Philosophy of AI: A Structured Overview

Philosophy of AI A Structured Overview Vincent C. Müller

Two point one. Topic and method

Two point one point one. Artificial Intelligence

The term Artificial Intelligence became popular after the nineteen fifty-six "Dartmouth Summer Research Project on Artificial Intelligence," which stated its aims as follows:

The study is to proceed on the basis of the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it.

This is the ambitious research program that human intelligence or cognition can be understood or modeled as rule-based computation over symbolic representation, so these models can be tested by running them on different (artificial) computational hardware. If successful, the computers running those models would display artificial intelligence. Artificial intelligence and cognitive science are two sides of the same coin. This program is usually called Classical AI:

a) AI is a research program to create computer-based agents that have intelligence. The terms Strong AI and Weak AI as introduced by John Searle stand in the same tradition. Strong AI refers to the idea that "the appropriately programmed computer really is a mind, in the sense that computers given the right programs can be literally said to understand and have other cognitive states." Weak AI is means that AI merely simulates mental states. In this weak sense "the principal value of the computer in the study of the mind is that it gives us a very powerful tool."

On the other hand, the term "AI" is often used in computer science in a sense that I would like to call Technical AI:

b) AI is a set of computer-science methods for perception, modeling, planning, and action (search, logic programming, probabilistic reasoning, expert systems, optimization, control engineering, neuromorphic engineering, machine learning, etc.).

There is also a minority in AI that calls for the discipline to focus on the ambitions of (a), while maintaining current methodology under (b), usually under the name of Artificial General Intelligence.

This existence of the two traditions (classical and technical) occasionally leads to suggestions that we should not use the term "AI," because it implies strong claims that stem from the research program (a) but have very little to do with the actual work under (b). Perhaps we should rather talk about "machine learning" or "decision-support machines," or just "automation" (as the nineteen seventy-three Lighthill Report suggested). In the following we will clarify the notion of "intelligence" and it will emerge that there is a reasonably coherent research program of AI that unifies the two traditions: The creation of intelligent behavior through computing machines.

These two traditions now require a footnote: Both were largely developed under the notion of classical AI, so what has changed with the move to machine learning? Machine learning is a traditional computational (connectivist) method in neural networks that does not use representations. Since approximately twenty fifteen, with the advent of massive computing power and massive data for deep neural networks, the performance of machine learning systems in areas such as translation, text production, speech recognition, games, visual recognition, and autonomous driving has improved dramatically, so that it is superior to humans in some cases. Machine learning is now the standard method in AI. What does this change mean for the future of the discipline? The honest answer is: We do not know yet. Just like any method, machine learning has its limits, but these limits are less restrictive than was thought for many years because the systems exhibit a non-linear improvement - with more data they may suddenly improve significantly. Its weaknesses (e.g., overfitting, causal reasoning, reliability, relevance, and black box) may be quite close to those of human rational choice, especially if "predictive processing" is the correct theory of the human mind (Sections two point four and two point six).