Connectionist models of conditioning: A tutorial.

Marr (1989) wrote a tutorial, not an experiment. The paper shows how simple neural networks can mimic Pavlovian conditioning. It argues that the same math explains both computer learning and dog salivation.

The author invited behavior analysts to borrow tools from artificial intelligence. The goal was to make conditioning principles clearer and to help build smarter machines.

The tutorial maps respondent procedures onto network weights. A tone becomes an input node; food becomes a teaching signal. After repeated pairings, the tone alone activates the output node, just like salivation.

The paper claims the setup can be stretched to operant work. Actions that produce reward would strengthen certain connections, letting the network "learn" without rules.

McDowell (2004) and Kulubekova et al. (2013) turned the idea into working code. Their digital organisms replicate matching law curves seen in live pigeons, giving the 1989 sketch real data to stand on.

Gallistel (2025) updates the story. Instead of adjusting weights, Gallistel uses information ratios to predict learning. Both camps agree computation helps, but they fight over the best math.

Demello et al. (1992) push the tutorial into applied AI. They outline software that learns verbal behavior through reinforcement, showing the 1989 dream can move from lab to product.

You can use these models to teach staff why contingencies work. Show a five-minute network animation of a token economy and the visual sticks. If you collaborate with tech teams, pitch behavior-based machine learning instead of rule-based code. The papers give you citations that bridge Skinner and Silicon Valley.