What is Artificial 'Intelligence'?

Artificial intelligence is often treated as a modern concept, associated with computers, machine learning, neural networks, and large language models. Historically, however, the idea is much older. At its broadest, “artificial intelligence” can be understood as the attempt to create artificial systems that can perform tasks associated with human cognition: performing limited forms of reasoning or calculation.

Throughout history this ambition has lead us to ancient counting devices, formal systems of logic, mechanical automata, calculating machines, the programmable digital computer in the 1940s, and contemporary neural networks. All of these systems express the same basic ambition: to externalize aspects of human thought into tools and machines, allowing a non-human system to carry out operations that previously seemed to require the human mind.

Yet throughout its history, the meaning of “intelligence” has never been fixed. In ordinary language, intelligence usually suggests some kind of flexible ability to adapt to new circumstances, and solve unfamiliar problems. In a more technical sense, intelligence often refers to the ability to process information and produce useful outputs.

This precise definition matters, because it shapes how we interpret the achievements of artificial systems. A mechanical calculator can perform arithmetic better than most humans, and though people may have referred to it in the past as a 'mechanical intelligence', we usually think of calculators as tools that follow a procedure. We typically no longer describe them as intelligent systems.

A new layer of programmability

Over the 2020s we have made amazing progress in the way computational systems can learn abstract representations from text and image data. These new systems work so well, that they can generate images, essays software, summaries, translations, and conversations that are often difficult to distinguish from human-created work.

At the same time, these systems remain deeply dependent on description, structure, and context. They are powerful when users can specify what they want in sufficient detail, follows well established existing patterns.

This is visible in practices such as “vibe coding,” where people use AI systems to generate applications or prototypes through natural-language instructions. The process feels radically new, because users no longer need to write every line of code themselves. But it also reveals a continuity with earlier forms of programming: the system still needs detailed and consistent instructions, constraints, examples, and goals.

In this sense, modern AI systems may be understood better as a new kind of programmable system (or Turing machine), rather than as a new kind of 'intelligence'.

Traditional computers required precise formal instructions written in programming languages, which were then translated by interpreters or compilers into machine code and processor instructions. Today’s LLM-based systems form a layer on top of these older systems, that can now be guided through ordinary language. This may feel like a new level of intelligence, as if we have finally created an artificial brain that can understand human language. But under the hood, we are still dealing with the same computational systems as before.

The main difference is that we have created artificial systems that can translate ordinary language into computational actions.

A programmer no longer needs to express every idea in detailed computer syntax that appears unintelligible to outsiders. In principle, anyone can now try to control a computer using plain English. But how well this works still depends on how clearly and precisely the user can describe what they are trying to achieve into ordinary language.

In a way this is very similar to what we have been used to with programmable digital computers.

In a similar way, these models, and pipelines that we build with them are programmable. When we give them detailed instructions, they can follow them. And the 'creature' that comes out of this is like a Turing machine.

What is 'Intelligence'?

This does not make the current wave of AI insignificant. On the contrary, it may become one of the major technological breakthroughs of the twenty-first century, comparable in impact to earlier waves of computerization and automation. Just as programmable computers made it possible to automate vast numbers of administrative, scientific, industrial, and creative tasks, these new AI systems make it possible to automate or accelerate many forms of linguistic and visual work. The consequences may be profound.

But the central philosophical question remains unanswered: is this the same as human 'intelligence'?

To answer that question, we first need to ask what we mean by intelligence. Like consciousness, intelligence is an unstable and partly ill-defined concept. We use it confidently in everyday language, but its precise meaning depends a lot of the cultural and historical context in which it is used. Different definitions lead to very different conclusions about whether artificial systems are intelligent.

Historically, we often reserve the word “intelligence” for abilities that still seem uniquely human. Once a task can be outsourced to a machine, we tend to reinterpret it as procedural rather than truly intelligent. Calculation is a good example. Before mechanical calculators and digital computers, arithmetic was closely associated with human reasoning. Once machines could perform it reliably, it came to seem like a mechanical procedure rather than a sign of intelligence.

Simply put, we have no idea what we mean when we are exactly talking about when we refer to 'human intelligence'.

I think we will need to go through some kind of paradigm shift in the 21th century, where we will more sharply distinguish between different kinds of 'intelligence', and what we mean by them.

AI might not solve the puzzle of intelligence. But sharpens it. Not that it shows us what machine intelligence is, but that it forces us to reconsider what we thought human intelligence was.

So that we will be able to more accurately distinguish what we mean with 'human intelligence', and how that differs from what we mean when we talk about 'artificial intelligence'

Next sections

In the following sections I will do my best to summarise my thoughts on this topic as well as possible, by trying to go over the following topics

• 1. how does human intelligence relate to language?
• 2. what is language? how does language describe the world? to what extend do general AI systems depend on human language?
• are there limitations to how AI based systems can ?
• to what extend do LLM-based systems depend on human language?
• what is 'computability' (and can we validate statements made in language using computations?)
• can everything that can be expressed in terms of language also be computed?