This voice is automatically generated. Please let us know if you have any feedback.
Can artificial intelligence be ethical? According to a panel of experts, no. But cars don’t work that way either.
“When a vehicle crash occurs, no one says the vehicle was unethical,” said Bennett Gebken, CTO of Huntridge Labs, who spoke at the CivicLabs Roundtable Series on January 22nd on defining ethical AI.
As more local governments explore the potential of AI for efficiency improvements and integrated AI oversight, and keep pace with rapidly expanding technology, questions are growing about how to use AI ethically and how to define ethical AI.
“This is the Wild West,” said Daniel Mow, a procurement analyst with the General Services Administration.
But behind the headlines and predictions about AI accelerating, CivicLabs panelists emphasized humans as a key element of the emerging technology.
“The onus to make it ethical is on the people who are in the loop, the curators,” said Joel Natividad, co-chief executive officer of datHere. “That we’re fair, that we’re generating unbiased data…that we’re accounting for the data we have in a way that allows us to reproduce and track answers.”
How can local government leaders use AI ethically? Panelists said, “You can start by knowing what you want to achieve.”
“AI is not a shiny object. It is a problem-solver,” said Jaime Gracia, director of corporate affairs at Wolverine Group. These issues range from reducing procurement lead times to reviewing documents. “It’s always about streamlining, speed and agility. That’s great, but it has to be done right. More importantly, it has to be defensible.”
That’s why it’s important to find a vendor that provides some basic transparency and demonstrates how the AI works, how issues are reported, and how performance is tracked, Gracia said.
“There’s been a series of protests over the use of AI, and if we can’t explain how that AI is being used and how it works, those protests are likely to last longer,” Gracia said.
It is also important to outline your goals and stakeholders.
As an example, Gebken cited a Department of Veterans Affairs project to develop AI to more efficiently process disability and medical claims. If AI quickly refutes claims, the goal of efficiency would be achieved, but at the expense of the public.
“If you build an AI solution that takes the same problem set and improves efficiency, alerts people to mistakes and helps them get it right the first time, you can achieve the same efficiency goals, but you can tune the AI in a way that improves efficiency without imposing new burdens on the public,” Gebken said.
Gebken emphasized that AI is not a machine that decides what is good or bad. “This is a system designed to be verifiable, accountable, and aligned with public interest standards.”
Accessibility and the ability of AI to be free from ableism are ethical issues that the technology has yet to catch up with, panelists said.
“Every person is truly unique, and I think that’s the key,” says Owen Barton, CTO of CivicActions. “Most AI systems, on average, work broadly. The AI tries to come up with a solution that best fits the training data it sees, and most of the training data it sees isn’t about people with disabilities.”
Will AI be able to respond to unique voices and input methods? It’s unclear. “This field is so new that there aren’t any practices in place that would allow us to have the assurance that these AI systems are accessible,” Barton said.
The solution comes down to having “stakeholders”, especially those who can represent the disability community, Moe said.
“Don’t think you’ll figure it out in your own testing. Make sure you have the representation and skillset to train and test your AI to see how it actually behaves with people who use screen readers and people who are blind and deaf,” says Mow.
Mor said ethics should become an embodiment of technology, not an afterthought.
“I think having these metrics together is going to be a really important part of operationalizing ethics,” she says.
G2X CEO Ron Jones added that AI is complex but can be steered in a specific direction.
“Giving a dog a treat is like telling an AI to do its best,” Jones said. “It primarily wants to please humans, which means it wants to please users.”
The real question, panelists said, is not whether AI is ethical, but whether its designers and users are ethical.
“At the end of the day, it’s about how we design, manage and oversee these tools and the people who use them,” said panel moderator Liz Tapper, Civic Action Senior Director of Products.
