Apple and Google have both joined the generative AI race, introducing features that will bring AI capabilities to our smartphones. This means exciting new possibilities for our phones, but it also raises privacy concerns.
Apple unveiled a suite of features on June 10th that bring generative AI to iPhones, iPads, and Macs. These features allow users to transcribe recorded calls, generate emails, summarize notes, remove objects from images, create illustrations and animations, and even generate custom emoticons.
Last September, Google announced similar features, including real-time transcription of audio recordings. Samsung is also in the game, offering a native AI model on its flagship devices to process audio recordings, phone calls, and summarize web pages.
But are these features running locally on our devices? Not all of them. Apple clarified that while Siri can handle basic queries locally, more complex queries might need to access OpenAI’s ChatGPT, relying on the GPT-4o multimodal AI model. Google’s Pixel phones rely on an internet connection to access large AI models hosted on cloud servers for many of its features. Samsung takes a similar approach, allowing users to choose between using local AI features, such as transcribing a voice note, or accessing a cloud-based AI model to generate summaries of recorded voice notes.
Will all phones get generative AI soon? Apple’s AI features are currently limited to its ‘Pro’ iPhones, while Google’s latest phones all support its AI capabilities. Samsung only offers AI on its flagship Galaxy S24 series, but plans to expand to new devices next month. The availability of these features depends on the phone’s processor, with only flagship processors from Qualcomm and MediaTek supporting local AI processing.
What are the trade-offs? Privacy experts have raised concerns about the personal data that local AI models can access on our phones. Apple assures users that no personal ‘contextual data’ will be transferred online, maintaining privacy on OpenAI’s servers. Google and Samsung claim they don’t collect any sensitive personally identifiable information but do collect ‘anonymized metadata.’ Independent audits are needed to verify whether all personal information truly remains local.
Privacy isn’t the only concern. AI is prone to hallucinations and misinformation, and local ‘narrow’ AI models may lack the context for comprehensive understanding. This could lead to inaccurate or misleading texts and images generated by these models.
Local AI models offer the benefit of working offline, but this can turn them into black boxes of unexplained decision-making. Conversely, relying on internet connectivity raises privacy concerns. As a result, most companies are playing it safe by offering basic features, which may have limited value. The future of generative AI on smartphones hinges on striking a balance between convenience, accuracy, and data privacy.
