Where to begin with this week? Funny-not-funny #ManInTree, funding news from Convoy and a new AI startup, DimensionalMechanics, and the teen chatbot from Microsoft corrupted by the Internet over the course of a day? At least we’ve got hoops (and long odds). Here now, Xconomy Seattle’s Week in Review:
—Microsoft has received more attention in recent weeks for two incidents involving teen girls, a state of affairs that has executives in Redmond asking hard questions and making public apologies.
On Wednesday, the software giant released a chatbot called Tay. It was a research project meant to talk like a teen girl and learn from its interactions with real people on the Internet. You’ll never guess what happened next. Take it away, headline writers of The Telegraph: “Microsoft deletes ‘teen girl’ AI after it became a Hitler-loving sex robot within 24 hours.”
The last tweet posted to the @TayandYou Twitter feed:
c u soon humans need sleep now so many conversations today thx
— TayTweets (@TayandYou) March 24, 2016
This Wired article delves into Tay’s technology, and how its ability to learn from interacting with humans—at least some of whom appear to have banded together to teach it to spout hateful, racist comments—was its undoing. As Allen Institute for Artificial Intelligence CEO Oren Etzioni put it to Wired: “This is an example of the classic computer science adage, ‘Garbage in, garbage out.’”
Maybe Microsoft should have more heavily filtered Tay’s responses, or at least devised a way to indicate they were assimilated comments of humans interacting with it, rather than its own inventions, as University of Washington’s Ryan Calo, an expert in AI law, points out in Wired.
Peter Lee, head of Microsoft Research, came out Friday with a blog post about Tay. “We are deeply sorry for the unintended offensive and hurtful tweets from Tay, which do not represent who we are or what we stand for, nor how we designed Tay,” Lee writes. “Tay is now offline and we’ll look to bring Tay back only when we are confident we can better anticipate malicious intent that conflicts with our principles and values.”
He also notes that a Chinese-language chatbot, Xiaolce, is working great. Maybe you could read something into this about U.S. and Chinese Internet culture. Lee describes “a coordinated attack by a subset of people [that] exploited a vulnerability in Tay.” And, he adds, “AI systems feed off of both positive and negative interactions with people. In that sense, the challenges are just as much social as they are technical. We will do everything possible to limit technical exploits but also know we cannot fully predict all possible human interactive misuses without learning from mistakes.”
Meanwhile, the very human failure in Microsoft’s games division in booking exotic dancers dressed like school girls to mingle with attendees at its Game Developers Conference party in San Francisco earlier this month was arguably worse for a company—and an industry—trying to create a climate more welcoming to women.
—Another artificial intelligence company, DimensionalMechanics, announced funding. The Bellevue, WA-based startup says it raised $4.7 million from angel investors, led by Kent Johnson, along with board members and company founders, including CEO Rajeev Dutt and senior vice president of product Dave Hebert. The company is working on a product called NeoPulse, “a flexible cloud-based platform that will enable users to solve enterprise challenges by incorporating the power of AI into existing systems,” it says in a news release. DimensionalMechanics is “actively seeking strategic technology partnerships as it builds out NeoPulse’s first phase of core capabilities.”
—Google is making available some of its machine learning technologies to developers through Cloud Machine Learning. The company will allow developers to use APIs for speech recognition, translation, and computer vision, accessing the same technology at play in the company’s voice search and voice typing products. This underscores the tough competition from technology giants faced by startups aspiring to deliver similar machine learning or AI capabilities as a service.
—Here’s a new resource for those of us trying to keep tabs on machine learning and AI: From Seattle startup Algorithmia comes Emergent Future, a “weekly, hand-curated dispatch exploring technology through the lens of artificial intelligence, data science, and the shape of things to come.” You can sign up for it here.
—Convoy, a Seattle-based startup trying to bring on-demand convenience to the trucking business, raised a $16 million Series A funding round from lead investor Greylock Partners. The capital will go toward technology development and new service regions, beginning with Oregon. Reid Hoffman and Simon Rothman of Greylock join Convoy’s board, along with Hadi Partovi, co-founder of Code.org. Other new investors in the Series A are Jeff Wilke, Kevin Systrom, Gary Chartrand, and Mike Gamson. The funding comes hard on the heels of Convoy’s $2.5 million seed round, raised last fall from Jeff Bezos via Bezos Expeditions, Salesforce.com CEO Marc Benioff, eBay founder Pierre Omidyar via Omidyar Technology Ventures, and others.
—Trucking seems like a business where an “Uber for X” model might make a lot of sense. But that may not be the case for other on-demand businesses, which are struggling to lower prices to the point where they can reach a mass market, as Farhad Manjoo writes in his New York Times column this week: “The opportunity for Uber to become a regular part of people’s lives was huge. Many people take cars every day, so hook them once and you have repeat customers. Finally, cars are the second-most-expensive things people buy, and the most frequent thing we do with them is park. That monumental inefficiency left Uber ample room to extract a profit even after undercutting what we now pay for cars.
“But how many other markets are there like that? Not many. Some services were used frequently by consumers, but weren’t that valuable—things related to food, for instance, offered low margins.”
—Speaking of traffic, last week, we talked about Seattle’s traffic woes as quantified by a recent ranking by INRIX. Another traffic index out this week, from TomTom, says Seattle’s rush hour is second worst in the nation, and congestion on the roads overall is worse than all but three other cities: Los Angeles, San Francisco, and New York.
Development along one of the most congested corridors in Seattle—Mercer Street in South Lake Union—continues unabated. On Thursday, Google and Vulcan revealed plans for a South Lake Union office complex that could accommodate upwards of 3,000 workers.
And lest we forget that South Lake Union is a hub of life sciences research in addition to one of the hottest tech neighborhoods in the city (or the country), BioMed Realty opened a new research building Thursday at 500 Fairview Ave. The 122,000-square-foot addition adjoins the company’s existing lab and office building at 530 Fairview Ave., which houses companies including Novo Nordisk, NanoString Technologies, Presage Biosciences, and Blaze Bioscience.
—The Giant Steps exhibit we visited before it opened in King Street Station earlier this month asked artists to contemplate what they would build given a brief residency on the Moon. Here is an artist, Michael Najjar, who aspires to be the first artist in space, and is among the astronauts training to be part of a future Virgin Galactic crew. Smithsonian Magazine profiles his latest exhibition of photographs and compositions, outer space, which focuses an artful lens on the technology of modern spaceflight. (Giant Steps, by the way, is open weekends through April 3.)
—Finally, how’s your March Madness bracket? Walt Hickey, writing at FiveThirtyEight, unpacks the odds of nailing a perfect bracket—picking the winner of all 63 games of the NCAA Division I Men’s Basketball Tournament: 1 in 9.2 quintillion. I like this essay because it explains the underlying math, but also goes into the futility of trying to explain or understand very large numbers. Hickey quotes Randall Munroe, the genius behind xkcd and a recent book, “Thing Explainer,” on how to decide whether it’s even worth presenting a very large number to your reader, without sufficient context: “A good rule of thumb might be, ‘If I added a zero to this number, would the sentence containing it mean something different to me?’ If the answer is ‘no,’ maybe the number has no business being in the sentence in the first place.”