark. ‘We are looking at how to transfer learnt knowledge from one computer to another as a start. But computers are still very precise and don’t make the link between new situations and old experience. So we are bringing together the country’s top people to develop new ways of machine learning.’
SRI has received $22m (£13.8m) for the first stage of its project, called Cognitive Agent that Learns and Observes (CALO).
The project will also focus on visual and audio capabilities for human computer interaction. This could mean an AI would monitor a meeting, determine what information is needed, and only record pertinent speech and images.
Universities sharing the work include Massachusetts Institute of Technology,University of Stanford, University of California at Berkeley, and the State University of New York. Californian wireless IT provider Dejima, internet data integration specialist Fetch Technologies and Boeing’s Phantom Works are also involved.
The other major partner in the programme is Carnegie Mellon University, Pennsylvania, which has received £7m (£4.4m) for the first phase of a project called Reflective Agents with Distributed Adaptive Reasoning (RADAR).
The PAL programme is being funded by the US department of defence’s Defence Advanced Research Project Agency (DARPA), as part of its cognitive information processing technology initiative.