Artificial intelligence – and why we’re a very long way from building Skynet

Shannon Vallor arrives in Scotland from Silicon Valley in February – and she’s keen to bust some myths around AI


Professor Shannon Vallor has been hunting for sun lamps on- line. A born and bred Californian, she is preparing to depart the warm climes of America’s ‘golden state’ - for a new life in Edin- burgh. “It will be an adjustment,” she says. Helpfully, she’s yet to meet a single person for whom the charms of Auld Reekie have not more than compensated for the weather, which in February, when she arrives, will be a test. “I will come prepared with my sun lamps and my winter gear but in all seriousness I have never met a person who has spent any time living in Edinburgh who wasn’t absolutely enchanted with the city,” she says. Prof. Vallor is one of the US’s

best-known academics whose work on the ethics of data and artificial intelligence (AI) has helped some of the world’s big- gest tech companies – including Google, for whom she has been a visiting researcher and consulting AI ethicist – to ensure that their

technologies are rolled out in way that is in line with ethical prin- ciples underpinning them. Work- ing with the company for just over a year, she is understandably guarded about the detail of that work, but at the University of Santa Clara’s Markkula Center for Applied Ethics she has been influential in shaping the training courses that have been adapted by Google, particularly for its cloud service, and other large Silicon Valley companies.

What will be of interest, no doubt, is exactly how you figure out what those principles might be, but we start off by discussing her new role, which will be based at the Edinburgh Futures Institute – one of the five data driven inno- vation (DDI) centres attached to the University of Edinburgh, and which will come to play a large role in a £1.3bn City Region Deal economic and social stimulus programme – signed off by the Scottish and UK Governments. Her official title will be the first Baillie Gifford Chair in the Ethics


of Data and Artificial Intelligence, and she’s excited about the pros- pect of not only developing a new academic programme, but also working with the public sector as it grapples with how services might be deployed using AI, and how she can help local industry partners, in much the same man- ner she has worked with the tech companies in the US. “I’m thrilled and couldn’t be

more excited,” she says. “It’s a fantastic opportunity and one that I have been looking forward to for a couple of months now ever since the position was announced.” Te multi-disciplinarity of the research base at EFI has also enticed Prof. Vallor to the role: she will be working with a cohort of academics that span com- puter science, machine learning and AI, and combines the social sciences, arts and humanities, and is informed by statutory and legal frameworks, in a way that has perhaps not been done in the past, and is highly experimental. And it’s that kind of configuration that is designed to answer some

of the pressing societal questions that perhaps cannot be answered from within the resources from any one academic field. What comes out of that mix will in- evitably inform the way technol- ogy can perhaps be applied in new or existing industries – for example automation in medicine or financial services – but, for Vallor, the most important thing is that it’s grounded on an ethical framework.

Coming back to what that will look like, she gives a few ex- amples of where technology has gone wrong, which is perhaps the most helpful way of illustrat- ing the problem. She mentions a case in the US where AI technol- ogy has resulted in biases against certain racial and gender groups in the way bail was granted in a criminal justice setting, and also of a project at DeepMind – an AI company owned by Alphabet – where engineers developed an AI to make an earlier diagnosis of acute kidney disease. Te problem, however, was that the training algorithm relied on data from military veterans that were heavily skewed towards male subjects, therefore the AI was able was much more effective at predicting disease rates in men than in women.

Page 1  |  Page 2  |  Page 3  |  Page 4  |  Page 5  |  Page 6  |  Page 7  |  Page 8  |  Page 9  |  Page 10  |  Page 11  |  Page 12  |  Page 13  |  Page 14  |  Page 15  |  Page 16  |  Page 17  |  Page 18  |  Page 19  |  Page 20  |  Page 21  |  Page 22  |  Page 23  |  Page 24  |  Page 25  |  Page 26  |  Page 27  |  Page 28  |  Page 29  |  Page 30  |  Page 31  |  Page 32  |  Page 33  |  Page 34  |  Page 35  |  Page 36