Google to make artificial intelligence more people-friendly
Google is grappling with how to make sure artificial intelligence (AI) benefits humans more than it hinders them, as the powerful new technology starts to permeate all of its services.
The Internet Company on Monday, announced a research initiative aimed at finding ways to put people at the centre of AI development, from giving engineers better tools to control the technology to making sure ordinary people are not disadvantaged by decisions coming from intelligent machines.
The idea “grew out of a set of questions that started to emerge in various corners of Google about human/AI interaction,” said Fernanda Viégas, one of the researchers behind the plan.
Google’s growing awareness of the potential downside of AI reflects a broader concern in the tech world that the methods companies have used to build new products in the past might not be adequate to control the new technology. Incomplete or distorted data used to train AI systems, for instance, could result in biased “thinking” by the machines, and it is difficult to subject the algorithms they employ to outside scrutiny.
Even when AI systems work effectively, they risk embedding views of the world that put some groups of people at a disadvantage, or they can sideline humans altogether rather than encouraging new forms of computer interaction that give people a role.
The aim should be to create “people-centric AI systems” and “democratise” the technology by making it more understandable to engineers and subjects alike, Viégas and fellow Google research scientist, Martin Wattenberg, wrote in a blog post announcing the research initiative on Monday.
The Google call follows a concerted attempt by Microsoft last year to put a more humanistic approach at the centre of its own AI development. Chief Executive Satya Nadella, compared the urgency to Microsoft’s overhaul of its development processes 15 years ago after the first wave of computer viruses exposed the need for more robust software.
Google’s drive is led by two senior research scientists from Google Brain, the name given to the first research project at the company to use deep learning — the most advanced form of the machine learning technology that is behind the current advances in AI.
Most AI research so far has been narrowly focused on “technical performance”, the two wrote. They called instead for a new development process that starts with end users in mind — a technique known as design thinking, a popular approach to product development that spread had to many industries.
“How can we make sure this technology benefits and empowers everyone?” the authors asked.Their call echoes a growing concern that the difficulty of explaining how AI systems work could lead to a lack of trust in their results, holding back their use. In one attempt to overcome this problem, the U.S. defence department’s research arm has backed studies into how to make the “thinking” of AI systems more intelligible to humans.