Mozilla wants to develop an Open Source AI and for that it requires Training Data
Many of these companies like Facebook, Google have near patents in areas like search and social media. Their situation improves them to collect data, which supports them create better AI, which benefits them stay ahead of rivals. For the firms themselves, it’s a virtuous cycle, but without viable competition, companies can and do damage their dominance.
Now a new project from the Mozilla the non-profit producer of the Firefox browser is trying as an alternative to data copyrights, by requiring users to pool data in order to power open-sourced AI forces. The company’s first design is called Common Voice, with Mozilla claiming missionaries to donate vocal units to build an open-source speech recognition system similar the ones powering Siri and Alexa.
“Currently, the ability to manage speech identification could finish up in just a few hands, and we didn’t need to see that,” Sean White, vice president of developing technology at Mozilla, tells The News. He states to get data, the big businesses “can just refine everything coming in,” but for other players, there need to be additional rules. “The interesting topic for us, is, can we do it so the characters who are building the data also benefit?” he asks.
At the time, Mozilla is just getting data but intends to have its open-source voice identification available by the end of the year. Will it go in the Firefox browser? White won’t say, but continues: “We have some operations designed for that.” Currently, anyone can go to the Common Speech website and “donate” their speech by reading out sample sentences. They can also provide biographical data like age, location, gender, and accent. This data will help Mozilla avoid prejudice in building its voice recognition systems, says White, and guarantee that the technology can work accents something Google and Apple still fight with.