Google DeepMind calls for ‘responsible’ approach to AI usage

The chief business officer at Google’s artificial intelligence research lab says the world is having a “eureka moment” around artificial intelligence, but we have to be responsible with the technology.

The explosion of interest around AI has come from recent advances in the technology that have allowed people to use it with conversational language, rather than the programmers who predominantly dabbled with it before, said Colin Murdoch of Google’s DeepMind.

“It’s kind of all of a sudden been much more accessible because my mum and dad can do this,” he said in an interview with The Canadian Press.

“Anyone can do it.”

The surge in people and companies experimenting with AI was triggered by last year’s release of ChatGPT, a generative AI chatbot capable of humanlike conversations and tasks that was developed by San Francisco-based OpenAI.

The release kick-started an AI race with other top tech names including Google and its rival product Bard, putting an additional spotlight on DeepMind, which is headquartered in the U.K., but has space in Montreal and Toronto.

Now, everyone from health care companies to oil and gas firms and tech businesses are touting their use of or plans for AI.

But Murdoch said that ubiquity must be met with a careful approach and thoughtful consideration about all of the risks that AI carries.

“The way we think about this is being bold and responsible because it is a balance,” he said.

“What we want to make sure of is that we are doing this in a way that enables society to benefit from the incredible potential for this technology, but also the exceptional promise also does need exceptional care, which is why we have to act responsibly and why we have to pioneer responsibly.”

But what does responsible AI look like?

At Google, for starters, it’s meant being open to criticism at every step of the AI development process.

The company relies on internal and external review committees from the day an idea is generated to when it is unleashed for public use, Murdoch said.

“We’re making sure that we have the right oversight of our work, so, for example, we have ethicists sitting alongside policy experts sitting alongside machine learning experts,” he said.

Sometimes they prod staff to talk to even more external experts about ramifications, like when they were building AlphaFold and 30 people ranging from biology experts to biosecurity professionals and farmers were consulted.

AlphaFold can predict 3D models of protein structures. Murdoch reckons the technology has mapped all 200 million proteins known to science, saving one billion years of research time in the process because it can determine the structure of a protein in minutes and sometimes even seconds rather than years.

It has been used by researchers at the University of Toronto to identify a drug target for liver cancer.

Aside from ensuring products involve external reviews, Murdoch said responsible AI also takes bias into account. Many say bias crops up in AI because of a lack of diversity and opinions in the building and training phase.