The term "synthetic intelligence" is sometimes used to describe "strongartificial intelligence" and sometimes to make the point that"artificial intelligence" as a term is an oxymoron. I'm not going toenter into that debate but want to here appropriate the term todescribe a collective form of intelligence where collaborative kinds ofwisdom can be aggregated toward some greater insight and evenpredictability.
Synthesizing intelligence--I prefer the verbal form to emphasize process--may, likefolksonomy, be a product only visible in retrospect. By this I mean that one person impressionistically tagging doesn't count for much. When hundreds or thousands provide tags, we learn not only a different system of categorizing, a different taxonomy, but also the multiple variant ways of seeing and categorizing content. It's the opposite, in other words, of the multiple choice answer where a question, however complex, has one and only one correct response. Folksonomies (and other forms of synthesizing intelligence) in aggregate display the wide variety of choices that count as answers to the question: What is this phenomenon? How do you define it?
I am brought to thinking about this today by Daniel Gilbert's bestseller STUMBLING UPON HAPPINESS. Gilbert is an eminent psychologist, at Harvard, who has conducted extensive research on the deceptions of memory, preference, and imagination. In this popular book, he makes the point that humans change so much over a lifetime that we cannot predict what will make us happy. Instead, he says, the best way of knowing what in the future will please us, we should stop imagining and, instead, find someone who already occupies that future and interrogate them about how they like it. Imagination, like memory, homogenizes experience, Gilbert argues. It misguides us precisely by not remembering all the details but only the overall narrative. Yet, he suggests, humans are far more alike than unalike (no matter how much we prize our uniqueness) so the best way to predict our future happiness is by asking someone else.
I don't find Gilbert's assumption of similarity among humans troubling in itself. What I find troubling is the concept of "similarity." His strokes are too broad. Of course we are similar in so many ways but it is the difference that makes for the "long tail" that existed long before the internet made so much choice and selectivity accessible. I would suggest that the small differences among us may be the most important. Those are what account for arguments, tedious department meetings, rebellious children, or broken friendships. I don't think we can overestimate the importance of small differences. Given that, I don't believe asking someone else about their experience necessarily tells us much about what we will or won't like unless we've already done a presort to know how much we share affective qualities with the other person, how much we agree on the particular range of small differences relevant to a certain situation.
So what does Gilbert have to do with Synthesizing Intelligence? Artificial Intelligence is based on a certain universalism, although the linguistic systems that generate AI carry the specificities of culture with them (something many AI specialists will heartily disavow). Gilbert's psychological approach is both too individualistic and too universal (i.e. happiness is personal, but all persons are pretty much alike). I am suggesting that the small differences not only matter but that, if aggregated and folksonomied, can point to a much more interesting form of computational preference choice (if not AI per se).
Synthesizing Intelligence, I am suggesting, is a way of sampling through the equivalent of preferences--a way of data mining affect--to determine likeness, affinity, and points of comparison. Within that set of like-minded individuals, one well may be able to learn more about one's happiness in the future by interviewing those who already occupy that future. Will I like this restaurant or movie or destination? Perhaps, I will have a better way of knowing that if I consult others who share other tastes. Match.Com for Life, not just for relationships, for minor preferences and not only for major choices.
What's the difference between what I'm calling "synthesizing intelligence" and the semantic web? A good question. But I think we need the aggregated wisdom of crowds to answer it.
[Photo by myself, posted courtesy of Flickr: Portofino Promontory on a gorgeous Ligurian April day with new friends. Relevance to this blog posting? None. But such an obscure posting deserves a pretty image!]