Michael P. Holmes and Dr. Bryan Morse, Computer Science
Imagine going through your late father’s belongings and finding that he had recorded many of his thoughts, ideas, and experiences in a special computer program. As you begin to use the program, recording the thoughts of your mind, the program brings to your attention various insights of your father’s that relate to what you are thinking. Thus a priceless part of his mind is retained and explored, even after he is gone. Suppose you are a lawyer at trial, and the opposing side brings up some evidence or arguments to which you had previously designed thorough refutations, but now you can’t quite remember what they were. Suddenly the PDA where you are taking notes flashes the forgotten plans to you, which you promptly use to dismiss the opposing argument. Perhaps you are a scientist trying to piece together a theory on some new phenomenon, and as you are taking notes into a special program, it suggests some links to previous notes that seem related to what you are writing. You examine them, and suddenly the pieces fall together into the theory you have been searching for.
The possibilities go on and on, but this fact is clear: Intelligent agents can augment human thought, memory, and creativity, and will be doing so in the very near future. Sufficient technology already exists or is on the verge of existence, and the need is obvious to anyone who has ever been flustered as they strove to recall a half-forgotten insight that would be so useful in their current task, if only they could remember. However, the pieces have yet to be put together in a user-friendly way with strong artificial intelligence.
This project started out as an idea I had while taking notes on my studies in the mission field. I accumulated volumes of notes, many of which I considered quite precious and illuminating, but once written, I tended to never look at them again! They were not indexed, just written in chronological order, and if I ever searched for a vaguely recollected entry, it was a considerable bother to find it. What if I had a computer do it for me? What if the computer could somehow figure out which notes were relevant to what I was currently studying, and bring them to me so they wouldn’t go to waste? I was reminded of Jane, an intelligence that lived among the vast interplanetary computer networks of Orson Scott Card’s Ender’s Game series. She had access to all the computer files, sensor readings, news, etc. of a hundred worlds, and was linked to Ender (the main character) by an electronic implant in his ear. Through that link, she could bring him any information that might help him in whatever situation he found himself. If there was someone coming his way, she could see it through the satellites. If decisions that would affect him were made far away, he would know about it almost instantly. Whatever the context, she could bring him relevant information precisely when it was needed most. I wanted one of those.
Related work has been done by Bradley Rhodes, formerly of MIT’s Intelligent Agents group. He developed Remembrance Agents, a type of what he more generally calls Just In-Time Information Retrieval Agents, or JITIR Agents. My objectives in beginning this project were to build on my initial concepts as well as what I learned from Mr. Rhodes’ work to create an easy to use, graphically based, accessibly-platformed (i.e., Microsoft Windows) intelligent program that would allow the user to enter ideas, then suggest enlightening connections to previous ideas. In essence, it would be a digital add-on to the brain. I named the program ThoughtWeaver.
One of the first things I realized was that the way in which suggestions were presented by the program had to be subtle and non-distracting, yet easily accessible. The big breakthrough here was the use of an Intellisense™-style suggestion window (see Figure 1). Having the agent run in the background while popping up links as it found them was tricky, but essential so that the user could carry on their normal business while the agent performed its job.
Getting the agent to only make suggestions of high relevance was key. Traditional search techniques make connections based on the presence of key words. I wanted a technique that would make connections between the meanings of ideas. What I came up with is a system I call proximity matrices. The innovation here was in considering relative positions, rather than simply the presence, of key words. This method requires more processing power than traditional techniques, but I hoped the payoff in capturing meaning and relevance would be worth the cost.
I was able to develop a fully functional ThoughtWeaver that uses Intellisense™-style alerts to display the suggestions of a proximity matrix-based search agent. Feedback from user-testing has been encouraging. All users have found the ThoughtWeaver program to be useful, smart, intuitive, and non-distracting. When a diversity of ideas is accumulated by the program, its suggestions are deemed relevant 50% of the time, on average. I personally have already had a number of distinctive experiences where ThoughtWeaver made suggestions that very much illuminated and expanded my train of thought in ways that would not otherwise have taken place.
The development of this program has generated many plans for its evolution into a constantly used and relied-on companion. It has also led to some great ideas on how to represent and associate meaning in computers, from synonym matching to more complex and unorthodox techniques that I hope will find application in future, much smarter ThoughtWeavers, translators, and other “understanding” computer applications. Ultimately, I see intelligent agent programs becoming like Jane of Ender’s Game fame. They will be with us almost constantly, actively gathering information about our thoughts and experiences (e.g. through monitoring speech), and proactively making suggestions and giving information that will enhance our abilities in many situations. They will be no substitute for inspiration or human ingenuity, but they will augment and assist us. Things will be dreamed, conceived, and accomplished that would not have been without their help. And it’s already happening.