The question was: how intelligent can artificial intelligence be? This paper has shown that so-called artificial intelligence has deficits in comparison with human intelligence. That is nothing new; the question is which deficits are involved:

(1) Understanding and meaning sense is possible on the basis of a pre-understanding, i.e. on the basis of available possibilities of sense that are brought to application in the proposition. An AI system such as ChatGPT does not understand the sense of a sentence and does not mean the sense of a sentence; it knows no semantics and no pragmatics, actually no syntactics either; it does not generate sentences on the basis of rule knowledge, it computes sentence constructs on the basis of statistically probable possibilities of combining sentence building blocks.

(2) A pre-understanding, the available possibilities of sense, is not always sufficient for understanding sense. In this respect a learning process is required in which possibilities of sense are modified or replaced and new ones are found and learned. An AI system such as ChatGPT is conservative, not creative; it learns no new possibilities of sense. Everything that can be new for AI is heteronomous, dependent on the data, patterns and algorithms with which it is fed.

(3) AI can draw inferences in the mode of deduction. This is the principle of the old expert systems, but combinations of expert systems and LLMs are not yet best practice, and in effect deduction is only an application of rule knowledge and leads to no new knowledge. The induction of AI rests exclusively on probability calculus: a probabilistic model is generated quasi-inductively through the statistical evaluation of observations and measurement data. AI cannot draw inferences in the mode of abduction (on which understanding also rests) at all.

(4) An AI system is a self-referential, closed system in which paradoxes are possible and in which self-replication of texts, and in this respect a reduction in the quality of texts, can occur.

What consequences could be drawn from this? The more human beings concede intelligence to artificial intelligence, the more they call human intelligence into question. In this respect the question whether an AI system has a quasi-human intelligence or not cannot simply be filed ad acta, even if this question is in itself meaningless. For the fact that human beings have a consciousness of their freedom, autonomy and responsibility shows that their intelligence must be fundamentally different from that of the machine, however symptomatically and deceptively genuinely intelligent the latter may appear.