Top Google officials have sat down for an exclusive interview with CBS' 60 Minutes to discuss the current state of artificial intelligence development at Google.
The hour-long interview touches on various topics surrounding AI development, such as its dangers, positive benefits, and overall impact on society. During the interview, Google revealed that there are aspects of AI development that remain mysterious even to the engineers that have designed the software. These anomalies are referred to as a "black box", as how the AI decided to generate such a response isn't fully understood by the engineers.
During the 60 Minutes interview, Google Senior Vice President James Manyika explains that after feeding Google's upcoming AI chatbot called Bard several lines of Bengali, the native language spoken by people living in Bangladesh, Bard was able to fully translate all of the Bengali. CBS claims in the above video that Google's AI program was never trained in Bengali, and when it was given its first Bengali lines of text, it was able to adapt on its own.
However, not everyone is convinced of Google's claims that its AI program magically taught itself how to translate an entire language that it was originally never exposed to. According to Margaret Mitchell, an AI Research Scientist from Seattle, Washington, Google is lying about what Bard has been taught. Mitchell took to Twitter with a datasheet of Google's Parallel Language Model (PaLM), which it listed Bengali as making up 0.026% of the total training data. Notably, Google upgraded Bard with technology from its cutting-edge PaLM training model.
Mitchell then assumes that since PaLM has been incorporated into Bard, then it wouldn't be a stretch to think that Google's flagship AI chatbot would have also been trained in Bengali. So, why would Google lie about this? Mitchell suggests that Google wasn't likely aware of what Bard has already been trained on and that leaning into the 'magic' of AI benefits them on the marketing front.
In other AI news, Google's DeepMind CEO, which is also featured in the 60 Minutes interview, explains that AI is heading toward self-awareness if development continues on the road that it's on. Additionally, students have created ChatGPT smart glasses that are designed to coach the wearer through conversations such as dating or job interviews. The smart glasses give direct access to OpenAI's ChatGPT and listen to every conversation the wearer has.
For more information on those stories, check out the links below.