Echoes of Google’s GOOG-411 Experiment Stir Concerns Over AI Training Practices
OpenAI has introduced a new feature aimed at simplifying access to artificial intelligence for newcomers. The service, which streamlines interaction with ChatGPT compared to its web-based platform, is described as a “low-cost way to try it out through familiar channels.” However, OpenAI clarified that users looking for advanced capabilities should continue using the full ChatGPT service via traditional methods.
This announcement has sparked comparisons to Google’s 2007 experiment with GOOG-411, a free voice-based directory assistance tool. Although the service was discontinued in 2010, many believe its primary purpose was to collect voice samples to enhance Google’s speech recognition technologies.
At the time, Google executive Marissa Mayer openly explained the initiative’s underlying goal. “The speech recognition experts … say: If you want us to build a really robust speech model, we need a lot of phonemes… from a bunch of different speech samples,” Mayer said. GOOG-411 allowed Google to collect the necessary data to refine its technology for future applications, such as video voice extraction and other AI-driven tools.
While OpenAI has assured users that it will not use voice interactions to train its language models, the precedent set by Google has prompted some skepticism. OpenAI spokesperson Taya Christianson emphasized that these calls would not contribute to the training of large language models, aiming to allay potential concerns about privacy and data use.
