There are two schools of thought about the market impact of Big Data and artificial intelligence, although, paradoxically, both may be simultaneously true.
The first says that only the biggest tech companies know how to swim in the deep end of AI. To build state of the art models requires massive amounts of data, computational power, expertise and capital. Only the likes of Google and Microsoft in the US and Alibaba and Tencent in China can ever hope to make a splash.
The second argues that AI, like almost every other modern technology, will eventually become “democratised”. The tech companies are themselves empowering others to access powerful AI tools, such as Facebook’s PyTorch and Google’s TensorFlow, and cloud computing. Researchers are becoming better at building smarter models with less data. Traditional companies with specific industry expertise and innovative start-ups with entrepreneurial vim can also dive into the AI pool.
Policymakers in the US and Europe have been spending a lot of time recently thinking about how to break the grip of Big Tech. That may indeed be necessary to create a freer marketplace. But they should also be spending more time thinking about how to open up access to data and AI to encourage a broader range of competitors to emerge.
One intriguing way to do just that is being championed by the Stanford Institute for Human-Centered Artificial Intelligence, led by John Etchemendy and Fei-Fei Li. Their idea is to build a National Research Cloud, a three-way collaboration between federal government, private enterprise and research universities, to provide affordable access to powerful computational resources and public data sets in a secure cloud environment.
The aim of the research cloud is to enable more academic and industry researchers to work at the leading edge of AI, help train a new generation of experts in the most sophisticated techniques and — with luck — deliver breakthroughs in highly complex areas, such as climate change and healthcare.
Prof Etchemendy says that some cutting-edge AI research is currently inaccessible to universities because of the cost. He points to the example of GPT-3, the deep learning language prediction model developed by the Microsoft-backed Open AI research company, which machine-read all text accessible on the web, at an estimated cost of $12m.
Remarkably, given the current political environment, Stanford’s public cloud initiative has won bipartisan support in both houses of Congress and is likely to be added to the National Defense Authorization Act after November’s election. This would authorise a task force to come up with specific proposals to develop the research cloud, which has also been backed by 22 universities and several companies. The federal government would then inject public data sets and subsidise researchers to access the cloud, layered on top of existing industry infrastructure. “I can see this being funded at anywhere from $1bn to $10bn a year,” says Prof Etchemendy. “This will benefit the US and democratise the technology, whether for industry or academia.”
Several European countries and companies are attempting to develop their own cloud computing infrastructure, known as Gaia-X, to reduce their dependence on the powerful US tech groups. The EU’s draft Digital Services Act also envisages forcing the big US companies to share some of their data.
But Willem Jonker, chief executive of EIT Digital, an innovation institute, says too much of Europe’s policy is geared towards the transition of established companies rather than the creation of new ones. There is little point in creating a level playing field in Europe if that field is still dominated by big US tech companies. “Europe has to create the makers,” he says. “Digital sovereignty is about having a choice.”
A public research cloud could be a great way to stimulate such innovation. The US proposal is such a good one that Europe should build something similar itself.
Get alerts on Technology sector when a new story is published