Government R&D unit unveils £42m project to ‘redefine’ computing and cut costs of AI by 99.9%

ARIA, a public body specialising in ‘high risk, high reward’ STEM research, has revealed that its first programme aims to apply principles of the natural world to deliver simpler processing

The government’s high-end research-and-development agency has unveiled its first full programme, which aims to “redefine our current compute paradigm” and cut the costs of delivering artificial intelligence by 99.9%.

Plans to launch the Advanced Research and Invention Agency (ARIA) – a public body intended to specialise in “high risk, high reward” science, tech and engineering – were unveiled by government in early 2021. The unit, which operates as an arm’s-length body of the Department for Science, Innovation and Technology, was formally established in January of last year.

ARIA describes its operational model as being to “back people, then projects” and the body has appointed a range of programme directors that have “creative control as they define an opportunity space, set the direction of programmes, and decide how to best fund the R&D community to advance their vision”.

The unit’s first full programme – titled Scaling Compute – AI at 1/1000th the cost – will be led by director Suraj Bramhavar, who co-founded cloud-optimisation firm Sync Computing.

The project, which is supported by £42m of funding, is intended to address the issue that “for the first time in history, increased performance requires increasing costs and this coincides with an explosion of demand for more compute power driven by AI”.

As the “traditional methods for improving hardware performance have become economically unviable”, the thesis for the programme explains that the natural world may provide a solution to this problem.

According to the programme documentation: “We see an opportunity to draw inspiration from natural processing systems, which innately process complex information more efficiently (on several orders or magnitude) than today’s largest AI systems.

Related content

“Nature has evolved to process complex information in an entirely different way, and has proven incredibly efficient at doing so,” the thesis says. “Looking to nature provides us with an existence proof that it is fundamentally possible to accomplish sophisticated information processing much more efficiently than current computers. Key principles that distinguish natural systems from existing computers include the facts that: they typically do not distinguish computing elements from memory elements; they incorporate noise and are not purely binary; [and] they do not operate in discrete time intervals The field of neuromorphic computing was created in this vein, primarily adopting [the first] tenet to demonstrate impressive performance. One aim of this programme will be to show that embracing a combination of all three can take this work significantly further and help radically reduce the manufacturing and operational cost of today’s AI hardware.”

To bid for a piece of the programme’s funding, individuals or organisations have until 27 March to submit a short concept paper in one of three technical areas: bold solutions; bold ideas; and system-level software simulation.

ARIA will review these and then advise on whether a “full proposal would be competitive”. After this feedback period, bidders will have a window of about four weeks before formal proposals must be submitted by 7 May.

Former Downing Street adviser Dominic Cummings is understood to have been a significant driving force in the creation of ARIA, having taken inspiration from the US’s Defense Advanced Research Projects Agency (DARPA). During its near 70 years in operation, DARPA is credited with playing a key role in the development numerous technologies that are now part of everyday life – including the internet, GPS, voice-activated virtual assistants, and drones.

In the hope of replicating such success, ministers have committed to providing £800m to fund the delivery of ARIA’s programmes over the next few years.

Sam Trendall

Learn More →

Leave a Reply

Your email address will not be published. Required fields are marked *