People

Sleeping with Agents
Two days in the life of Anthony Pounds-Cornish

A PhD student from the Department of Computer Science is currently the guinea pig in an exciting research project that could revolutionise the way we live.

Our world could be populated by intelligent agent based devices working on our behalf. Items in our everyday life will contain networked computers working together 'intelligently' to improve the quality of our lives in the home and at work. Helping to get us there is PhD student Anthony Pounds-Cornish who has agreed to live in a recently created Intelligent Dormitory (iDorm) to test out the ground-breaking technology.

Anthony with the Agent

IDorm is a multi-use, multi-user, multi-agent environment in which most services used by occupants are provided by networked embedded computer based gadgets. The room will effectively get to know Anthony, his needs and when and why he carries out certain basic tasks, such as opening the blinds or turning a lamp on. The aim of the project is that the room will know him so well, it will eventually carry out those tasks for him.

The project is a collaborative venture between Computer Science and the Estates Management Section.

Wednesday 2pm
Having packed two day's worth of clothes, received a farewell kiss from my girlfriend and signed in with security, I was about ready to start the experiment. I had been chosen as the human guinea pig in an experiment to see how much a computer could learn about human behaviour in a 50-hour period. For a little over two days, a machine (the Agent), slightly bigger than my palm, was to look at everything I did in the iDorm and make decisions on my behalf based on what it learnt. This is an account of a 24-hour segment of that experiment

6pm
I had been using the computer interface to change the environment for various activities such as having the table lamp on when I was at the computer and having the bed lamp on when I was sitting on the bed. It was at this point that the sunlight began to reflect on the computer monitor and I adjusted the blinds to block the light. I then left to grab some dinner in the bar.

7.30pm
On returning to the iDorm, the Agent had switched off the lights in my absence. When I sat back down on the chair, it switched on the table lamp and opened the blinds. It had worked out that the light falling on the monitor was why I angled the blinds and thus, when the sunlight disappeared, it had put the blinds back to the way they were when I was out of the room.

Wednesday Night
Overnight, the Agent learnt that when I was out of the bed, it should switch the bed lamp on and when I was in bed, it was to switch it off. I woke at 3am to find the bed lamp was switching on and off at apparently random intervals. Cursing technology, I got out of bed and examined the pressure pad wiring. One of the pressure pad wires had come loose and so, as far as the Agent was concerned, I was getting in and out of bed more frequently than someone with bladder problems. Not particularly fancying a soldering job at 3.15am, I overrode the Agent and kept the light permanently off.

Thursday 8.30am
Having got up, washed using the toilet down the corridor, and put some clothes on (not necessarily in that order as the Graduate Secretaries are on the same floor), I fixed the pressure pad and went to get breakfast. Because the iDorm was designed as a multi-function room, I could work from there as well. I rang my girlfriend to tell her that I hadn't been cooked by the air conditioning and wasn't being held against my will by the Agent.

2pm
Throughout the day, I had to make less changes to the environment than I had the previous day. This was because the Agent was using what it had seen previously to pre-empt my actions and carry them out itself. This was one of the experiment's measures for success - less interaction meant more correct decisions by the Agent and less corrective measures by the user.

The Future?
The Agent has already been refined, based on the results of this experiment. It will also be able to see far more about the room by being attached to more sensors in time for the next experiment, which will see the human guinea pig stay in the iDorm for an entire week. I wonder if I can get a reduced rate of rent for the period?

Also in the printed June edition of Wyvern:

• Computer scientist attends artificial intelligence conference