Who's Who in AI
Behind artificial intelligence stand not companies but, first of all, people with very different questions. This overview sorts the most important names into five groups, so that you know who stands for what when they are quoted.
The statistical forefathers
Long before there were computers, the mathematical tools that AI works with today were being developed.
- Thomas Bayes (eighteenth century) formulated how to adjust a conjecture as soon as new evidence comes in. Every learning system does exactly that at its core.
- Pierre-Simon Laplace turned this into a fully worked-out theory of probability and thereby into a usable tool.
- Claude Shannon founded information theory in 1948 and measured for the first time how much information a symbol carries.1 Without that concept there would be neither data compression nor language models.
- Norbert Wiener shaped, with cybernetics, the idea of feedback: a system measures its own result and corrects itself.
The founding generation
- Alan Turing defined in 1936 what computing actually means, and in 1950 recast the question of machine intelligence as a question about observable behaviour.
- John McCarthy christened the field "artificial intelligence" in 1956 and developed the programming language Lisp.
- Marvin Minsky was a co-founder and long-time spokesman of the symbolic, rule-based direction.
- Frank Rosenblatt built the perceptron in 1958, the ancestor of all neural networks.
- Herbert Simon and Allen Newell studied human problem solving and built the first programs that found logical proofs.
The early critics
- Joseph Weizenbaum wrote ELIZA and was alarmed at how readily people credited a simple program with understanding. His warning was directed less at the capabilities of machines than at our willingness to leave decisions to them.2
- Hubert Dreyfus argued philosophically that human skill rests on bodily experience and therefore cannot be captured in rules.3
- John Searle put forward the Chinese Room thought experiment in 1980: processing symbols correctly does not amount to understanding them.4
The neural network generation
This group stuck with neural networks long after the field had written them off.
- Geoffrey Hinton made the training of multi-layer networks practicable in the 1980s, received the Turing Award for it in 2018 and the Nobel Prize in Physics in 2024. Since 2023 he has publicly warned of the risks of the technology he helped bring into being.
- Yann LeCun developed the network architecture that proved decisive for image recognition and takes the relaxed counter-position to Hinton in today's debate.
- Yoshua Bengio shared the Turing Award with both and works today on international reports on AI safety.
- Fei-Fei Li built ImageNet, the image database without which the 2012 breakthrough would not have happened.
- Jürgen Schmidhuber developed methods for processing sequences in Europe and regularly points out how much of it was there far earlier.
The present
Today research stands alongside very large economic interests. Whoever speaks usually also speaks for a lab.
- Demis Hassabis heads Google DeepMind and received the Nobel Prize in Chemistry in 2024 for the protein structure prediction system AlphaFold.
- Sam Altman (OpenAI) and Dario Amodei (Anthropic) lead the two best-known language model labs.
Just as important are the voices that ask about the consequences:
- Emily M. Bender and Timnit Gebru described large language models as "stochastic parrots": systems that reproduce patterns of language without grasping meaning.5
- Joy Buolamwini demonstrated that commercial face recognition fails dramatically more often for dark-skinned women.6
- Kate Crawford investigates the raw materials, energy and labour behind AI systems.7
- Melanie Mitchell and Stuart Russell write accessibly about the capabilities and limits of these systems.
How you should read these names
Three questions help with any quotation: what period does the statement come from? Which discipline is the person speaking from, mathematics, computer science, philosophy or social science? And on whose behalf are they speaking, a university or a company whose valuation depends on the answer? Prominence is not an argument, but the point of view often explains why two intelligent people judge the same system in entirely different ways.