by Lisa Riste, Research Fellow, NIHR Greater Manchester PSTRC, The University of Manchester
I never thought I’d enjoy ‘being sent to Coventry’ quite so much, but from 18-22 February 2019, the city was home to the UK’s first Citizens’ Jury on Artificial Intelligence (AI).
The process was the first of two weeks of juries commissioned by National Institute for Health Research (NIHR) Greater Manchester Patient Safety Translational Research Centre (PSTRC) and the Information Commissioners Office, led by Citizens Juries C.I.C. using methods developed by the Jefferson Center in the USA. It was designed to help guide future policy around what people would want to know about how decisions are made by AI and what trade-offs people are prepared to accept between accuracy and explain-ability.
The jury comprised 18 members aged between 20 and 70, whose occupations ranged from the arts, hospitality, banking, health and tourism, who had been selected to reflect the demographics of the local population. After learning about the various types of AI systems, the jurors were asked to consider evidence from expert witnesses around four scenarios:
- a system that diagnosed strokes
- a recruitment system that filtered job applications and decided who should be interviewed
- a system that determined whether patients awaiting kidney transplant were likely to be a match or not
- a criminal justice system that could potentially filter people for rehabilitation rather than a court trial.
My poem describes the deliberative process that I was privileged as a researcher to go and observe.
Artificial Intelligence Citizens Juries
Monday morning came, not sure what to expect
Just a venue and a time slot, the rest was all a guess
Eighteen citizens jurors, had been selected to attend
Responded to the adverts, looked an interesting event
Jurors were ‘curious’, but ‘all in the same boat’
Learnt the AI jargon, and the process of how to vote
They heard about the funders, and their interests in it all
Why they were invested, and data protection law
Impact and explanation post-its, were arranged upon the wall
Pregnancy, trains and dentists, similar rules applied to all
Then lots about AI, and systems A and B and C
How they process data, verses its explain-ability
A week of consideration, the impact and how and why
Are performance and accuracy trade- offs? That they must decide
Then came the best bits, scenarios to debate
Considering evidence from experts, determining AI’s fate
For the health examples, they all were pretty clear
For measurable data, explanation didn’t have a need
Seems they’d virtually all trust, speed and accuracy of AI
Especially in medical emergencies, where it potentially can save lives
But recruitment and rehabilitation, were both trickier to call
Humans were still needed, where data could be flawed
Seems rationale was important, these decisions had implications
And humans need to trust the intelligence, of this next AI generation.
Leave a comment