Beyond the Phenomenon: Kant, AI, and the Limits of Knowledge [Introductory]

Hello, everyone. My name is Alejandro Leal. I'm a Senior Analyst with KuppingerCole.

Today, I'll be doing a session. I'm aware it's a little bit more experimental, more philosophical, but I'm going to try to connect it with the overall theme of the conference, which is digital identity and AI. Here's the agenda.

First, we're going to look at Immanuel Kant's philosophy, try to understand it, and then we will jump into Gen-AI and the importance of the human element in this scenario, as well as looking at the critiques and the main takeaways. In a way, we can understand Kant. He was born in 1727 in Königsberg. Today is Kaliningrad, so not too far from here. His life can be understood as an attempt to fix human knowledge and to establish its proper limits.

He was a thinker of the Enlightenment, and his main work was The Critique of Pure Reason, which was published in the 1780s in two editions, between the American and French Revolutions. Those were very interesting times, with a lot of scientific, political, and social discoveries. One of his main principles is that every human has the right to question and to understand the limits of human knowledge. Why is this topic important?

I'm aware that it's a little bit strange to talk about some thinker from the 18th century, but I think that if we trace back the foundations of Western philosophy, of Western epistemology, that could shed some light into the current challenges that we face, especially in the age of artificial intelligence. I believe that if we understand the boundaries of human and machine cognition, that could be very important for us in the way we use these technologies, and the way we apply them in our own personal lives, as well as in our businesses.

The famous French philosopher Michel Foucault, in an interview in the 1970s, was talking about the history of philosophy, and he said that in the Middle Ages and before the Enlightenment, before the 18th century, philosophy was mainly concerned with religious matters, morality, and other topics. But according to Foucault, Kant was the first thinker that really asked the question, where are we? What is our reality? What is our factuality? What do we have in front of us?

So, if we take that position, where are we? We are at the Congress Hall in Berlin. This building was built in the 1960s. It has the Soviet realism architecture style, and it was constructed during the era of the German Democratic Republic. They used to have a lot of concerts and congresses and events during the 60s and 70s.

But today, well, it's EIC Day, and we are talking a lot about AI. So, if we take Kant seriously, then we can also ask this question of where are we, and what do we have in front of us?

So, before we get into Kant, just a little brief introduction on some philosophical concepts. So, metaphysics is the branch of philosophy that studies the ultimate source of reality, things like time and space. Then ontology is the study of the nature of being, so what it means to be, what it means to be a human. Then epistemology, which will be the main focus of today's session, which is a branch of philosophy that studies the origins and the limitations of knowledge.

So, in a nutshell, you can argue that Kant argues that knowledge is derived from experience, but what we experience must conform to the mental structures that our brain and our mind has. And, of course, again, this was a thinker of the 18th century.

Of course, if we look at current research on neuroscience or psychology philosophy, which I will talk a little bit later, it's a different study, it's a different definition, let's say. But this skepticism is called, or Kant calls it, transcendental idealism, which means that we cannot really know things in themselves, but only as how they appear to us, because our minds, they will understand what's in front of us based on how our minds function.

There's a German word, I don't know if I'm pronouncing it right, but it's umwelt, which in a way means that if you look at an insect or an animal, the way the reality, even if that animal or that insect is in this room, the reality and the nature of the experience is very different for that organism, for that being, or, as a matter of fact, for a human being. So, this picture in a way illustrates that. We know our reality, but if we really look deep, according to Kant, we cannot really know the things in themselves. He calls that the noumena. And for him, the phenomena is what we experience.

So, back in the 18th century, there was a debate between philosophers from the school of rationalism, thinkers like Rene Descartes, they would argue that you can only know things by using reason. And then there were other thinkers, empiricism, mainly from the UK, thinkers like David Hume, or John Locke, or George Berkeley, they said that knowledge can be derived only through experience.

So, Kant was the one that synthesized and combined the two in some way. So, according to him, the knowledge arises from the interplay between perception, so experience, and understanding, so rationality.

So, according to him, the human mind is actively structuring data that we obtain from experience. So, he categorizes these structures, he calls them categories, and these are, as he calls them, a priori.

So, these are things that are built up in our minds. So, things like time, space, causality. And in the next slide you will see, after this one, you will see the rest of those categories.

So, according to Kant, time and space are not out there. They are the lenses through which our minds filter experience.

So, time and space are preconditions for experience. They are, as I said, a priori. They are independent of experience, but they are necessary for experience to be processed by humans. And here's just a reminder of what transcendental idealism means.

So, we do not know things as they are in themselves, only as they appear to us. So, I think it's an interesting human framework, because the Enlightenment was a very humanistic, intellectual development. And I think that putting the human at the forefront in these days is very important, and it's a reminder that, despite the technology that we have, it's also important to know the limitations of the human mind before we completely give up our autonomy and our operations and workflows and epistemology to machines.

So, here are the categories that Kant talks about. According to him, there are 12, and they can be divided into four sections, quantity, quality, modality, and relation. And he talks about how these categories are preconditions of intelligibility.

So, if we apply this, if we compare it to how AI models are built, and I know it's a strange comparison, but for the sake of the presentation and to get the point, I think we need to make that comparison. The AI is built with data representations. It has a model and algorithm assumptions that pre-structure how the AI system sees and knows reality.

So, reality. AI does not really discover reality. It models it based on data.

So, in a way, it's similar to how Kant argues that humans cannot perceive things as how they truly are, but how they filter through our human minds. But there's a key difference, and the difference is that, although the AI lacks this self-consciousness, and these categories, the algorithms, the data representations, they are externally imposed by us humans.

So, the AI is not conscious. It doesn't obtain whatever the AI system knows through experience or through self-consciousness or through being.

So, that's a critical difference, and we see many people talking about how one day there will be an AI that will be intelligent and self-conscious, and, well, if we understand the limitations of the epistemology of the human and the system, then that's up for debate, I would say. So, if we take this Kantian epistemological framework seriously, according to Kant, critical reflection is a precondition of responsible knowledge.

So, the fact that we as humans can have this discussion about how much can we know, why do we know the things that we know, that's, in a way, what makes us human, and an AI system cannot really do that. It lacks critical self-awareness of their conditions of knowledge and the limitations of their knowledge.

So, as we use the AI more and more, I think that should challenge us to examine whether the output that we get from the AI system is true knowledge or is just a synthetic projection. John, the other week, shared with us this article about how some lawyers used an LLM to summarize a case study, and then the judge found that there were a lot of inconsistencies and false cases.

And, as John said, the LLM does not care about the truth. It's just text prediction based on highest probability.

So, we shouldn't trade our epistemology to the machine, and we hear this all the time. It's important to preserve human oversight. It's not just good practice, but it's a Kantian imperative, and that applies not only in our personal lives, but when you have a business and when you have a team, it's important to always have communication with your team members and ensure that all of them understand what's at stake, how are they using the technology, if that's going to bring value to the team, to the business.

So, the human element is always going to be there. So, I'm aware I'm running a little bit out of time, so I'm going to go a little bit fast here.

So, if we compare this to how Kant understands human knowledge and its categories, it's not too different. In a way, these models should have accurate data that could be used first to train the model in order to produce high-quality output and more accurate information. I won't go too deep here, but if you just compare traditional AI with Gen AI, we can see that Gen AI is really creating output that a human being can do, such as text or images or videos, which is, of course, something that has completely shifted from what we understood initially of what traditional AI was.

So, what is Gen AI lacking? What are machines lacking?

So, from this Kantian perspective, do they have these transcendental structures? Do they understand time, space, causality?

Well, not really in the Kantian sense. Gen AI doesn't really possess a priori forms of knowledge because they haven't obtained that through lived experience, but it's mainly through symbolic correlation. Does Gen AI have categories? Metaphorically, yes, but these categories are limited to the data used for training, and they're not transcendental as understood by Kant.

So, does Gen AI have limits of knowledge? Yes, but different than humans, because for humans, our limitation is more epistemological and it's more metaphysical rather than being computational, architectural, and ontological.

So, again, ontology, the study of what it means to be, does not really apply to a machine. But I think it's also worth to criticize a little bit Kant.

So, if you look at the contemporary critique from a neuroscience perspective, we understand now that knowledge emerges from a dynamic interaction between the brain, the environment, and our body. And it's not really made as these fixed internal forms or these categories as argued by Kant. There's also a historical critique.

So, he comes from the Enlightenment, so it was a very Eurocentric movement, and he is sort of lacking a deeper understanding of concepts such as time and space. There's a well-known German historian named Oswald Spangler who wrote a very famous book almost a hundred years ago, and he goes very deep into how different civilizations have understood time and space, and in a way, it's a very pluralistic way of understanding this.

And then there's also a philosophical critique that there is really no Kantian subject, that it's just an idealization, and it ignores biological, linguistic, political, social, and technological conditions that are embedded in our minds. And here is something that could be related to what we talked about in this conference.

So, concepts such as identity, trust, these are very human concepts, as well as responsibility. So, do you trust the output of the LLM? That's a question. Do we trust it? Probably not. We always have to double-check and understand the limitations of that system. And the last slide is, again, just a reminder to keep the human in the loop, that it's important in cybersecurity and IAM that the integration of AI is often seen as a silver bullet. Everyone is talking about how they have incorporated AI features, but this can lead to uncritical trust.

It can make us comfortable, and it could make us outsource epistemology to machines. So, one of the main lessons, the main takeaway would be that understanding arises not just from processing data, but from reflecting on the conditions of knowledge itself. And that makes us to be more responsible, to be more critical of ourselves and what we experience. And I think that we are also at a risk of maybe having a new sort of dogmaticism where we will just completely blind trust the output produced by the machine. And as humans, we must remain that transcendental subject.

You know, Kant says, there's only an object to a subject. So, we should be the ones questioning, interpreting, and holding meaning, especially when we have conversations around trust, digital identity, and ethics. And digital identity, I mean, many of you have been coming to EIC for many years in the past, and we know that we're talking about human, well, now we're talking about non-human identities, but the human is there. And the human is the one that creates also these non-human identities, right?

So, just to keep in mind, in crisis scenarios, the human insight is going to be essential for navigating trade-offs, for making context-aware decisions, as well as ensuring resilience in your business. So, I know it was a philosophical session, but I hope that you can take some lessons into how you can use technology in your own private life, as well as in your business.

And, yeah, that's it. Some information. And if you want to talk about it or have any questions, feel free to reach out. Thank you.