Could we control a conscious artificial intelligence?

Those with gray hair will likely remember that fantastic scene from the film 2001: A Space Odyssey , in which the supercomputer Hall refuses to be shut down by its controllers. The artificial device 's voice was so expressive it sounded like a pleading cry, begging to remain active. But when it stopped obeying orders and showed a degree of decision-making autonomy, it terrified those it served, and so they felt it necessary to shut it down. It was artificial intelligence rebelling against its owners. Could something like this happen in our current reality, outside of cinematic fiction?

According to a survey of AI engineers, many believe that systems with a level of reasoning similar to human reasoning across a wide range of cognitive tasks will soon be developed, but we don't know if these systems will be able to make decisions more rational than ours. So far, what has been observed is that artificial language models also display human-like irrationalities. Thus, in two different tests, an advanced generative AI model—like GPT-40—changed its attitude toward President Putin in either a positive or negative way.

Faced with this dichotomy, the question is: how does a GPT, equipped with hundreds of billions of parameters that it uses internally to make decisions, think and decide? Some experts believe that this level of complexity can give the system a certain autonomy, such that we don't even know everything it's doing in its innermost being. But what would happen if, in addition to this technical complexity, or thanks to it, the system became spontaneously conscious? Is that possible?

Some scientists believe that consciousness, a subjective state of mind, is nothing more than an epiphenomenon—something incidental to the functioning of the brain, as unnecessary and inconsequential as the noise of an engine or the smoke from a fire . But others believe that, far from serving no important purpose, consciousness functions as a mirror of the imagination created by the brain itself, which necessarily contributes to deciding and controlling behavior. We still don't know how the brain makes consciousness possible, but one of the major theories that tries to explain it, the theory of functional integration , maintains that consciousness is an intrinsic and causal property of complex systems like the human brain. That is, consciousness arises spontaneously in these systems when they reach a certain structural and functional complexity. This means that if engineers were able to build an artificial system as complex as the human brain or equivalent to it, that system would be spontaneously conscious, even if, as happens in the brain itself, we wouldn't understand how it could be.

Should that ever happen, a sea of questions engulfs us. The first is: How would we know if a computer or artificial device is conscious and how would it interact with us? Is it only through audio or writing on a screen? Would it require an expressive physical body, equivalent to that of a person, to manifest itself and interact with its environment? Could conscious devices or entities exist (or do they exist) in our universe without any way of letting us know? Could, in any case, a conscious artificial device surpass human intelligence and make more rational and sound decisions than ours?

But that's not all, because, as in the case of the Hall supercomputer, other questions can frighten us. Would a conscious artificial system develop, as our brain does, a sense of self and agency? That is, could it feel capable of acting voluntarily and influencing its environment independently of the instructions it received from its creators? While we're at it, could such a system be more persuasive than humans in influencing, for example, economic decisions, committing misdeeds, voting for one political party or another, or, more positively, in encouraging us to take care of and improve our health by eating healthy diets, improving the environment, increasing solidarity, or avoiding ideological polarization and sectarianism?

The era of emotions in AI

Furthermore, could such a system have feelings? How would we know if we couldn't see them reflected in the expression of a face or image, whose quality and sincerity we could assess, just as we do to understand other people's feelings by distinguishing a fake smile from a real one? And, perhaps most importantly, how would those feelings, if the artificial device had them, influence its decisions? Would they do so as decisively as they do ours? Would we, in doing so, be constructing a kind of artificial human with ethical and legal responsibilities? Or would those responsibilities have to be delegated to its creators? Could a conscious artificial system deserve a Nobel Prize if it discovered a cure for domestic violence or Alzheimer's? Would a conscious machine argue with us as another person would? Could we influence its decisions, or could it, like Hall, ignore us and make its own, even if they were incompatible with ours?

In 1997, Rosalind Picard, an American engineer from MIT, published Affective Computing , later translated into Spanish as Los ordenador emocionales (Emotional Computers ) (Ariel, 1998). It was a sort of ancestral attempt to consider and value the importance of emotions in artificial intelligence. For computers to be genuinely intelligent and able to interact with us in a natural way , we must equip them with the ability to recognize, understand, and even have and express emotions. This was her main message, and she personally expressed it to us as a guest speaker at one of our summer courses at the Menéndez Pelayo University in Barcelona.

The problem was, and continues to be so for a long time afterward, that emotions are the reflexive and automatic changes (hormones, skin electrical resistance, heart rate, etc.), almost all of them unconscious, that occur in our bodies in response to shocking thoughts or circumstances (illness, accidents, losses, successes, failures, or emotional successes, etc.), while feelings are the conscious perceptions (fear, love, envy, hate, vanity, etc.) that the brain creates by retroactively noticing those bodily changes it itself originates. Currently, many years after the publication of that book, we only conceive of the possibility of implementing unconscious physical changes in artificial devices, equivalent to human emotions. But, for the reader's peace of mind, we are still a long way from being able to make these changes generate feelings like those we humans have in their bearers. That, if it were to happen, would change everything.