(Page 2 of 2)
data-driven design is just as important as recommendations, which influenced the ongoing development of the app.
Some red-faced, embarrassing moments have emerged in the course of developing AI technology. Microsoft’s Tay chatbot went live on Twitter in March—and within hours began to spout offensive language and statements. Tay was created to imitate the speech patterns of a 19-year-old girl. It also learned by interacting with human users of Twitter. Within one day, Tay was mimicking the worst of the Web’s trolls, which led to Microsoft taking the AI offline and apologizing for the inflammatory tweets.
This came as a bit of a shock for the company since its natural language chatbot called Xiaoice met with a far different reception in China, Tamblyn said. Xiaoice learned if users were not feeling well, he said, and then checked on them the next day. It could also handle simple financial banking transactions.
Looking to try something similar domestically, Microsoft introduced Tay on Twitter and a few other platforms. “We realized very, very quickly, the way in which the U.S. market interacted with this bot was very different than they did in China,” Tamblyn said. “This thing turned into a racist nightmare in about four hours.”
Despite Tay’s unexpected turn to the dark side, Tamblyn said the test still proved to be a good learning experience for Microsoft. “If you’re going to play in this space, you have to be prepared to make some mistakes,” he said. More intelligence is filtering into applications, Tamblyn said, such as Microsoft Office, in ways that were not done several years ago.
Even with the progress made so far, Tamblyn said, “We are not even yet at basecamp in terms of the evolution of this AI space.” The companies that will be most successful in advancing the capabilities of artificial intelligence will be the ones who have control of vast amounts of data, he said. Weinstein said there is room for startups to come into this sector, yet consolidation is underway as the different players seek to gain access to more and more information.
“From the perspective of IBM, we are doing it ourselves,” said Weinstein, “We recently bought The Weather Channel, pretty much purely for their data because there’s so much to be learned from weather data. You can use it in so many ways for machine learning.”