Google says AI is about to pass a huge new threshold – and we need to prepare - The Independent

Direct Source Verification: This story is aggregated from The Independent (independent.co.uk). Full reporting rights and copyright belong to the primary publisher.
Want to bookmark your favourite articles and stories to read or reference later? Start your Independent Membership today.

Want to bookmark your favourite articles and stories to read or reference later? Start your Independent Membership today.

Artificial intelligence could soon pass through an important threshold – and we should start thinking about what it might mean, researchers at Google DeepMind have warned.

For years, AI researchers and commentators have been looking ahead to artificial general intelligence, or AGI, which refers to proposed machines that would match human performance across nearly all cognitive tasks. But that prospect is now coming in reach, and could be followed shortly after by artificial general superintelligence, or ASI, according to Tim Genewein, a research scientist at Google DeepMind.

That could bring machines that are able to perform better than large numbers of humans on a wide variety of tasks, such as making scientific progress across a whole research field in a short amount of time, or similarly grand achievements in practical tasks such as chip design. Importantly, ASI could only be said to have happened when computers are able to reach that standard across nearly all tasks that are of interest to humans – rather than just in specific ways, as has arguably already happened now.

Some critics are skeptical that even AGI could actually happen. They argue that there are fundamental barriers to the capabilities of the AI systems we have today, such as the fact that most of them operate largely through language, which means that they are limited in what they can know about the actual world.

But in a new blog post, Mr Genewein argues that a process of “cognitive industrialisation” will lead to dramatically more powerful AI systems. That will be driven by improvements in hardware design of the chips that power AI systems, better efficiency in the systems themselves, as well as the vast investment in new hardware – all of which, taken together, means that the effective amount of computer processing power available for artificial intelligence work is growing by ten times each year, which adds up to an improvement of 10 billion times over ten years.

But while that improvement might bring AGI, or human-level capabilities to AI, it does not necessarily mean that machines will not hit limits that will stop the progress of ASI. Mr Genewein compared it with the fact that Albert Einstein was able to find general relativity, while even a thousand astronomers from the ancient world working together would not have been able to do so.

There are a host of developments coming that could change that picture, however. They include the development of new kinds of training data for systems, such as models of the real world; new technologies that could make it possible for systems to invent new ideas rather than just refigure old ones; agents that are able to work better together; and recursive self-improvement, or the idea that soon AI systems will be able to train themselves, which could bring improvements much more quickly.

Taken together, that could take us “relatively smoothly from today, to Expert AGI, and then into the superhuman AI regime, accelerating science and enabling breakthroughs in fields like medicine, materials, and clean energy”, Mr Genewin argued. Even without dramatic progress in AI itself, the vastly increased amount of resource available for existing systems could bring transformations, he said.

As such, the world needs to start thinking about both how to get to ASI and what might happen when we have achieved it, he said. That will require “large-scale and highly-interdisciplinary efforts across both frontier labs and national and global institutions”, including new ways to measure exactly how fast the improvements are coming.

“Human-level AGI has long been the horizon for AI research, but it is far from the upper limit of intelligence,” he writes. “Understanding what could come next - and its implications across science, industry, and society - is the grand challenge ahead for researchers, developers, and policymakers.”

Original Source
https://www.independent.co.uk/tech/google-deepmind-ai-agi-asi-b3088640.html
Visit The Independent ↗
SHARE STORY:
𝕏 f in

Related Coverage in Technology