In the press.

WIRED followed ArtAdvisor into Artsy and made the distinction that mattered. Artsy’s Art Genome described traits of artworks and powered recommendations; the art brain I had built sought a wider account of cultural significance and value. At Artsy, institutional, gallery, browsing, buying, and market evidence could meet—without pretending that art had become certain.

A broad account of Artsy’s attempt to make the art world legible through data. Carter Cleveland describes why the ArtAdvisor acquisition mattered: recommendation could say what someone might like, while my work tried to model the larger cultural context behind why an artist or work becomes valuable.

The trade paper placed the work inside the art market itself. It identified me as Artsy’s first Chief Scientist and connected the earlier taste research, trend prediction, and artificial intelligence to Artsy’s effort to give collectors useful, personalized intelligence.

Before the Artsy acquisition, this account gave ArtAdvisor a substantial section of its own. The premise was not that data should replace expert intuition; it was that large cultural datasets could help expose some of the evidence and learned judgments from which expert intuition is made.

The Magazine’s 2005 Year in Ideas selected the Synesthetic Cookbook: a “smart” cookbook that searched existing recipes by sensation and emotion as well as ingredient and method. Its deliberately unruly example asks for a dish that is sweet, salty, sour, meaty, dessert-like, primal, and sophisticated; the database answers with Brunswick stew.

The magazine’s annual portrait of twenty accomplished single Bostonians profiled Hugo first: then chief scientist at Hunch and a taste researcher at MIT, working where technology met style and taste. The profile moves from emotional recipe search and custom Scotch blended for Proustian memory to Coco Chanel as favorite philosopher. It closes: “A guy to watch, whether you want tips on couture or the hottest IPO.”

The Times’s September 4 front page reported from a flooded playa, where thousands were stranded and asked to conserve food and water. In the opening column, Star declined the language of helplessness: “When it’s time to leave, we’ll leave. Burners aren’t victims.”

A full Food-section portrait of the attempt to “help people become fluent in taste.” It follows Taste Spaces, the Synesthetic Cookbook’s analysis of 160,000 recipes, and Gulp Fiction’s attempt to compose new ones from learned patterns. The reporter preserves the experiment’s limits too: generated directions could be lyrical and absurd, and the “kitchen chemistry” had not caught up with the computational imagination.

The Economist introduced Åzone Futures Market, the Guggenheim’s first online exhibition and a forecasting game. Visitors received 10,000 units of Åzone currency to invest in possible technologies across the coming decades: lab-grown replacement limbs, laser-regenerated teeth, genome-edited plants. As the clipping put it, “the real action is online: place your bets.”

Fast Company named the people who made Åzone Futures Market, including me as the data scientist who built its market algorithm. The work was a web-based simulated market in possible technologies, with a public terminal at 181 Front Street in New York’s Seaport Culture District.

The surviving continuation page is headed “Gadgets finding their feminine side.” The broad feature on gender assumptions embedded in products included GenderLens, built from 150,000 blog entries. Its news-filtering experiment created men’s and women’s feeds; in four of five categories, they predicted which stories each group preferred. The article’s larger question was practical: what changes when an interface stops assuming one default kind of user?

Elise Ackerman’s feature introduced Hunch while it was still new: a question-and-answer site that learned a person through a short sequence of questions, then used those answers to personalize decisions in unrelated domains. The piece survives through its newspaper syndications.

Metafor treated an ordinary story as a partial program specification: nouns suggested objects, verbs suggested actions, and relations suggested structure. Technology Research News made it a cover story; Technology Review, MIT Technology Insider, IEEE Software, Optimize, Slashdot, and ACM TechNews also followed the experiment.

A ring or watch reads the tag on a physical object; Ambient Semantics consults InterestMap and sends a recommendation to the wearer’s phone. The 2004 report describes working wearable prototypes and a map of 50,000 items. Technology Review returned in 2005, by then describing two million relationships among 100,000 items—and a next step in which the same sensor could notice shared interests between strangers before the opportunity passed.