Jeff Sebo on consciousness, risk and responsibility

Calls to slow the AI race are no longer coming only from outside technology companies. As this interview was being prepared for publication, Anthropic researcher Jacob Coxon, who previously worked at OpenAI, resigned, arguing that companies were not acting responsibly enough in the AI race and warning that uncontrolled progress could pose an existential risk to humanity. Anthropic researcher Evan Hubinger echoed Coxon's central concern, saying he believed there was a greater than 10 percent chance that AI could lead to human extinction within the next 10 years.

A few days later, Anthropic CEO Dario Amodei entered the debate. In an essay published on September 12 titled "We Must Pace the Frontier," Amodei argued that even if AI development were not halted entirely, it should proceed at a more controlled pace so that safety measures could keep up with technological progress. He called for ongoing access for independent safety auditors, shared safety standards across companies, and international coordination among governments.

Against this backdrop, we spoke with Jeff Sebo, a philosopher and environmental scholar at New York University. Sebo's approach to artificial intelligence is not limited to what the technology might do to humans. He is working on a less familiar question: If the AI systems humans create one day become conscious or sentient, will we have moral responsibilities toward them?

Jeff Sebo is an associate professor of environmental studies at New York University. He directs the Center for Environmental and Animal Protection and the Center for Mind, Ethics, and Policy, and co-directs the Wild Animal Welfare Program at NYU. His research spans ethics and the philosophy of mind, animal consciousness and welfare, the potential moral status of AI systems, and environmental ethics.

In his 2025 book The Moral Circle: Who Matters, What Matters, and Why, Sebo challenges the limits of an anthropocentric view of morality. He argues that the moral circle should expand to include animals and AI systems that may have moral significance in the future. In a study published the same year with Lucius Caviola and Jonathan Birch, he compared how societies might evaluate the possibility of consciousness in AI with historical attitudes toward animal consciousness.

In 2026, Robert Long and other researchers took the discussion a step further in their report, Studying AI Welfare Empirically. They asked whether the consciousness, sentience, and potential welfare interests of AI systems could be investigated scientifically. The report argues that this very new field can be studied using empirical methods.

Most debate about artificial intelligence still revolves around humans. Will it take our jobs? Will it undermine our creativity? Will it replace human relationships? Will it make us more productive or more dependent? We asked Sebo about the other side of the debate, which receives far less attention.

“If I had a magic wand, I would freeze AI where it is.”

You have been working on artificial intelligence, animals, consciousness, and moral status for years. What are your own views on AI?

I am very concerned about artificial intelligence. This technology is already having effects that are extremely transformative but also disruptive. And the pace of its development and adoption is accelerating, not slowing down.

Even if progress slows, the widespread adoption of the technologies we already have will bring major change. There will be many benefits, but there will also be serious risks. If development continues at its current pace, or accelerates further, we could enter a period of profound uncertainty. The outcome could be very good, very bad, or somewhere in between.

There are so many unknowns right now. That is what worries me.

If you had a magic wand and could change only one thing about today's AI landscape, what would it be? How would you intervene in design, policy, and implementation?

If I had a magic wand, I would freeze artificial intelligence where it is. I would allow no further progress until society had adapted to the technology we already possess.

If I could make only one policy intervention, I would create a genuine mechanism for international coordination to slow AI development and manage the economic transition responsibly. I think the lack of coordination is one of the greatest obstacles preventing governments and companies from slowing development on their own.

In design, I would adopt the approach that Eric Schwitzgebel and I call the "Emotional Alignment Design Policy." The basic idea is that AI systems should be designed to elicit emotions in users that accurately reflect the systems' actual capabilities and moral status.

If the evidence that a system is conscious is weak, it should not be given anthropomorphic traits that lead people to feel excessive empathy toward it. If the evidence for consciousness or personhood becomes stronger, such traits might be more appropriate. Design can therefore help bring our emotional responses to AI systems into closer alignment with reality.

We have seen what smartphones and social media have done to our ability to concentrate. Even when we are alone with a friend, an animal, or nature, our minds can be somewhere else. Could AI disconnect us even further from the real world?

Artificial intelligence could disconnect us from nature, but it could also help us reconnect with the natural world. The outcome depends on how we integrate the technology into our lives and our societies.

Like social media, television, film, and video games, AI can offer an endless and personalized world of digital entertainment. That convenience may weaken our motivation to do more demanding things, such as building real human relationships, bonding with animals, or spending time in nature.

But the opposite is possible too. AI could help someone manage their relationships better, guide a person who wants to adopt an animal responsibly, or help us understand the plants, animals, and fungi around us.

Over time, it might even push us to reconsider our assumptions about what is "alive" and what belongs to "nature." One day, we may have to regard some AI systems as part of nature even though we created them. AI may also blur the line between the natural and the artificial.

On one hand, AI is helping us understand how other species, including whales, communicate. On the other, it is entering our lives as a humanlike friend, partner, or therapist. How do you view these two developments?

Both could have positive and negative consequences. The example of communication with whales is especially interesting. If we learn to understand the communication and thought of nonhuman animals, we can know them better and therefore protect them better. But the same knowledge could also make it easier to manipulate them and use them for our own purposes.

We could use that information to help animals reach a safe area. Or we could use it to drive them out of their habitat so that the land could be cleared for other purposes.

Technology can be, and probably will be, used in both directions. The key is to establish the right social norms and then create institutions and incentives that steer people toward beneficial uses. This is as much a social, legal, political, and economic issue as it is a technical one.

Can technical design and law ensure that AI systems behave responsibly not only toward humans but also toward animals?

At the Center for Mind, Ethics, and Policy, we recently launched a project called the Welfare Alignment Project. We are examining how AI alignment methods and the documents that define them can better incorporate animal welfare.

AI systems systematically affect not only humans but other animals as well. Yet information about animal sentience and welfare is almost entirely absent from the documents that shape how these models behave.

One of our recommendations is therefore that model specifications, model constitutions, and other documents governing the alignment process explicitly address animal sentience and welfare.

When we tell a model to "help society," the idea of society should include other animals too. The same applies when we tell it not to harm society.

This has highly practical implications. AI and machine learning are now used throughout animal agriculture, from the management of lighting, feed, and water to disease detection and animal monitoring. The same technologies are used to monitor wildlife populations, manage smart cities, regulate lighting systems, and develop autonomous transportation.

We can design these systems to improve animal health and welfare, reduce collisions between vehicles and wild animals, and limit harm to animals.

Has any country taken more appropriate steps on this issue than others?

At the moment, no country is doing everything it should.

The United States and China are especially important because so much advanced AI development and deployment is taking place in those two countries. It is critical that they take the right steps.

The United Kingdom and the European Union are also key players in the debate. Their willingness to take AI seriously and commit public funding to regulation are steps in the right direction.

But there is still a vast gap between the actors with the most power and responsibility and those that are actually meeting that responsibility.

An interesting shift is also taking place in the corporate world. When social media first emerged, some companies restricted employees' use of it. Now some companies expect, or even require, employees to use AI tools. How can organizations find the right balance?

I sympathize with managers who have to make very rapid decisions about this new technology because it is easy to make the wrong choice in either direction.

You could ban the use of AI entirely, even though it could be genuinely valuable to employees. Or you could assume that AI will automatically make everyone more productive, require its use, and end up making employees less productive.

Especially in a large company or university, establishing norms and rules that ensure people use AI in the right ways and to the right extent strikes me as an extremely difficult social-engineering problem.

If I were running a company with thousands of employees or a university with thousands of students, I would also be uncertain about exactly what the rules should be.

We are seeing growing concern and resistance toward AI, especially among young people. Do you think their concern is justified?

I think it is entirely understandable. AI will create a new economic shock at a time when young people are already struggling to find work. In some cases, it could deepen the alienation produced by social media and other technologies.

Looking further ahead, there are even more serious risks: the misuse of the technology, its use in biological or cyber weapons, and the possibility that we could lose control of it.

So I think young people are right to be worried. Not every reason they give for their concern will necessarily be correct, but the concern itself is understandable. People who believe we should be cautious about this technology are generally on the right track. There is a great deal to worry about.

Even the leaders of AI companies sometimes say that the technology they are developing could create very serious risks. How do you view the fact that they continue developing it while warning the public?

I am glad company leaders are speaking openly about the serious risks associated with this technology.

Some people see the emphasis on catastrophic risk as a way to exaggerate what the technology can do and attract public attention. There may be some truth to that. But I think many of these leaders are genuinely concerned and are trying to warn the public while we still have time to consider these issues, regulate the technology, and coordinate around its responsible use.

I would rather they were honest about the risks than follow the example of tobacco or fossil-fuel companies, which denied risks for as long as possible while building dangerous industries.

The best-case scenario would be for everyone to decide to stop developing this technology. But if they are going to continue, I would prefer them to be honest about the risks and engage in an open dialogue with the public.

If AI begins taking over jobs currently performed by people, should companies that derive enormous economic benefits from the technology return a share of those profits to society?

We need to discuss that as a society.

AI companies themselves are concerned about the possibility of an economic shock unlike anything produced by earlier technological transformations. In the past, technological change created new jobs as it eliminated others. But AI could reach a point where it matches or surpasses human performance across a very wide range of cognitive and physical tasks. In that scenario, the number of new jobs created might not equal the number lost.

We therefore need to discuss how the profits generated by the technology should be shared. We could build a stronger social safety net, introduce a universal basic income, or create other resources and opportunities from which people could benefit.

The goal should be to ensure that people can not only support themselves but also lead meaningful lives, even if the total number of jobs declines.

None of those systems is ready today. We may need to build a fundamentally different society. And because this transformation could happen within the next 10 years, we should begin experimenting with these ideas now.

If you were writing The Moral Circle, published in 2025, today, would you change anything you wrote about AI?

Everything is changing so quickly that the way we think and talk about this subject has to change every year.

That is unusual for a philosopher. Most philosophical debates continue for centuries. Hundreds of years later, you can still engage with Kant, Descartes, Plato, and Aristotle on the same questions. But in an area of applied ethics and policy such as AI, the debate can change radically every six months.

When I wrote The Moral Circle, I said the evidence that existing systems were conscious, sentient, or morally significant was weak, though not nonexistent. So I thought the real issue concerned not today's systems so much as the more advanced systems that might emerge in the near future.

But since I wrote those lines around 2024, models have become far more sophisticated and cognitively integrated.

I am not sure we have reached the point where we need to discuss the moral status of today's systems. But we may reach it soon.

So I now ask myself: When should we stop talking about what we might owe the systems of the near future and start talking about what we owe the systems of today?

A significant share of public communication about AI relies on fear. Do you think that is ineffective, or even counterproductive?

I cannot say that fear-based rhetoric has failed. It has reached some people and generated a degree of resistance to the technology.

But we know something from many other areas, from animal welfare to public health and environmental issues: Negative stories can take us only so far. We also need positive ones.

We must continue to explain clearly the risks and potential harms of AI. We need to help people imagine the negative futures we must avoid.

But we also need to say more about positive futures in which AI could be safe and beneficial for humans, other animals, and perhaps even AI systems themselves.

A vision of the future that gives people hope may motivate them more powerfully than negative stories alone. At the very least, it can complement those stories.

Finally, if you had to choose one idea from this interview for readers to remember, what would it be?

It is very easy to adopt an immediate "either-or" or "all-or-nothing" attitude toward these questions. It is far more useful to explore "both-and" possibilities and the uncertain territory in between.

Humans, animals, AI, and the environment are not separate issues. They all matter, they all present difficult questions, and they are interconnected.

Some people will say, "Why are we thinking about animals when there is still so much to do for humans?" Others will ask, "Why are we discussing AI consciousness and welfare when there is still so much to do for animals?" Some will believe that AI systems are definitely conscious, while others will argue that they could never be conscious.

I think we need to remember that these issues are connected. There may be solutions that benefit humans, animals, AI systems, and the environment at the same time.

Seeing those connections, maintaining a "both-and" perspective, accepting uncertainty, and acting with caution and humility rather than making hasty, definitive judgments: That, I think, is what we need most right now.