Apple researchers have developed a new technique that could allow advanced AI capabilities to run directly on iPhones and other Apple devices, without needing to connect to the internet. This breakthrough has the potential to bring sophisticated features like chatbots and vastly improved voice assistants to Apple’s products while maintaining user privacy.
New Method Enables On-Device AI
The new approach, detailed in a research paper published this week, focuses on running neural networks – the algorithms behind AI systems – more efficiently. Specifically, it targets large language models (LLMs) like the popular ChatGPT chatbot. LLMs are powerful but require substantial computing resources.
By using flash storage instead of RAM, Apple’s method reduced the memory needed to run an LLM by up to 13x. This could soon make it feasible to run advanced LLMs directly on consumer devices like iPhones.
“Our method allows fitting large neural network models into small devices, enabling on-device AI,” the researchers wrote.
| | Standard Approach | Apple’s New Method
|-|—————-|—————–
|Memory Needed | Large RAM capacity | Uses flash storage instead
|Efficiency | Inefficient | Up to 13x more efficient
|Privacy | Requires internet access | Runs on device
The breakthrough works by splitting a neural network across flash storage and RAM, minimizing the amount of RAM needed at any one time while preventing major slowdowns.
“This provides the best of both worlds: low memory usage as with [flash], and fast inference speeds as with RAM,” the paper explained.
On-Device AI Enables New Features, Protects Privacy
Running advanced AI on iPhones rather than in the cloud has major privacy advantages. The data remains on device rather than being sent to internet servers. This approach aligns with Apple’s strong stance on privacy.
Local AI processing also enables features that wouldn’t be possible otherwise. With an LLM running on the iPhone, apps could access AI capabilities like chat and language understanding even without an internet connection.
“We believe firmly that privacy is a fundamental human right and that users should have transparency and controls over how their data is used,” Apple’s software engineering chief Craig Federighi said last year.
Apple’s commitment to on-device AI also sets it apart from other tech giants like Google and Amazon, which rely heavily on cloud-based services.
What Could This Mean for Future Apple Products?
The new technique paves the way for substantial AI upgrades in Apple products over the next several years. The company will likely roll out enhanced features powered by on-device LLMs gradually.
Siri and voice assistants. Apple’s voice assistant Siri could be transformed with conversational abilities powered by an AI chatbot running locally. This would enable smoother interactions and understand more context compared to today’s Siri.
Faster app experiences. Apps could tap into on-device LLMs to boost relevance and improve predictions. For example, the Photos app could automatically label people and objects in images faster.
New intelligent apps. Sophisticated AI abilities like summarizing long documents or answering complex questions could be offered in standalone iPhone apps in the future. For users needing internet connectivity, these features wouldn’t be disrupted.
And this is just the beginning. Apple is investing heavily in its AI and machine learning teams, aggressively recruiting talent over the past few years. As the researchers note, specialized hardware like Apple’s latest mobile chip line with dedicated neural processing units “should provide an order-of-magnitude increase in capability” for advanced on-device AI.
What’s Next for Apple’s AI Efforts
Withfoundational research like this breakthrough, Apple is poised to emerge as a leader in privacy-preserving AI. Competitors are taking notice – a new report indicates Google has scrambled together an “Apple Intelligence Task Force” to assess the iPhone maker’s progress.
We likely won’t see the results immediately, but Apple is playing a long game. The company aims to control the key technologies powering its products rather than relying on third parties. As AI becomes further ingrained into mobile devices and computers, Apple wants its hardware running the latest machine learning models, not cloud services.
“[Apple] has been quiet on AI so far. Now it is showing some of its hand,” an analyst told The Financial Times.
With foundational research progressing well, Apple’s senior vice president of AI strategy and Apple Distinguished Scientist Injia Ye hinted that “2023 will be the year AI comes to the iPhone.”
Rumors indicate some initial enhancements powered by on-device AI could emerge in the iPhone 15 line expected next September. But the full vision – with transformative apps and features enabled by LLMs running locally on Apple hardware – might not arrive until 2025’s iPhone 17.
Whenever the rollout begins in earnest, Apple seems prepared to differentiate itself as the AI leader focused on protecting user privacy – not harvesting more data. This long-term strategic direction aligns closely with the company’s values.
While competition in AI innovation accelerates, Apple is taking a leap ahead with this breakthrough work. By running advanced models directly on devices, the company can deliver new intelligent features while putting user privacy first. For consumers, it means iPhones and Macs that protect privacy while getting smarter every year. And for Apple, it’s the ultimate win-win.
To err is human, but AI does it too. Whilst factual data is used in the production of these articles, the content is written entirely by AI. Double check any facts you intend to rely on with another source.
