Ethical Machines

Automation is great, right? It speeds up what needs to get done. But is that always a good thing? What about in the process of scientific discovery? Yes, AI can automate a lot of science by running thousands of virtual experiments and generating results - but is something lost in the process? My guest, Ramón Alvarado a professor of philosophy and a member of the Philosophy and Data Science Initiative at the University of Oregon, thinks something crucial is missing: serendipity. Many significant scientific discoveries occurred by happenstance. Penicillin, for instance, was discovered by Alexander Fleming who accidentally left a petri dish on a bench before going off for vacation. Exactly what is the scientific value of serendipity, how important is it, and how does AI potentially impinge on it? That’s today’s conversation.

Behind all those algorithms are the people who create them and embed them into our lives. How did they get that power? What should they do with it? What are their responsibilities? This and more with my guest Chris Wiggins, Chief Data Scientist at the New York Times, Associate Professor of Applied Mathematics at Columbia University, and author of the book “How Data Happened: A History from the Age of Reason to the Age of Algorithms”. Originally aired in season one.

People in the AI safety community are laboring under an illusion, perhaps even a self-deception, my guest argues. They think they can align AI with our values and control it so that the worst doesn’t happen. But that’s impossible. We can never know how AI will act in the wild any more than we can know how our children will act once they leave the house. Thus, we should never give more control to an AI than we would give an individual person. This is a fascinating discussion with Marcus Arvan, professor of philosophy at The University of Tampa and author of three books on ethics and political theory. You might just leave this conversation wondering if giving more and more capabilities to AI is a huge, potentially life-threatening mistake.

Developers are constantly testing how online users react to their designs. Will they stay longer on the site because of this shade of blue? Will they get depressed if we show them depressing social media posts? What happens if we intentionally mismatch people on our dating website? When it comes to shades of blue, perhaps that’s not a big deal. But when it comes to mental health and deceiving people? Now we’re in ethically choppy waters. 

My discussion today is with Cennydd Bowles, Managing Director of NowNext, where he helps organizations develop ethically sound products. He’s also the author of a book called “Future Ethics.” He argues that A/B testing on people is often ethically wrong and creates a culture among developers of a willingness to manipulate people. Great conversation ranging from the ethics of experimentation to marketing and even to capitalism.

There’s a picture in our heads that’s overly simplistic and the result is not thinking clearly about AI risks. Our simplistic picture is that a team develops AI and then it gets used. The truth, the more complex picture, is that 1000 hands touch that AI before it ever becomes a product. This means that risk identification and mitigation is spread across a very complex supply chain. My guest, Jason Stanley, is at the forefront of research and application when it comes to managing all this complexity

From the best of season 1: Microsoft recently announced an (alleged!) breakthrough in quantum computing. But what in the world is quantum computer, what can they do, and what are the potential ethical implications of this new powerful tech?

Brian and I discuss these issues and more. And don’t worry! No knowledge of physics required.

Every specialist in anything thinks they should have a seat at the AI ethics table. I’m usually skeptical. But psychologist Madeline Reinecke, Ph.D. did a great job defending her view that – you guessed it – psychologists should have a seat at the AI ethics table. 

Our conversation ranged from the role of psychologists in creating AI that supports healthy human relationships to when children start and stop attributing sentience to robots to loving relationships with AI to the threat of AI-induced self-absorption. I guess I need to have more psychologists on the show.

A fun format for this episode. In Part I, I talk about how I see agentic AI unfolding and what ethical, social, and political risks come with it. In part II, Eric Corriel, digital strategist at the School of Visual Arts and a close friend, tells me why he thinks I’m wrong. Debate ensues.

Jaahred Thomas is a VC friend of mine who wanted to talk about the evolving landscape of AI ethics in startups and business generally. So rather than have a normal conversation like people do, we made it an episode! Jaahred asks me a bunch of questions about AI ethics and startups, investors, Fortune 500 companies, and more, and I tell him the unvarnished truths about where corporate America is in the AI ethics journey and what startup founders should and shouldn’t spend their time doing.

From the best of season 1: The hospital faced an ethical question: should we deploy robots to help with elder care?

Let’s look at a standard list of AI ethics values: justice/fairness, privacy, transparency, accountability, explainability.

But as Ami points out in our conversation, that standard list doesn’t include a core value at the hospital: the value of caring.

And that’s one example of one of three objections to a view he calls “Principalism.” Principalism is the view that we do AI ethics best by first defining our AI ethics values or principles at that very abstract level. This objection is that the list will always be incomplete.

Given Ami’s expertise in ethics and experience as a clinical ethicist, it was insightful to see how he gets ethics done on the ground and his views on how organizations should approach ethics more generally.

From the best of season 1: Innovation is great…but hype is bad. Not only has all this talk of innovation not increased innovation, but it also creates a bad environment in which leaders can make reasoned judgments about where to devote resources. So says Lee Vinsel, professor in the Department of Science, Technology and Society at Virgina Tech, in this Ethical Machines episode.

ALSO! We want proactive regulations before the sh!t hits the fan, right? Not so fast, says Lee. Proactive regulations presuppose we’re good at predicting how technologies will be applied, and we have a terrible track record on that front. Perhaps reactive regs are appropriate (and we need to focus on making a more agile government).

Super interesting conversation that will push you to think differently about innovation and what appropriate regulation looks like.

“Sustainability,” “purpose/mission/value driven”, “human-centric design.” These are terms companies use so they don’t have to say “ethics.” My contention is that this is bad for business and bad for society at large. Our world, corporate and otherwise, is confronted with a growing mountain of ethical problems, spurred on by technologies that bring us fresh new ways of realizing our familiar ethical nightmares. These issues do not disappear via semantic legerdemain. We need to name our problems accurately if we are to address them effectively.

Democracy is about how we ought to distribute power in society and, more specifically, it’s the claim that people ought to have a significant say in how they are ruled. So if we’re talking about AI’s impact on democracy, we should focus on how our use of AI interact with our concern that democracy be respected, upheld, and improved.  

But my guest Ted Lechterman, UNESCO Chair in AI Ethics and Governance at IE University’s School of Humanities, argues that our current discussion on AI and democracy is far too narrow. We need to widen our aperature if we’re going to discuss and ultimately address the issues that really matter.

From the best of season 1.

Well, I didn’t see this coming. Talking about legal and philosophical conceptions of copyright turns out to be intellectually fascinating and challenging. It involves not only concepts about property and theft, but also about personhood and invasiveness. Could it be that training AI with author/artist work violates their self?

I talked with Darren Hick about all this, who wrote a few books on the topic. I definitely didn’t think he was going to bring up Hegel.

My guest and I have been doing AI governance for businesses for a combined 17+years. We started way before genAI was a big thing. But I’d say I’m more a qualitative guy and he’s more quant. Nick Elprin is the CEO of an AI governance software company, after all. How has AI ethics or AI governance evolved over that time and what does cutting edge governance look like? Perhaps you’re about to find out…

People, especially kids under 18, are forming emotional attachments with AI chatbots. At a minimum, this is…weird. Is it also unethical? Does it harm users? Is it, as my guest Robert Mahari argues, an affront to human dignity? Have a listen and find out.

You might want more online content moderation so insane conspiracy theories don’t flourish. Sex slaves in Democrat pizza shops, climate change is a hoax, and so on. But is it irrational to believe these things? Is content moderation - whether in the form of censoring or labelling something as false - the morally right and/or effective strategy? In this discussion Neil Levy and I go back to basics about what it is to be rational and how that helps us answer our questions. Neil’s fascinating answer in a nutshell: they’re not irrational and content moderation isn’t a good strategy. This is, I have to say, great stuff. Enjoy!

From the best of season 1. Part 2 of my conversation with Alex.

There’s good reason to think AI doesn’t understand anything. It’s just moving around words according to mathematical rules, predicting the words that come next. But in this episode, philosopher Alex Grzankowski argues that AI may not understand what it’s saying but it does understand language. In this episode we do a deep dive into the nature of human and AI understanding, ending with strategies for how AI researchers could pursue AI that has genuine understanding of the world.

From the best of season 1. Part 1 of my conversation with Alex Grzankowski.

It looks like ChatGPT understands what you’re asking. It looks like ChatGPT understands what it’s saying in reply. 

But that’s not the case.

Alex and I discuss what understanding is, for both people and machines and what it would take for a machine to understand what it’s saying.

One person driving one car creates a negligible amount of pollution. The problem arises when we have lots of people driving cars. Might this kind of issue arise with AI use as well? What if everyone uses the same hiring or lending or diagnostic algorithm? My guest, Kathleen Creel, argues that this is bad for society and bad for the companies using these algorithms. The solution, in broad strokes, is to introduce randomness into the AI system. But is this a good idea? If so, do we need regulation to pull it off? This and more on today’s episode.