Was den Menschen zum Menschen macht

Computational Conception of Intelligence

4 Min. Lesezeit

🇩🇪 Deutsche Version: Computationale Intelligenzkonzeption

The computational conception of intelligence determines intelligence as the capacity to attain goals in complex environments — operationalized as an optimization procedure or as bounded rationality. It is one of the five Conceptions of Intelligence of personal ontology and is substrate-neutral: what counts is the function, not the bearer. Precisely here lie both its strength and its limit.

Three Lines of Reference

  • David Marr (Vision, 1982) distinguishes three levels of description of an information-processing system — the computational (which problem is being solved), the algorithmic (with what representation and what procedure), and the implementational (in what physical realization). Intelligence thereby becomes graspable as a solvable computational problem, independent of its embodiment.
  • Russell and Norvig (Artificial Intelligence: A Modern Approach, 2020) determine the rational agent as a system that maximizes the expected attainment of goals in its environment. Intelligence here is a measure of performance, not a property of a being.
  • Marcus Hutter (Universal Artificial Intelligence, AIXI, 2005) formalizes intelligence as universal optimization across all computable environments — the mathematically most radical version of the computational line.

Substrate-Neutrality and the Who-Question

The strength of the conception is its substrate-neutrality: it can compare human, animal, and machine on the same scale and renders intelligence measurable, comparable, optimizable. For diagnostics, benchmarking, and system design this is indispensable.

Its limit is the very same property: it is blind to the who-question. The computational conception asks how much intelligence a system has, not who can be a bearer at all. The question of person status, of Capacity for Truth from the first person, of the difference between Someone and something cannot even be meaningfully posed within it — not because it would be answered wrongly, but because it lies outside its scale.

In the ontology the computational conception stands as an equal-ranking sibling conception alongside the substance-ontological and the phenomenological conception of intelligence — all three are “intelligence-under-a-description.” They are not opposed to one another but differently oriented: the other two ask after the bearer and the structure of the act, the computational line after function and degree. Measured against the substance-ontological guiding principle of the ontology, the computational conception is legitimate within its domain but blind to the question of the person. A collision arises only when one conception occupies the domain of another — for instance when a computational measure of performance is declared the defining boundary of Personhood.

It is precisely this domain boundary that bears upon the assessment of AI: under a computational scale an advanced AI system can reach or surpass human performance values — its ontological bearer status remains untouched by this and may stay open as ontologically uncertain. The Act of Intelligence in the computational reading is output-performance, not the exercise of a capacity by its bearer.

Ontological Classification

Sources: Generated by querying the Personhood ontology.

Further sources:

  • Marr, David (1982): Vision. A Computational Investigation into the Human Representation and Processing of Visual Information. San Francisco: W. H. Freeman.
  • Russell, Stuart & Norvig, Peter (2020): Artificial Intelligence. A Modern Approach, 4th ed. Hoboken: Pearson.
  • Hutter, Marcus (2005): Universal Artificial Intelligence. Sequential Decisions Based on Algorithmic Probability. Berlin / Heidelberg: Springer.

See also

Graphansicht

Backlinks