AI and intelligence: aren't humans also token parrots?

Many people are frightened by AI. Nobody understands exactly how it works. Some say that AI is "only" based on statistics and, in the case of language models, merely spits out the most likely next word or part of a word (token). Intelligence is thus subliminally or deliberately denied to "AI". What is the point?

What is intelligence?

This question is not easy to answer. The naive approach would be to refer to humans as a yardstick. This is completely unjustified. Even just mentioning animals that can display intelligent behavior (in our eyes) shows the problem with a human-centered perspective.

The EU's AI-Regulation defines intelligence itself not, but tries to define Artificial Intelligence:

System of artificial intelligence“ (AI system) a machine-supported system, designed to operate with varying degrees of autonomy and produce results such as predictions, recommendations or decisions that can influence the **physical or virtual environment

Source: Article 3 No. 1 AI Regulation of 14.06.2023 (bold added)

The latest accepted version is slightly different, but qualitatively just as practical distant. The version in the previous version from 21.04.2021 still had humans as a yardstick in the definition.

These EU definitions offer room for criticism. For example, they question whether ChatGPT is intelligent. However, ChatGPT was probably the first (prominent) AI system to pass the Turing Test and thus imitated intelligent behavior indistinguishable from that of a human.

In addition, the EU sometimes assumes objectives in their definition that are not present in most generic programs, at least not in a task-specific manner. The fact that an artificial system must be machine-supported is not only doubtful, but must be described as false.

That would be like wanting to ban diesel just because one can't imagine that diesel could also be produced in an environmentally friendly way… However, I lack the imagination of how a very expensive and many drawbacks afflicted battery of the current technology level should bring about the traffic turnaround as a battery.

Further Criticisms* of the EU's concept of intelligence and more details can be found in a separate article on the definition of Artificial Intelligence. There, a new definition for the term "Artificial Intelligence" is also proposed.

What is the difference between artificial and non-artificial intelligence?

Most people can't seem to answer this question or don't want to. The answer is: artificial intelligence is artificial and non-artificial intelligence is non-artificial. That's all it is. Now you can argue at length about what artificial means exactly. Just decide for yourself. It is irrelevant to the question of whether something or someone is intelligent or not.

It's not about how and why intelligence occurs exactly. There is only one particularly outstanding criterion that should be considered a feature of intelligence: The result.

In order for a random generator not to be considered intelligent, something still has to be refined. Similarly, it often does not even come down to the final result in the form of a final answer. For example, only the so-called "Assessment Center" is mentioned here. In this case, applicants are invited to have their suitability for a specific position in the company tested. Some tasks are set up in such a way that they cannot be solved within the available time and with the available information. Nevertheless, the tasks are set. Why?

It's about how the candidate wants to solve the task. "How" means it's about the solution path, namely the one that is visible from outside. "Wants" means it's not primarily about presenting a final result.

This leads to the following definition of intelligence and thus also of Artificial Intelligence:

Next follows the definition of Artificial Intelligence:

Replace "system" with "artificial system".

It's very simple, inspired by Simyo.

Is AI a token parrot or not?

The question is irrelevant in itself. See definition of intelligence or artificial intelligence above.

Let's answer the question with "Yes" and ignore the fact that this question is irrelevant. Isn't a human then also a token-parrot? A token, by the way, is a short word or a word part. Tokens are used as a form of compression so that AI-language models work more efficiently.

How does the human brain work?

No idea. Nobody knows exactly. There is a neural network that converts all input signals into analogue signals and processes them. The result are again analogue signals.

How does an artificial brain work?

I have no idea. Nobody knows exactly. There is a neural network that converts all input signals into digital signals and processes them. The result is digital signals again.

That was no copy & paste error. The fact that both statements sound identical is intentional. One just has to be willing to accept that analog and digital are two systems that can be converted into each other.

Back to the human brain (we are not talking about our bodies here; if we are, then we are talking analogously about robots with a built-in computer; this will be a frightening reality in a few years' time).

A 2021 essay reads: Can the Brain Build Probability Distributions?. The underlying scientific investigation in humans aimed to determine "whether the brain can build probability distributions to represent experienced data and use this global representation to evaluate the likelihood of new information