Ethical Machines

We use the wrong metaphor for thinking about AI, Shannon Vallor argues, and bad thinking leads to bad results. We need to stop thinking about AI as being an agent or having a mind, and stop thinking of the human mind/brain as a kind of software/hardware configuration. All of this is misguided. Instead, we should think of AI as a mirror, reflecting our images in a sometimes helpful, sometimes distorted way. Our shifting to this new metaphor, she says, will lead us to better, and ethically better, AI.

Canada Air blamed the LLM chatbot for giving false information about their bereavement fare policy. They lost the law suit because of course it’s not the chatbot’s fault. But what would it take to hold chatbots responsible for what they say? That’s the topic of discussion with my guest, philosopher Emma Borg.

California just signed a bill to drastically decrease deepfakes on social media. The worry, of course, is that they are already being used to unjustifiably sway voters. In this episode, one of the best from Season 1, I talk to Dean Jackson and Jon Bateman, experts on the role of deepfakes in disinformation campaigns. The bottom line? Deepfakes aren’t great but they’re not half the problem.

What does it look like to integrate ethics into the teams that are building AI? How can we make ethics a practice and not a compliance checklist? In today’s episode I talk with Marc Steen, author of the book “Ethics for People Who Work in Tech,” who answers these questions and more.

Doesn’t the title say it all? This is for anyone who wants the very basics on what AI is, why it’s not intelligent, and why it doesn’t pose an existential threat to humanity. If you don’t know anything at all about AI and/or the nature of the mind/intelligence, don’t worry: we’re starting on the ground floor.

Are we dependent on social media in a way that erodes our autonomy? After all, platforms are designed to keep us hooked and to come back for more. And we don’t really know the law of the digital lands, since how the algorithms influence how we relate to each other online in unknown ways. Then again, don’t we bear a certain degree of personal responsibility for how we conduct ourselves, online or otherwise? What the right balance is and how we can encourage or require greater autonomy is our topic of discussion today.

From the best of season 1: I talk a lot about bias, black boxes, and privacy, but perhaps my focus is too narrow. In this conversation, Aimee and I discuss what she calls “sustainable AI.” We focus on the environmental impacts of AI, the ethical impacts of those environmental impacts, and who is paying the social cost of those who benefit from AI.

Is our collective approach to ensuring AI doesn’t go off the rails fundamentally misguided? Is our approach too narrow to get the job done? My guest, John Basl argues exactly that. We need to broaden our perspective, he says, and prioritize what he calls an “AI ethics ecosystem.” It’s a big lift, but without it it’s an even bigger problem.

Many researchers in AI think we should make AI capable of ethical inquiry. We can’t teach it all the ethical rules; that’s impossible. Instead, we should teach it to ethically reason, just as we do children. But my guest thinks this strategy makes a number of controversial assumptions, including how ethics works and what actually is right and wrong.

It’s common to hear we need new regulations to avoid the risks of AI (bias, privacy violations, manipulation, etc.). But my guest, Dean Ball, thinks this claim is too hastily made. In fact, he argues, we don’t need a new regulatory regime tailored to AI. If he’s right, then in a way that’s good news, since regulations are so notoriously difficult to push through. But he emphasizes we still need a robust governance response to the risks at hand. What are those responses? Have a listen and find out!