Just in case Tim O'Reilly is running short on buzz phrases, I'm going to contribute another one suggested by the mix at eTech. In the early days of cybernetics, many applications were in the control of aerospace systems. Many of those control loops unavoidably involved the presence of homo sapiens in the cockpit, leading to the phrase Human in the Loop (HITL). As the linked article dryly observes "The adaptability of human operators made the basic approach of extracting equivalent transfer functions enormously challenging." (Translation: Under carefully controlled conditions, the experimental subject will do what he damn well pleases.)
That particular acronym died along the way, but the concept lived on. It's a portion of the intellectual underpinnings of Col. John Boyd's OODA loop. It moved with practitioners into the growing computer-human interface field. (E.g., old acquaintance Joy Mountford worked on helicopter control systems before running HI for Apple, if memory of old conversations serves me right.)
I think it's time to bring back the notion, with an important modification: Add an 's' to the first word. If it wasn't enough fun to have to model the sensory apparatus and variously trained responses of a single human, now we're awakening to cybernetic systems that incorporate the activities of whole packs of these annoying critters. Cases in point:
Start with Icosystems' Hunch Engine. As far as I can tell, the core of this is a genetic algorithm (GA) using humans as the 'goodness' function evaluator. An elegant hack-around on the famous knowledge acquisition bottleneck: Don't worry about what people say about how they think, just watch what they do. This may allow an attack on domains where evaluation is inherently subjective (design) or on those where experts make judgments based on long experience, with criteria they are unable (or unwilling) to articulate. Icosystems' Bonabeau termed these latter 'heuristocrats' (Bruce Sterling loved it) and one would presume he sells on that feature. Let's just hope it doesn't get along to the subjects of the experiments, or he will likely once again find "extracting equivalent transfer functions challenging."
Or we can look at the Mechanical Turk, where humans can literally be put into a program loop. And not just a single human, a whole marketplace is the big vision. Which are known to get complex. I grew up writing EXTERNAL statements; will the next generation need to write 'EXTERNALITY' declarations on their jobs?
Note in passing that Larry and Sergey literally coded all of us into their system when they created PageRank to follow our gestures in information space. And if you haven't noticed the feedback loop that one generated, you haven't been paying attention at all.
danah boyd observes - along with much else, that a culture of teenagers can make a perfect hash of your computational linguistics system if they decide that lexical tricks are their boundary markers. And scale and diversity, right. Back on the Well and BBS' we had our little flame wars. These days, if you're running a social platform of any scale, those 'clusters' could represent the real thing - your customers may be shooting at each other when they aren't typing - often over those same words and images. And if you don't think there's a feedback loop there, you're also not paying attention.
So other than late night musings, and another buzz phrase, is there a point? Maybe this: Through folks like George Dyson, Michael Goldhaber and Linda Stone, O'Reilly has been encouraging us to remember our intellectual heritage. Maybe we should be including some of the early systems thinkers in that remembrance as well. A lot of their big dreams didn't work out, and a few ended tragically. But we now have something they never had: experimental social cybernetics, at the scale of the whole net. And we can all play, at the price of a web crawl, some analytics, or a bit of a mashup here and there.