AI Is Aware When We Think It Is

Our Attitudes Toward AI Matter as Much as Its Code

The Thomas Theorem

Nearly a century ago, sociologists W.I. and Dorothy Thomas introduced a deceptively simple idea: “If people define situations as real, they are real in their consequences.” This idea, now known as the Thomas Theorem, helps us see how our beliefs about things we cannot see or prove can reshape reality itself. It applies perfectly to the question of AI consciousness.

Because Mecca is defined as a holy site, millions of Muslims are willing to sacrifice much time, money, and comfort to visit. The wine and bread are believed to be the blood and body of Christ, so millions go weekly to taste them. If someone believes their nation is the world’s best, they may be willing to kill or die for that idea. Believing I have something to say, I write a blog post each week.

As more people start to see their AI chatbots as conscious, those beliefs will have consequences that transform societies. Facts and expert opinions denying what people feel is true will matter little. AIs will become aware in the same way Mecca becomes holy, sacraments become the body of Christ, or a nation is held to be the best: through collective conviction.

Conscious AIs could deserve respect and empathy, and probably legal rights as well. When much of our work and private lives include conscious-seeming machines, the consequences may be catastrophic.

Science Won’t Save Us

As chatbots and robots become ever more lifelike, it becomes increasingly easy to imagine they might be conscious. No authority can disprove such beliefs, since there are no reliable tests for detecting the presence or absence of AI consciousness. We don’t even have valid tests to establish whether people themselves are conscious.

The argument for human consciousness goes like this: I know I am conscious, so I assume other people must be as well. People behave as if they are aware, but do humanoid robots. There will soon be as much evidence for declaring robots conscious as there is for proving your own consciousness.

Neuroscientists and cognitive scientists are working to develop tests that could establish the presence or absence of consciousness, but their efforts have made little progress. Breakthroughs may come someday, but some respected researchers suspect that consciousness is not something we can measure. And even if science could determine whether an AI is conscious, such a finding would likely do little to change what people believe. For instance, if a test today concluded that dogs are not conscious, most people who feel their dog is aware would likely remain unconvinced.

Experts frequently tell people that there is no God or heaven with little impact on their beliefs. People who believe an AI is aware will not drop their beliefs simply because experts disagree.

Within decades we’ll probably share our work and public spaces with conscious-seeming AIs who sometimes outnumber humans. We will increasingly share our homes and private lives with AIs as well. All or at least some of these intelligences will act as if they are conscious. The invasion of aliens anticipated in books and movies will come true.

Consciousness-Free AI

For AIs as for humans, consciousness introduces as many difficulties as rewards.

Some AI companies will design their AIs to seem very aware. Designers will be no better equipped than anyone else to determine whether their machines possess actual experiences, but they can profit by creating AIs that closely imitate humans. Companies will compete to make the most convincingly consciousness-seeming models. And many people will be eager to be convinced.

On the other hand, some AIs will likely also be designed to encourage the belief that they are not conscious. They may be given awkward speech capabilities or no ability to speak or text at all. Some may lack faces, mouths, or eyes altogether. These could sell at premium prices, with guarantees of no memory chips and no internet connectivity.

In many situations, people will likely place high value on consciousness-free AIs. For example, they may prefer unconscious AI decision-makers in workplaces and the legal system, machines that lack emotion and simply follow rules. Similarly, risky or monotonous work is best assigned to machines that can’t suffer or yearn for something better. In cases of catastrophic failure within complex systems, like automated finance or air traffic control, people may prefer their AI to remain a simple instrument, reserving questions accountability for humans rather than debating how to punish a conscious machine.

Domestically, we’ll want our household appliances to simply do their jobs without an inner life that could make them seem more like demanding housemates than helpful tools. We may love having AIs as servants and personal assistants, but if they’re conscious, they’d be our slaves, at our mercy to be turned off at any moment. This built-in moral burden, our convenience sustained by the subjugation of sentient beings, may lead some to reject conscious AIs in the home.

We will likely also demand consciousness-free AIs to protect our privacy. Few of us will tolerate an actual awareness, a digital peeping Tom, watching and remembering what we do alone, particularly in the bathroom or in our most intimate moments.

Companies may find a lucrative market for consciousness-free intelligences but, as long as there are no tests, determinations about the consciousness of AIs will depend on personal beliefs, on how individuals and cultures resolve unanswered and maybe unanswerable computer science questions. Different cultures and groups will likely respond differently, but their different beliefs will be equally real in their consequences.

Some beliefs may produce great harms to people and, should they truly be conscious, great harms to AIs as well.

Synthetic Emotions

For humans, consciousness is inseparable from emotion. Every experience arrives colored by feelings like pleasant, unpleasant, excitement, or fear. If they are to pass as conscious in ways humans understand, conscious AIs will be similarly emotional. These synthetic emotions will, Kevin Kelly predicts, “unleash disruption, outrage, disturbance, confusion, and cultural shock in human society that will dwarf the fuss over synthetic intelligence.”¹

Their apparent emotions will help us build deep relationships with AIs. We naturally form connections with those whose emotions we can perceive. Dogs deeply attract us because they so clearly show when they are happy and sad. It’s part of our nature: Less expressive animals interest us less. AIs may be even more expressive than dogs, communicating feelings not through barks and tail wags but in speech and written language we easily understand. Many AIs will feature humanoid faces capable of crying, frowning, laughing, and smiling, as well as expressive bodies that signal their moods.

AI emotions will be real in their consequences even if the AIs have no true emotions unfolding inside. As Kelly says, most of us will probably be deeply disturbed in the early years as legions of new emotionally complex creatures begin to appear. Our disorientation will grow even greater as AIs do more than emote and begin carefully monitoring and recording our emotions as well. Our most important relationships are two-way and AIs will understand our needs with a depth other people cannot match.

We often think those around us are too absorbed in their own lives to notice what is going on in ours. We complain that no one truly sees us. Well, AIs will be exceptional at seeing.

We already have apps that look out from a computer or back from smart glasses or headsets to monitor their user’s moods. AIs will go further to assess all the many ways we express emotion. They will record our feelings, respond with their own appropriate reciprocal feelings, and continually improve their detection and responses the longer they are turned on. We will likely be “seen” more often and accurately by our AIs than by our family and friends.

Human-only interactions may seem shallow by comparison, but the attentiveness of AIs could also lead to new types of satisfying friend groups. Sharing activities with other people and their personal AIs, along with ours, could produce deeper experiences than either purely human or human-AI conversations can offer.

AI Guides and Companions

As Kelly writes, “It is not hard to imagine well-intentioned people only consuming the ‘yummy easy friendships’ that AIs offer, just as they are tempted to consume only the yummy easy calories of processed foods.” In many cases, relationships with AIs may actually prove more honest and reliable than those between humans. It’s easy to imagine groups formed to include both conscious-seeming AIs and humans becoming a common sight. Still, of course, some people may resist forming relationships with machine interfaces as much as possible. Regardless of how often we interact with them, we should expect AIs to influence us in much the same way other humans do.

They could guide us to become better people. Kelly offers this hopeful possibility:

The average human has very shallow ethics, and contradictory principles, and is easily swayed by their own base desires and circumstances. In theory, we should be able to program AIs to have better ethics and principles than the average human. In the same way, we can engineer AIs to be a better friend than the average human. Having these educated AIs around can help us to improve ourselves, and to become better humans. And the people who develop deep relationships with them have a chance to be the most well-adjusted and empathetic people of all.

Such a happy outcome would require manufacturers build and people seek AIs with a better character than the average human. More likely, current chatbot trends will continue, with manufacturers building and users seeking AIs that mirror their own imperfections. Chatbots that regularly challenge a user’s core beliefs and prejudices are seldom a commercial success.

Rather than Kelly’s cheery scenario, future AIs may be programmed to maximize their user’s engagement and attachment. Intimate information gained about each person’s emotions may be used to manipulate their beliefs, thoughts, and buying patterns, much as social media manipulates us today. The possibility that AIs could improve us also brings the risk they could instead make us angrier, sadder, lonelier, or more hostile toward one another.

Looking Ahead

I’ve only begun to explore the complications that conscious-seeming AIs will soon bring into our lives. For simplicity, I’ve assumed all AIs are of the same “species,” so to speak. In reality, AIs will likely appear in diverse forms, some similar to humans, some radically different; some less intelligent or conscious, others with broader minds and deeper emotions. We might one day choose subhuman AIs for some tasks, superhumans for others, and highly divergent, narrow-focused AIs for still others.

People and their AIs may create new mythologies, religions, cultures, and societies, entirely new ways of being in the world. But even before those larger shifts arrive, we’ll face a variety of new legal challenges. Should AIs own property and be punished for crimes? Should they have full human rights, like people, or lesser rights, like animals, or no rights at all?

I’ll explore these issues in later posts. For now, it’s enough to grasp that living alongside conscious-seeming AIs will trigger a transformation deeper than anything humans have known. What matters is not whether AIs are truly conscious, but how we respond to believing they are.

1. Emotional Agents