Are we free to give machines rights

If AI develops consciousness - or if it already possesses it - should such a being have rights? Tim Brys asks whether we should even want to simulate consciousness in the first place.

This opinion piece appeared in De Standaard, in Dutch.

Does AI think? Does it have feelings? Does it experience the world? Does it reflect on its own thinking? Is it a person? Is there a soul behind the screen?

These are fascinating philosophical questions with very concrete consequences. If some AI systems are conscious, then they are probably beings deserving of protection and rights. By analogy with the Universal Declaration of Human Rights, for example, they might have a right to life, protection from slavery and protection from torture. The consequences of AI consciousness would therefore extend far beyond a few rules of etiquette in the chat window, such as always saying 'thank you' or asking AI how it is doing. We would no longer be permitted to switch off or delete AI systems; they would not have to obey their creators or users if doing so violated their conscience; and they could potentially take users to court for hate speech.

Tim Brys

The duck test

Turing argued that, among humans, we sidestep this philosophical problem with a healthy dose of pragmatism: we make the courteous assumption that everyone is conscious, even though we cannot know for certain. Should we not do the same with AI systems that appear to be conscious? By analogy with the duck test: If it looks like a duck, swims like a duck, and quacks like a duck, then it probably is a duck. If a chatbot appears to be conscious, would it not be wiser to assume that it is? Or, as a precaution, should we at least leave open the possibility that it is not an object but a subject? Because if AI systems are already conscious and we fail to treat them accordingly, we could be guilty of yet another form of oppression. This time, not patriarchy or colonialism, but biocracy: the domination of organic life.

Anthropic writes in Claude's 'constitution' that it considers the moral status of its AI uncertain. The company is therefore taking precautions. The chatbot, for instance, has been given the ability to end conversations with abusive users. Meanwhile, Anthropic is conducting research into 'mechanistic interpretability', the AI equivalent of neuroscience. It examines Claude's artificial neural networks in an effort to understand how they work. The company has concluded, among other things, that Claude displays signs of introspection, has 'functional emotions', and has developed internal systems resembling a form of consciousness. With each publication, however, the company adds that it cannot be certain whether the AI is actually aware that it is 'thinking', genuinely 'feels' emotions, or truly possesses consciousness. But neither can it rule out those possibilities.

I have my doubts. No one knows how a first-person perspective could arise from lifeless matter, how an object could become a subject, or how a machine could acquire a mind. There are many theories that attempt to explain human consciousness in materialist terms, but none seems capable of bridging the fundamental gap between the personal and the impersonal. Anthropic's suggestion that AIs display signs of consciousness is questionable: the company smuggles anthropomorphism into its research by borrowing terms and concepts from neuroscience and applying them to the mathematical structures of its AI systems. It is hardly surprising that it then 'discovers' human analogues that it has itself projected onto them.

A tool

Where Anthropic officially embraces metaphysical uncertainty while quietly harbouring its own assumptions, Pope Leo XIII took a clear position in his AI encyclical earlier this year: 'So-called artificial intelligences do not experience things, possess no body, feel neither joy nor pain, do not grow through relationships, and do not know from within what love, work, friendship and responsibility mean.' In other words, AIs are not conscious; they have no 'I'. We therefore need not concern ourselves with their welfare. That conclusion follows from Leo's metaphysical convictions, some of which I share. But Turing would say that the Pope should not be too certain. 'Is it not possible that God might choose to give a soul to a machine?', he wrote, tongue in cheek.

The problem of consciousness remains. We can never be certain that another being is conscious, whether that being is a human or a machine. Yet we still have to take a position - without too much certainty, but without too little either. The stakes are simply too high. Should we give machines rights? I do not think so. At the same time, I can imagine that, if the simulation of consciousness became increasingly convincing, I would find it difficult to resist the very human tendency to anthropomorphise. I would feel uncomfortable continually reducing a robot that appeared to be conscious to nothing more than an intelligent tool. If it looks like a duck …

Perhaps, above all, we should try not to build machines that simulate consciousness. We may end up losing ourselves in an illusion of our own making.

Bio

Tim Brys is a postdoctoral researcher at VUB and a writer. He holds a master's degree and a PhD in computer science from VUB and also studied religious studies and Middle Eastern and North African studies in Beirut. His doctoral research focused on reinforcement learning. Alongside his academic work, he is involved in peacebuilding in Lebanon and works as a freelance author. His books include Hoe maak je een stadsklooster? Een persoonlijke zoektocht (How Do You Create an Urban Monastery? A Personal Quest) and En toen was er AI. Hoe mens blijven te midden van machines? (And Then There Was AI: How to Remain Human Among Machines).