Athletes and others focused on physical fitness would likely recognize Fitbit, the smartwatch that helps them track their fitness goals.
On the other hand, most recognize Google as the world’s handiest search engine. This company owns Fitbit, so it decided to integrate its latest artificial intelligence into the smartwatch.
READ: The best Apple Watch apps
Soon, Fitbit users will have Google Gemini, enabling their smartwatches to become their personalized fitness coaches.
How will FitBit AI work?
Tech news site The Street reported that Google and Fitbit will build a Personal Health Large Language Model (LLM) that will help users better understand Fitbit data.
It will run using Google’s Gemini AI and receive training from “anonymous data from real, human case studies gathered from accredited coaches and wellness experts.”
For example, Google said the AI model could analyze how exercise impacts a person’s sleep quality or vice-versa.
Android Authority reported Fitbit AI will use the analyses to provide actionable insights. Moreover, it will let users ask questions about their health in natural language.
The online search company says the model will become available to a limited number of premium Android users on FitBit Labs in late 2024.
At the time of writing, Google hasn’t shared more details about the features and their rollout.
Why are scientists worried about Fitbit AI?
The Street reported multiple AI experts like cognitive scientist Gary Marcus are worried about using large language models for fitness.
He said LLMs are largely and intrinsically hallucinatory, unreliable, and biased. Google explained that it trained its artificial intelligence on anonymized human data.
Yet, many details like its training process, data types, and environmental impact remain unknown.
“I am negative about LLMs. We need a better approach, one that is reliable, safe, fair, and trustworthy. LLMs will never be that, so it’s time to move on,” Marcus said.
“We should worried about LLMs being used in high-stakes applications where they simply aren’t reliable enough,” he added.
CNET shared a statement from Google related to this concern. The company explained launching the model as an experimental labs feature will let it gain user feedback before a public release.
However, Google didn’t comment on AI hallucinations, safety, and training data.