AI impact: DeepMind director outlines potential of artificial intelligence

“AI is happening today. It is reality.”

That was the view of DeepMind chief business officer Colin Murdoch, as he shared examples of how AI is impacting the world as part of a presentation at the AI Summit in London.

DeepMind – the British AI firm founded in 2010 – is owned by Google-parent Alphabet Inc. and, according to Murdoch, the company has already seen many “amazing” breakthroughs in fundamental AI research.

But, he explained to delegates during a speech at London’s Tobacco Dock’s back in June, research isn’t the most important factor of AI – application is.

“AI isn’t hype, it has the potential to enrich the lives of billions of people, he added in his ‘Translating AI research and innovation into impact’ presentation, “We’ve already seen amazing breakthroughs in fundamental AI research but that’s not everything because AI is built to be applied in the real world.

A key question for Murdoch is around maximising the benefits of AI in the real world in a way that is safe, ethical and inclusive.

“I believe that AI is one of our most promising technologies. In fact, when we look back years from now, I think we’ll look back and see that AI was one of the most transformative technologies in the history of humanity,” he said.

Murdoch, also director of Applied AI at DeepMind, oversees a wide-range of teams including DeepMind for Google, which applies research breakthroughs to Google products and infrastructure used by billions of people. He also helps drive the growth of DeepMind, building and leading critical functions including finance and strategy and leading external and commercial partnerships.

“Bleeding edge AI has already demonstrated huge potential in really exciting areas like nuclear fusion, quantum chemistry and many others. We’re applying it to live products and infrastructure today.”

Early research at DeepMind focused on conquering the games such as Chess, Go and Atari because “games provide an amazing testbed to develop our algorithms”. However, Murdoch said that these algorithms are now being applied to Nobel Prize-level problems in science and society, such as nuclear fusion, quantum chemistry, digital biology, protein folding etc., “It really is quite incredible how far this has advanced,” he declared.

Murdoch shared four examples with delegates how AI is impacting the world today.

Data Centres

With more and more compute moving to the cloud, it’s very important to keep them cool. They’ve had amazing results in Google’s data centres, “reducing the energy required to keep them cool by 40%”.

“As you can imagine, these are systems that have already been highly hand-tuned by human experts,” explained Murdoch. “Climate change is an immensely complex challenge, but you can just imagine if this sort of technology could be rolled out to all the world’s data centres and infrastructure what sort of impact that might have.”

<< Read more here >>>

Video Compression

The MuZero algorithm was originally developed to play the games of chess and Go. However, DeepMind asked it “to play the game of YouTube video compression to gain intuition”. This reinforcement learning technique that works very well is also adaptable for video as video in essence is a sequence of individual pictures. Video accounts for a huge amount of overall internet traffic and that only continues to grow. “YouTube alone streams over 700,000 hours every minute of video. That’s literally a lifetime,” Murdoch explained.

This example showcases how DeepMind’s MuZero AI has entered the real world. “A small improvement in even video compression quickly scales up to have a really positive impact on the global data infrastructure that underpins our digital society today,” the DeepMind executive added.

<< Read more here >>>

WaveNet

WaveNet is DeepMind’s AI-powered speech-to-text technology.

“Tim Shaw was an amazing [American] football player and amazing campaigner for research but he very sadly lost his voice to the devastating disease ALS,” also known as motor neuron disease (MND), explained Murdoch.

However, by using the WaveNet AI-powered technology, DeepMind weren’t just able to give him a voice back, they gave him his own voice back. Having conducted many interviews to the media and with his fellow teammates on camera after games, DeepMind received a catalogue of videos of Tim’s voice. “With his permission and WaveNet we recreated his voice. An amazing demonstration of what AI can do,” hailed Murdoch.

<< Read more here >>>

AlphaFold

AlphaFold is a system that can predict the 3-Dimensional structure of proteins with unprecedented speed and accuracy.

“Proteins are the building blocks of life. If you can begin their 3D structure you can begin to get a better understanding of things from cancer to dementia and begin to understand how to create treatments for them,” added Murdoch.

“It used to take years of painstaking work and millions of dollars of equipment to map the structure of just one protein. We’re now mapping the structures of millions of proteins and we’re doing that in minutes per protein, which is outstanding.”

This update was followed up recently by the AI research lab announcing that it had cracked the structure of all known proteins, thanks to advances in the Alphabet-owned company’s AlphaFold Software.

AI is truly having an amazing impact on the world today. However, later in Murdoch’s presentation he stressed that there must be a greater focus on responsible and ethical deployments of AI. It’s important for companies like DeepMind to utilise “exceptional care” when unlocking the “exceptional promise” the technology can bring.

Talking about its deep-learning neural network, Murdoch explained how DeepMind took the decision to bring in external experts to review AlphaFold to have greater emphasis on responsibility.

Responsible AI is a huge area of concern but also governance framework for the ethical development and deployment of the technology. Developing AI responsibly in order to benefit people and society must be a core principle.

Also speaking at the AI Summit in London, Dr Nikos Loutas, head of Data and AI policy at NATO, stressed the importance of responsibility alongside industry collaboration and learning through use cases as the key pillars of NATO’s AI strategy.

AI is transforming all industries on a global scale and is a cutting-edge technology enabling us to address urgent yet unsolved problems in a broad range of fields. But, as Murdoch points out, we need to be responsible and ethical about its implementation to have a positive impact on our society and environment.