The new generative AI model for Alexa, dubbed AlexaLLM, will be trained with users’ conversations, according to Amazon’s senior vice president, Dave Limp.
Limp made the revelation during an interview with Bloomberg TV on Sept. 20. AlexaLLM will offer a more personalized experience, will remember context, and will be able to have natural conversations — just like speaking to a human.
Alexa will “adapt to your cues and modulate its response and tone akin to human conversations,” Amazon said in an article previewing the chatbot. “Ask Alexa if your team won, and it will respond in a joyful voice if so; if they lost, the response is more empathetic. Ask Alexa for an opinion, and the response will be more enthusiastic, as it would if a friend was sharing a point of view.”
While the prospect of a “more intuitive, intelligent, and useful Alexa” seems exciting, privacy experts have voiced concerns about the use of users’ voice data and real conversations to train AlexaLLM. Limp said by opting for a customized version of Alexa, users implicitly volunteer their data to train the chatbot.
AlexaLLM Privacy Concerns
Limp said user privacy is paramount to Amazon, pointing to features like the ability to delete voice histories. He also touched on the importance of regulatory compliance and building a trustworthy AI.
Limp said Alexa’s accuracy will be refined to avoid AI hallucinations, inaccuracies, or unexpected responses.
Meanwhile, several experts have voiced concerns about the effects of generative AI chatbots like AlexaLLM on privacy.
“I don’t think we should accept that Amazon needs to retain those data for product improvement. They need affirmative opt-in confirmation to join these programs instead of being set at default,” John Davisson from the Electronic Privacy Information Center told NBC News on Tuesday, commenting on the revelation that AlexaLLM will be trained on users’ conversations.
Davisson also warned of potential unintended consequences, especially when children’s voices are used for training the model, pointing out possible issues with political bias, factual accuracy, and unexpected behaviors with the AI model.
In an interview with TechCrunch on Wednesday, Signal CEO Meredith Whittaker also expressed deep concerns about the relationship between artificial intelligence (AI) and privacy.
AI amplifies the surveillance business model, Whittaker argued, pointing out that the use of AI in applications like facial recognition with emotion detection capabilities is invasive and could lead to potentially inaccurate data about an individual’s emotional state. This data is then used by powerful organizations, such as employers and governments, to make impactful decisions.
Whittaker noted that individuals have limited control over their personal information in the vast datasets used to train AI models.
New Features for Alexa
The updated Alexa includes an “Alexa, let’s chat” function, which can be enabled with Visual ID. This allows users to activate Alexa without cue words by simply facing the screen on their devices. Additionally, users can have extended talks with Alexa without repeatedly using the activation word. Amazon said no images or videos are stored or sent to the cloud, and cameras can be manually disconnected for privacy.
Amazon emphasized the importance of trust, promising that privacy and security remain its priority. “As with all our products, we will design experiences to protect our customers’ privacy and security, and to give them control and transparency,” Amazon said.
Like other AI chatbots, we recommend avoiding sharing sensitive or personal information with Alexa devices.
A free preview version of the new chatbot is expected to be launched in the U.S. before the end of the year. You can catch a glimpse of Alexa’s new conversational capabilities in the video below.
For more generative AI news, follow us on X (Twitter), Threads, and Mastodon!
