Sajid Javid, the former British chancellor, warned that Britain’s £900mn supercomputer plan will be far too modest to keep up with rivals and leave Britain behind.
Javid said at the London Tech Week that “billions of dollars” should be invested to build a machine with 10 times the computing power of the “exascale”, which was promised by Jeremy Hunt in his March budget.
Hunt announced £900mn of funds to implement recommendations of a independent “Future of Compute Review” for the supercomputer. He said that “AI requires computer horsepower”.
Javid said that the plans of the government would only result in a facility by 2026 with the same level of power as OpenAI’s ChatGPT did in 2022. This would leave Britain behind rivals like the US.
He said, “It is nowhere near where we should be.” “The investment can’t be in hundreds of millions, it will have to be billions.” “I don’t believe we have a choice.”
Javid, former adviser of US artificial intelligence company C3 AI said that speed was essential and the project had to be launched much faster.
He said that although Rishi Sunak was aware of the issues – the Prime Minister is hosting the first global AI Summit in Britain this autumn – the Whitehall system had been “worsely unprepared” for AI revolution.
Sir Tony Blair and Lord William Hague warned in a recent report that the UK could never catch up if it “doesn’t up its game”.
Exascale computers are highly specialised machines, requiring tens or thousands of processors as well as intensive cooling systems. These computers will have the ability to drastically reduce the amount of time needed to process large amounts of information, or even more complex data. They promise scientific breakthroughs, such as in biology, climate change, and security.
The UK’s supercomputer project is comparable to those in the US and EU, but the resources of Silicon Valley’s wealthiest companies pale in comparison.
OpenAI received US$1bn in 2019 from Microsoft to develop the “large-language model” (LLM), which is at the core of ChatGPT, a breakthrough chatbot launched in November 2018 to much acclaim.
The massive amount of data used to power chatbots, and other “generative AI”, systems that can create human-like text and images with the help of LLMs requires a lot of computing resources.
Microsoft has invested several billions of dollars in OpenAI this year. In March, the AI model GPT-4 was upgraded, resulting in a significant improvement in ChatGPT’s response quality.
In 2018, the US allocated up to $1.8bn towards two exascale-capable supercomputers. The first fruit of this investment was the Frontier supercomputer, located at Oak Ridge National Laboratory, a US Department of Energy facility in Tennessee. It became the first computer to surpass the exascale threshold.
It achieved performance levels of more than 1 quintillion calculation per second, or 1.1 exaflops.
The European Commission announced last year that Jupiter, Europe’s very first exascale supercomputer, would be hosted by the Julich Supercomputing Centre, in Germany. Jupiter is due to be launched later this year.
The Commission allocated a budget for Jupiter of up to € 500mn, which was jointly funded by the German and Belgian governments.
For a comment, the Department of Science, Innovation and Technology has been contacted.
Gillian Keegan will launch on Wednesday a request for evidence regarding the opportunities and dangers of artificial intelligence (AI) in education.
Keegan, in an address at London Tech Week, will solicit views from schools, colleges, and universities about the benefits, risks and ethical considerations, as well as necessary training. The evidence call will last until August 23.
AI, according to the government, can be used by academics and teachers to improve their results and reduce workloads. Ministers are concerned that the technology could be misused to write essays or cheat in exams.
Several institutions have banned ChatGPT, but it can be difficult to enforce because it’s not always obvious when the technology is being used.
In April, the student newspaper Varsity from Cambridge found that almost half of students used ChatGPT for their studies.