Luca Paltrinieri : “Intelligence has always been artificial”

To get your hands dirty and look under the bonnet of artificial intelligence, you need at least one philosopher. Luca Paltrinieri is Professor of Political Philosophy and Philosophy of Humanities and Social Sciences. In five questions, he turns our perceptions of intelligence, human or otherwise, on their head.

Fabien Seraidarian
Director of Knwoledge Transfer and of the Global Executive MBA, SKEMA Business School

Kevin Erkeletyan

Is artificial intelligence trying to imitate human intelligence?

More than imitating it, it’s surpassing it. In the 1950s, with the work of mathematician Claude Shannon and cybernetics, the first idea was to reproduce human behaviour. But after the f irst “AI winter”, mainly caused by lack of funding, a new model of AI is developing: we are trying to reproduce a task performed by an intelligent human, but in a completely different way. The central concern lies in looking for correlations in a large set of data and, from there, predict possibilities: this is a form of predictive intelligence, as well as being instructive. When Deep Blue won against Kasparov in 1997, it was clear that the computer was not imitating human behaviour, but seeking, though its computing power, to obtain a result that could be human using non-human means.

So is human intelligence seen as a calculating machine?

Yes, and that’s the real crux of the matter: why has human intelligence become all about calculation? In the 19th century, the idea that everyone made cost-benefit calculations was the basis of utilitarianism. Babbage, for example, initially wanted to create “human calculators” by combining the intelligence of several people (called computers) to obtain greater calculating power. The famous mechanical calculator (Difference Engine) he developed was intended to replace this kind of collective social intelligence: the machine simply imitated and replaced the division of labour involved in calculating. For Matteo Pasquinelli (The Eye of the Master, Verso, 2023), this is the true origin of the connectionist model that we f ind, for example, in the famous “neural network” of contemporary AI: the nodes in the network are not neurons; they were originally humans. But before we could call computing power “intelligence”, we first had to “rediscover” this human capacity for calculation and give it a positive meaning, as shown by the work of Stephanie Dick, among others. AI derives from the automation of this calculating social intelligence.

Are you saying that AI is based on not an imitation of our brains but our social behaviour?

Yes, this connectionist model also tells us something about the way we conceive the market – as a set of nodes connected by a system of relationships. If you look at a map of the British Empire in the 19th century, you’ll see trading posts connected by shipping routes. Later, in the 1930s, Hayek saw markets as self-teaching, self-organising systems where it was impossible to know what decision is made by each player. On the other hand, he believed that there was a rationale in the results, i.e. price formation, which arose from the interaction of all the players. In this way, through a pricing system, we can express a kind of collective intelligence: the intelligence of players who are in contact with each other. It’s no coincidence that Frank Rosenblatt, the psychologist who invented the f irst “artificial” neural network (the Perceptron), studied Hayek: he thought we could control the machine’s learning by randomly changing the weight of the neuron nodes, depending on the results obtained. Today, with deep learning models combining several layers of neurons, computer scientists don’t know how interactions are formed, i.e. what is really going on under the bonnet; they only look at the final result.

Why has human intelligence become a matter of calculation?

Luca Paltrinieri, Researcher in Philosophy

So is “algorithmic thinking” based on a form of mystery?

I wouldn’t talk of “artificial thinking” or “an artificial mind.” Generative AI – for example LLM (ChatGPT 3 or 4), certainly produces good results: the machine “speaks” but obviously it doesn’t know that it is speaking. I would even say that there is no “it” because the machine is not a subject. The machine returns a result by associating data on the basis of a statistical inductive model: it is a kind of linguistic skill devoid of comprehension. In a certain way, knowledge is produced in the form of a prediction, but it is a prediction for us, which we perceive as the automated production of knowledge. The confusion with human intelligence comes from the presupposition of unknowability: since we can’t really describe what’s going on in ChatGPT’s “brain”, we equate AI with human intelligence, whose workings are also mysterious. Yet there is a fundamental difference: we have always sought to produce definitions of human intelligence, whereas it isn’t necessary to describe the way ChatGPT works, since only the result counts.

But is “intelligence” the right word to describe artificial intelligence?

If we define intelligence as a skill, an ability to produce a result by adapting to certain circumstances: yes, it is intelligence. We could think, like the structuralists, that intelligence is a property emerging from a large number of correlations between data, rather than a system of rules describing the way in which an action is carried out. But the very principle of correlation – the idea that there is additional meaning in the connection between two elements – comes from symbolic thought, which is specifically human. From this point of view, current systems, algorithms and even generative AI do not add anything to thought; they simply project it outside mankind into media that have always enabled knowledge to be socialised through the construction of a collective memory. Today, AI also seems to be a response to a quest for objectivity: the machine seeks to be more neutral; it must be able to take us to an overarching, universal point of view. In HR, for example, the idea is to use AI to eliminate human bias in recruitment. But even this quest for objectivity only perpetuates the effort to construct artificial organs, supplements external to human beings, which began with cave drawings and continued with the invention of writing. Intelligence has always been artificial.