Apple has hired dozens of Google artificial intelligence experts and created a secret European laboratory in Zurich. The tech giant is building a team that will compete with rivals to develop new AI models and product.
A recent analysis of LinkedIn profiles, public job listings and research papers revealed that the $2.7tn firm has been on a hiring spree to expand its AI and machine-learning team globally.
Since 2018, the iPhone maker has targeted employees from Google. It has hired at least 36 AI specialists from their rival.
The majority of Apple AI team members work out of offices in California and Seattle. However, the tech group also has a large outpost in Zurich.
Professor Luc Van Gool, from the Swiss university ETH Zurich, said Apple’s acquisitions two local AI start ups — virtual-reality group FaceShift and Fashwell image recognition company — led Apple build a “Vision Lab” research lab in the city.
Apple has been working with Zurich employees to develop the technology behind products like OpenAI’s ChatGPT Chatbot. Their papers focused on more advanced AI models which incorporate text and visual inputs in order to respond to queries.
The company is advertising generative AI jobs at two Zurich locations, with one being a low-profile location. One neighbour said they had no idea the office existed. Apple has not responded to requests for comment.
Apple is notoriously secretive about its AI plans, even though rivals Microsoft and Google have announced multi-billion dollar investments in this cutting-edge technology.
Since the beginning of the year its shares have fallen, while those of rivals have increased. This has put pressure on the technology giant to introduce AI features that can boost sales.
Apple, according to industry insiders, is focusing on deploying generative AI for its mobile devices. This would enable AI chatbots and applications to run on their own hardware and software instead of being powered by cloud services.
Apple’s chief executive Tim Cook told analysts that Apple has been “doing research across a range of AI technologies”, and investing in and innovating around this new technology “responsibly”.
The tech group has been developing AI products, including its voice assistant Siri, for over a decade. The company has been aware of ” neuro networks”, a form AI that is inspired by how neurons interact in the brain, for many years. This technology underpins products like ChatGPT.
Chuck Wooters is an expert on conversational AI, large language models, and speech recognition. He joined Apple in December 2013, and worked with Siri for nearly two years. They were big advocates of neural network technology even back then, when large language models weren’t popular.
Apple’s interest in neural networks, which are the basis of AI models, led them to hire researchers.
Apple acquired Perceptual Machines in 2016, a company founded Ruslan Salkhutdinov, and two of his Carnegie Mellon University students, who worked on generative AI image detection.
Around that time, they were trying to build an infrastructure for training the models and hunting down many researchers.
Salakhutdinov, a key figure of the history of neural network, studied at the University of Toronto with the “godfather” of this technology, Geoffrey Hinton. , who quit Google in 2014, cited concerns over the dangers of AI generative. Salakhutdinov was Apple’s director of AI Research until 2020. He then returned to Carnegie Mellon.
Apple’s AI team now includes former Google key figures, such as Giannandrea who was the head of Google Brain, Google’s AI laboratory, which has been merged into DeepMind.
Samy Bengio is a senior director for AI and ML Research at Apple. He was formerly a top AI scientist at Google. Ruoming Pan, who heads Apple’s team of “Foundation Models”, which is working on LLMs (Learning Learning Models), was Google’s AI research leader.
In 2022, Goodfellow returned to Google to protest against Apple’s policy of returning to work.
Apple published a research paper in March that listed six former Google employees who were hired in the last two years as authors. The paper revealed Apple had developed a “MM1” family of AI models which use visual and text inputs to generate response.
Apple also purchased about two dozen AI start ups over the last 10 years. These companies were focused on applying AI reasoning for image and video recognition and data processing. They also had search capabilities, music curation, and data processing.
LinkedIn profiles show that founders of Musicmetric, Emotient and Silk Labs are still working at Apple.
Salakhutdinov stated that Apple was focused on “doing as much as possible on the device”. This will lead to the need for powerful chips, with so-called Dynamic Random Access Memory (DRAM) which can handle the huge amount of data needed to power AI models.
“The next big thing will be AI smartphones — and they’ll require a lot of Dram,” said Sumit Sadana, executive vice president and chief business office of Micron Technology.
Sadana said that today’s smartphones have less memory than is needed to run a LLM.
Salakhutdinov added that language models’ tendency to give incorrect or problematic responses was another reason why Apple had been slow in releasing AI. He said, “I believe they are being more cautious as they cannot release something that they can not fully control.”
Apple’s generative AI capabilities may be first seen at its Worldwide Developers Conference, which takes place in June.
Erik Woodring, a Morgan Stanley analyst, said that the next iPhone could “become much more of a smart, voice-activated personal assistant led by an updated Siri, which, for example, could interact with all apps on your smartphone through voice control”.
He said: “What I’m looking forward to at WWDC is a preview of one or two AI capabilities that could be game changers for consumers.”