Saturday, June 6, 2009

Perhaps Google Wave will form a good habitat for intelligent agents?

Google Wave is a significant innovation. See the stuff on the google pages for intro, prepare to be impressed, but I won't repeat it here.

It just struck me that perhaps Google wave will enable the growth of a bunch of interesting intelligent agents. These are of course not new, and the initial wave team has certainly gone out of their way to facilitate robots to participate in conversations. Intelligent agents in the web area had a boom in the late nineties but sort of faded from the public view after that. Patti Maes's research group at the MIT Media lab did a bunch of intersting stuff, I recommend reading through their publications.

One of the thing that are striking when reading these articles is number of hoops that needed to be jumped through in order to get anything working. Proxy servers, strange layouts, genetic programming and even then the performance was interesting but not altogether impressive, and certainly not ready for prime time in the form of mass market products.

This might change now. If wave is launched, as I hope it is, most of the "hygiene factors" necessary for a good user experience will be taken care of in a properly Googlish, Marissa Meyerish-pedantic manner. The wave team has aleady taken care of many but not all of the issues an intelligent agent needs in order to thrive, so this means that it might very well become common practice to include agents in conversations, and that means that agent-writers will finally have a big consumer market to offer their wares to. This will take place in two distinct ways:

Semi-covertly; Like for instance a spell checker that is quietly but consistently checking all input for spelling mistakes. Many other types of covert agents can be imagined: Looking up words and phrases and trying to find related texts in books, articles etc. If a covert agent does this work, it will be added as some kind of decoration to the actual content: In separate layout boxes, as links on the side of the main content (like google does sponsored links today) etc. The main difference being that you are in charge of which agents are doing the decorating, not Google.

Overtly: Agents that are more or less real subjects. They can present input somewhat like humans in various roles: Research assistant ("I found sixty nine references to this compound in the standard journals, but none of them refering to toxicity in larvae, however ..."(etc.)), and other roles as well I can imagine (boss, mother, social network surveilance agent etc. :-) This is of course a minefield for user interaction design, but that doesn't really matter: You see, this kind of behavior simply won't happen if it's not wanted. Agents are invited, not imposed on conversations. Furthermore agents are simple to add and since the substrate on which they act is already quite rich, work on the agents can be concentrated on adding interesting substance. When some interesting agents are produced, they can start breeding and mixing, perhaps metaphorically and perhaps almost literally through genetic programming. I believe this could be the start of some really interesting times. I just can't wait to get my hands on wave.

Please please please Google let me have access soon? ;)

No comments: