Designing Australian values into artificial intelligence systems will be the focus of a new research project launched at The Australian National University (ANU), in a keynote featuring Professor Shannon Vallor of the University of Santa Clara and Google.
The 'Humanising Machine Intelligence' project brings together philosophers, social scientists and computer science experts to design AI that represents and promotes "Australian values".
The project, which will receive over $1.5 million in ANU funding each year for at least the next three years, is led by philosopher Associate Professor Seth Lazar.
Associate Professor Lazar, the head of ANU School of Philosophy, noted AI now underpins all aspects of the economy, from smartphones and search to recruitment, credit scores, and high finance. At the same time, all branches of the Australian public service are looking at how AI can help deliver services more efficiently and fairly.
"As we increasingly come to rely on systems like these it is crucial to get our values right from the start-not wait for a process of trial and error, with potentially catastrophic results," Associate Professor Lazar said.
"Machine intelligence has the power to realise incredible positive change. But it can also replicate the social injustice embedded in the data on which it is trained.
"AI is a decision technology. It has a 'picture' of the world, and a set of goals to achieve. We have to make sure it sees the world in a way that we endorse, and that its goals reflect our priorities.
"This means making fundamental progress not only in machine learning and other areas of AI, but also in philosophy and social science. To design our values into AI systems, we have to understand them ourselves, and then represent them in a way that a computer can act upon.
"Beyond designing AI systems themselves, this is also about shaping the governance structures that make moral machine intelligence not just possible, but probable."
Associate Professor Lazar said that, because AI cannot be morally neutral, we have to decide what values we want to build into it. As a democracy, this is something that all Australians need to consider.
"When we say that we want to shape machine intelligence around Australian values, we're not presupposing what those values are, or that there is one set of 'Australian' values.
"In fact, the opposite is true. Australia, like any society, is filled with moral disagreement. And we settle that disagreement through the democratic process-not just by voting, but through deliberation, public discussion, press scrutiny, and review.
"Machine intelligence is no different: the values we build in should be democratically legitimate."
The Humanising Machine Intelligence project is funded under the ANU Grand Challenges scheme which brings ANU researchers from different disciplines together to solve the most pressing challenges facing the world today.
Learn more about the project at https://hmi.anu.edu.au/.