I’ve just returned—in the cir­ca­dian rhythms sense of get­ting back up in the morn­ing and going to bed at night—from GECCO 2006 in Seat­tle. Much to write about and dis­cuss, but here are some bul­let points I gleaned from the innu­mer­able ses­sions attended and break-​​outs dis­rupted with enthu­si­as­tic banter:

• Genetic Pro­gram­ming and related things can be used to cre­ate com­plex human-​​competitive designs. That’s com­mon knowl­edge, among the cognoscenti at least. But the leap from build­ing a new kind of lens sys­tem or a new kind of ana­log cir­cuit, to build­ing a new kind of com­plex object like a word proces­sor or a robot or a work of art or a liv­ing thing… you need peo­ple in the loop. User-​​centric design. Had beer with Ian Parmee and Chris Simons, for instance, and attended sev­eral tuto­ri­als and work­shops on deal­ing with user fatigue in auto­mated design sys­tems, fram­ing ques­tions for glean­ing user wishes, and saw the same ques­tions com­ing up over and over in other ses­sions where peo­ple weren’t think­ing yet about the role of the user (aka “cus­tomer”) in evo­lu­tion­ary design, search, and opti­miza­tion. Engi­neers need to think about those peo­ple. About not what this stuff is for, but who it’s for.
• Sym­bolic regres­sion is not for curve-​​fitting. It’s for model dis­cov­ery. Analy­sis. Under­stand­ing. That’s another broad class of prob­lems where user-​​centric think­ing has started to per­co­late into the prac­ti­tion­ers’ con­scious­ness: sym­bolic regres­sion is for exploratory data analy­sis, not design divorced from a human brain.
• Genetic algo­rithms, evo­lu­tion strate­gies, neural net­works, and other numer­i­cal opti­miza­tion and search tools are quickly falling into the canon of oper­a­tions research and the toolkit of opti­miza­tion. I saw sub­stan­tially fewer papers on over-​​abstract toy prob­lems this time around; instead, where the­ory was explic­itly the goal, there was much more focus on using results from GAs and ESs to explore the struc­ture of abstract toy prob­lems from other domains.
• Of 500+ par­tic­i­pants expressly work­ing on opti­miza­tion, search and design with evo­lu­tion­ary algo­rithms and related approaches, I can’t think of a sin­gle one con­cerned with opti­mal­ity cri­te­ria. Maybe one or two pay atten­tion to it, but nobody gives a damn for the­ory, for appli­ca­tion, for cus­tomers. Is this a good thing?
• Lots of peo­ple seem to be leav­ing Google, after work­ing there for a while. How does that come to be the case?
• Startup time is upon us once again. There was lots of buzz about pro­duc­tive, suc­cess­ful high-​​tech star­tups. A sub­stan­tial drop in the num­ber of morose burnt-​​out entrepreneurs.
• All that stuff high-​​tech peo­ple are sup­posed to like doing at con­fer­ences, like hav­ing a back-​​channel and blog­ging and crap? Bah. Nobody has a clue. How can the cul­ture of com­puter sci­en­tists be so back­ward? Oh wait… never mind. I get that one.
• What one wants is to be able to talk with a diverse club of smart peo­ple, arrange to do short one-​​off research projects and sim­u­la­tions, pub­lish papers or cap­ture intel­lec­tual prop­erty quickly and eas­ily, and move on to another con­ver­sa­tion. Quickly. Eas­ily. For a liv­ing. Can’t do that in indus­try. Can’t do that in the Acad­emy. Yet in my expe­ri­ence, sci­en­tists and engi­neers all want it. Maybe even a few math­e­mati­cians and social sci­en­tists do, too.