91��Ƶ

In what feels like no time at all, consumer-grade generative AI has rewired the most fundamental aspects of life and work. From fueling scientific discovery to automating how we plan our holidays, artificial intelligence is everywhere.

But AI innovations have also been accompanied by a fresh flurry of concerns. Gone are the days when the biggest worry was job losses. Today, the very people building AI agents—leaders of companies like Anthropic and OpenAI—also warn of their .

91��Ƶ has long been a leader in advocating for safe and ethical deployment of AI tools. On September 25, 91��Ƶ’s latest Actionable AI conference will bring together industry titans from Google, Zoom, Salesforce and others to explore how AI can be used for good.

In advance of that conversation, 91��Ƶ’s , PhD., a leading global thinker on the ethics of AI development, offers a nuanced assessment of the ethical challenges presented by AI, and why he remains bullish on technology for technology’s sake.

An underlying theme of 91��Ƶ’s Actionable AI Conference series is the ethical considerations of the technology. How do you define AI ethics?

AI ethics is the investigation of the values that surround human interaction with artificial intelligence. But what does it mean in practice?

Value in ethical terms means something that is worth having on its own. Think of the slogan on the New Hampshire license plate—“Live Free or Die.” The idea with that motto is that freedom is worth having—not to get something else, but simply to have freedom. Without it, life isn’t worth living.

Privacy takes on a different urgency with AI.

AI ethics is simply the pursuit and the investigation of those values that surround artificial intelligence.

What makes AI ethics different from other kinds of ethics is that some values have greater importance than others. Take privacy, for example. In the world of automobiles, privacy isn’t really a big deal. No one is too afraid that people are looking into your car window as you drive by. But privacy takes on a different urgency with AI, because so much personal data can be gathered about people and then used to form models of those people and predict what they will do.

Based on that definition, we might expect regulatory agencies to be active in the discussion of AI ethics. But in the United States, at least, they are suspiciously absent, leaving it to private companies to define the safety parameters. That’s like the fox watching the henhouse.

Until recently, we've essentially trusted consumers to take care of themselves, and that's because artificial intelligence had not penetrated very deeply into our society. But that’s changing quickly.

Our collective social response was built for the industrial age, and it’s unlikely …to work for the AI age.

One area to watch is healthcare. The same tools that allow students to compose essays for university classrooms in seconds have the same technological power to compose profiles of patients and their health histories, which will, very shortly, enable predictions of the maladies people will suffer, and even when they might suffer them. That will raise lots of questions for things like insurance.

Here’s an example of that conundrum which, for obvious reasons, I hope is fictional: What if AI predicts with great certainty that not only am I going to have a heart attack in five to seven years, but that I will then have brain cancer in 11 to 12 years. Will insurance over the cost of my healthcare for a heart attack if I'm going to recover just in time to fall victim to predicted brain cancer?

These are the kinds of questions we’ve never had to ask before but are going to have to start asking them soon.

Can regulation of AI even work in these contexts?

I'm skeptical, and here’s why. The idea of political regulation, as we’ve developed it, is basically set for the industrial economy. It’s an area where things move slowly: problems develop, witnesses report those problems, political figures convene hearings to talk about the problems, a voting process ensues, and finally, a measure is signed and promulgated. It takes years to pass any kind of regulatory measure.

That way of organizing our collective social response was built for the industrial age, and it’s unlikely it is going to work for the AI age.

Clearly, though, societies must find ways to address the new types of questions arising around artificial intelligence, right?

Absolutely, but there are several schools of thought on how. On one side are the precautionists who believe we should slow down and ensure AI tools are safe before we deploy or push ahead with innovation.

Then there are the accelerationists. They believe it’s best to push forward with innovation and allow innovation itself to solve the safety problems that previous innovation has caused. I subscribe to this school of thought. My answer to the precautionist’s approach is that if we slow the technology down, we're missing opportunities to develop tools that could protect us from the risk that AI itself presents.

Let me give you an example. One of my was about a Canadian telecommunications company, called Telus, and how they managed the risks associated with a new client-serving AI chatbot. Instead of slowing down the development of their agent and introducing it bit by bit to the public, they developed a second AI agent, which tested the first one, attacked it constantly, and tried to discover weaknesses or points of exploitation. In my view, what Telus did is ultimately the best way to address the current challenges posed by AI.

The question people must answer is whether we lean toward this kind of acceleration—using innovation to solve safety problems—or toward precaution, even at the cost of innovation itself.

Is it really a good idea to leave this to commercial companies?

In the field of business, we would call this a conflict of interest, where professional responsibilities are conflicting with personal interests.

But there’s nothing inherently new about this. Think about advances in automobiles. There were similar debates about safety regulations surrounding seatbelts. In fact, the push for government-mandated passenger restraints was the result of car companies themselves. They wanted seatbelts in the cars, but no car company wanted to go first because they thought that would signal to the public that cars were dangerous.

The question people must answer is whether we lean toward acceleration …or toward precaution, even at the cost of innovation itself.

What the car companies did was they all got together in a smoky back room and said, ‘Look, if we can just get the government to impose this on us, then we can get these safety belts, which we all think we should have, without having to worry about the bad publicity.’

Something like that is probably what's going on in the world of Anthropic and Open AI. They're the ones who benefit most from regulations, so it stands to reason they're the .

How so?

It's called regulatory capture: When a set of regulations create high barriers to entry. When an industry goes from being relatively regulation-free to heavily regulated, it tends to freeze the companies in the positions they are at. It happens in healthcare. It happens in the automobile industry. And it would likely happen with AI.

“Regulations operate in a murky world, where the incentives are often hidden.”

Regulations can make it very difficult for anyone else to enter the field, and that would mean that we will have forever OpenAI versus Anthropic, with a little bit of Meta and a little bit of Grok. That doesn't mean that regulation is a bad idea. It just means that regulations operate in a murky world, where the incentives are often hidden.