"Super-intelligence is already here"
Virginia Dignum on the collective intelligence we already have — and the mirror AI holds up to us
Virginia Dignum — computer scientist, director of Sweden’s WASP-HS programme, and one of the drafters of the EU AI Act — is among Europe’s clearest voices in AI ethics. Author of Responsible Artificial Intelligence and, most recently, The AI Paradox: How to Make Sense of a Complex Future, she writes and speaks about what AI actually is, what it isn’t, and what we do to ourselves when we forget the difference.
Beyond familiar concerns such as regulation or job displacement, what challenges do current AI systems raise for human existence, identity, or lived experience?
That’s a big one. I think the biggest challenge we encounter is that we will lose the capacity and the opportunity to take charge of our own decisions and our own will. More and more, we will potentially let it be done by AI — which basically means being done by someone else outside of ourselves. And then we lose the capacity to discern what is really ours — the capacity to make decisions, to decide for ourselves.
What deeper questions about human life might arise as AI takes over activities that were previously carried out by humans?
I’m not sure if life itself is the issue. But I would say that our ability to reason for ourselves, to take a stance on a specific situation, our ability to distinguish between fact and fiction — those are issues we are already seeing happening. And if you take that to an extreme, we become much more spectators of our own lives than the actors of our own lives.
How might we find ourselves in that situation — where we no longer take charge of our own decisions?
There is a very concerning development at the moment. We see AI being used and applied for the sake of AI itself. It becomes the goal, instead of a means for a specific issue. You see this at the level of governments all over the world, coming up with these strategies about what I call AI first. The idea is that you need to use AI, based on the feeling that all other governments are doing the same. This feeling of missing out. And then AI becomes the goal in itself, no longer a means for a well-formulated and understood issue.
The more we do that, the more we become dependent on mediated interactions and mediated decisions with these systems — to an extent that we will not be able to take decisions anymore without the mediation. I read recently from some psychologists that children in secondary school are already feeling very uncomfortable whenever they have to write something without the support of AI. All that type of dependency, taken to an extreme, is a real concern.
Here in Sweden, the government just recently issued a requirement for public services and civil organisations to use AI within the next year, and to report back to the government on how they are using it. And there is very little in that directive that asks why they are doing this. Even at the level of public service, the reasoning is much more that you have to use AI, and not so much asking yourself what for.
So what you are saying, if I understand correctly, is that it is not AI itself that reshapes what it means to be human — it is that humans might delegate their decisions to AI?
AI is a technology we develop. It is an artifact, and in itself it will not really do anything. It is much more about what we are using it for, and how we are using it, and how we are allowing the mediation of our interactions with these systems. AI itself, as I also say in the book, is a misnomer — maybe even an empty signifier. We are calling everything and anything AI, and at a certain moment it doesn’t have any meaning anymore. But what we are seeing very clearly is that we are increasingly dependent on the mediation of interactions and decisions using some type of artificial system.
In contemporary philosophy of AI, discussions of human distinctiveness often focus on identifying what capacities machines lack. How should we approach human distinctiveness in this context?
I’m not really sure we should approach this from the idea of human distinctiveness at all — because then we are already assuming that AI has any way to compare with us, and then we have to look at how we are different.
I think it’s more a matter of understanding that what you are building is a problem-solving system — a system that can address a very specific type of human intelligence: what I would call the logical-mathematical approach. Human intelligence is much more than solving problems. It is much more than only the logical-mathematical type of capabilities. And even within that specific part of human intelligence, AI is being developed in a very specific approach.
The current approach assumes that by identifying patterns in a set of data — which is basically a set of results of previous human actions — we can somehow reconstruct the reasoning that humans took. All of this is very limited. It limits the type of intelligence we are looking at, and the approach to that specific type of intelligence.
Rather than comparing ourselves with what AI can or cannot do, we need to really understand what makes us human. And it is nothing new. In the Renaissance, philosophers were comparing human intelligence to clocks. They have been comparing intelligence to water pipes and water mechanics. The last thing we did was to compare our intelligence to computing. We are always trying to compare ourselves to our latest technology — instead of realising that this technology is here only, and necessarily, because of our own capacity to create it.
So in that sense, we are not really facing two kinds of intelligence. AI is more like a simulation of certain branches of human capacity?
I wouldn’t even say it’s a simulation. It’s an alternative to some of our types of intelligence.
The AI field started with the idea that we would be able to understand human intelligence better if we could build artificial models of what we understood of intelligence. That’s what came in with the Dartmouth Workshop of 1956, and what Turing claimed we could do. We started the AI field as something that would simulate human intelligence, as a means to understanding that intelligence. Nowadays we are building systems with the capacity to act directly, to intervene in the world in which they operate. And what is relevant for an active system is not necessarily the same as what we need from a simulated one. So we again change the understanding of what we mean by AI — to mean something different than the original purpose.
You’ve written about AI as a mirror of human thinking — Shannon Vallor’s phrase. What becomes visible through this mirroring?
The AI mirror, as Shannon says in her book, is the idea that we build something in our own likeness, and then we fall in love with that likeness. That is fundamental — not only for AI, but for human curiosity and for science in general. Trying to understand what makes us. What we can change, what we can improve, what we can take away — and what still remains as being humanity.
I think this is basic human intelligence: the curiosity that drives us forward, trying to build things that look like us. It starts in childhood. When kids play with dolls, or even with stones, pretending the stone is a child or another person — those are exactly the ways we have to understand ourselves: by experimenting. And AI, if anything, is one of such experiments. What can we build that decides like us?
I hope this comparison will give us a better understanding of ourselves — an understanding of what is in us that is not in the mirror. And it is also important to understand the difference: the image in a mirror will never be the reality itself. That is an important way of looking at AI.
When we consider AI systems that take over or imitate forms of human thinking and judgment, which human capacities require renewed attention or active cultivation today?
I would say the capacity to question ourselves and to remain curious — to remain alert to what’s happening, what the consequences are, to try to extrapolate from what’s happening to other dimensions. Not so much the doing itself, but the capacity to reflect on what we are doing, why we are doing it, and what if we would do something different? Those are the core capabilities moving forward.
The what if. Why does it matter?
Because it is only when we ask ourselves what if that we can really discover alternatives. If we accept things as they are, we are not prepared to find the differences. Science is basically the capacity to ask what if. What if I throw the ball out of the tower? What if I move this from one place to another? What if I count things in one way or another? Science has always been moved forward by what if.
Do you feel this capacity is losing depth in human practice?
I don’t think it’s losing depth. If anything, we will need to go even further in the what if — perhaps at higher levels of abstraction — exactly because we have now created machines which can analyse what ifs very quickly.
What happens to this capacity when it is not actively practised?
That is the worry. If we don’t practise it, we will start giving up on it. This is exactly what is important, moving forward — in terms of education and the new generations, to cultivate the capacity for creativity and reflection. If we don’t do that, then like many of our other capabilities, we forget them very quickly. If we don’t use the muscles we have, they atrophy. If we don’t use these mental muscles, they might atrophy as well. That would be a loss for all of us.
Might that happen as a result of over-reliance on AI?
It could. Or, on the other hand, relying on AI for some of the tasks we currently do could potentially liberate us to have more time for the reflective and the what if. In the same way that if I don’t need to calculate the square root by hand, I have more time to look at the mathematical equations I’m working on. It could go that way. But we need to be very careful and very conscious of keeping a finger on what’s happening.
You make an intriguing point in the book about how social behaviour is the foundation of our intelligence.
That is actually not mine. It comes from African and other non-Western philosophies. The concept of Ubuntu, which comes from African philosophy, says exactly this: Ubuntu means I am because we are. That’s a fundamentally different understanding of the world from our own Cartesian one — I think, therefore I am. In the West, we put reason first, and then identity. Other cultures put the collective and the social first, and identity emerges from that. You see how completely opposite they are.
In the book I try to build on those concepts, to see the absurdity of focusing on thinking as the core function of our existence.
So the picture of AI as a lone thinker — a kind of lone wolf — is itself the problem?
Yes. And that is exactly why the capacity to question ourselves is so important. The willingness and necessity of questioning ourselves is what probably makes us stand apart from animals — and maybe also from AI.
How is that capacity actually cultivated in human life?
Through curiosity. Through the willingness to try further than we think we are able. And through engaging with others. Because, as I say later in the book, super-intelligence is already here. It is the collective intelligence of all of us. We are only able to achieve this super-intelligence by interacting with each other.
If we really want the super-intelligence that big tech is promising us, we had better look at our collective capability to work together, to be more than what each of us is. That is what makes us move forward — even as some parts of what we consider intelligence today might be replaced by machines. It is exactly by cultivating our capacity to interact with each other, to contribute to each other, to remain curious and willing to go further than the borders we perceive.
In a sense, we are waiting for a super-intelligence that would be an awakening of our own dormant awareness of what we already are.
Yes. And I think the current understanding of super-intelligence is just a commercial gimmick. Big tech companies have got to somewhere near the limits of what they can do with LLMs and transformers. So if they can keep us hooked on the idea that they are going to use these tools to generate super-intelligence, we keep investing in them. It is a very down-to-earth, very simplistic, but very real understanding of what is happening. They need to keep their own importance and their own investment coming. So they promise something we know they will not be able to do — but no one else will either.
Super-intelligence is not, by definition, one being more intelligent than we are. Together, we are more intelligent than any of us is. If we can build tools that extend and awaken the capabilities we have together, that is what we should be looking at. That ChatGPT will suddenly solve all our problems — that is a lost battle.
That is a striking claim, from someone with your depth of expertise in the field.
We need someone — I hope some kids soon — who will point out that the emperor has no clothes. Because that is what it is. And if we keep putting more and more excitement around it, we forget to notice that the emperor has no clothes. Many of those who are building these things do not have a background of forty years in AI. Many of them do not have a background in AI at all — and certainly not in the philosophy of the field. Maybe they should go back to school.
Is there anything you feel is missing from the discussion — anything I haven’t asked that would be important?
One thing I was just remembering. One of the first definitions of AI I used — that many people used — is the definition that says AI is what AI cannot do. Fifty years ago, calculating a square root automatically was a sign of intelligence. Now that any calculator can do it, it is not really AI anymore. That machines can play chess is not AI anymore. That machines can find routes on maps is not AI anymore. All of this was part of the discipline for many years. Even machine learning — nowadays people are saying, oh, that is not AI anymore, it is just pattern matching.
On one hand, this is a kind of limitation of AI. On the other, it is exactly the capability we have to move forward. Once we solve a problem, it is no longer something we find very intelligent — because we know how to do it.
Which means, perhaps, that we understand ourselves more deeply through what AI cannot yet do.
Exactly. And that is why I say it keeps moving forward. Because it is what AI cannot do yet. And that will move again next time.