This is largely in line with the work of another psychologist, Robert Lescollla. Robert Lescollla's work in the 70s and 80s influenced both Wasserman and Sutton. Rescorla encouraged people to think of people as “learning arising from exposure to relationships between events within their environment” and “the main means of representing the structure of the world,” rather than as “a low-level mechanical process.”
This also applies to laboratory pigeons where scientists are pasting screens and buttons in small experimental boxes where they carefully control and measure stimuli and rewards. However, pigeon learning spreads outside the box. Wasserman students transport birds in buckets between birds and the lab, and each time students open the door, experienced pigeons quickly jump into the bucket. As Rescorla suggested, they learn the relationship between the structure of the world in the lab and the parts like buckets and boxes, but they don't always know the specific tasks they face inside.
Comparative psychologists and animal researchers have been working on questions that suddenly seem urgent for AI.
The same association mechanisms that pigeons learn about the structure of their world allow them to open a window into the inner life, as Skinner and many early psychologists said. Pharmaceutical researchers have long used pigeons in drug discrimination challenges. For example, they are given amphetamines or sedatives, and food pellets are given to correctly identify which medications were taken. The success of birds suggests that they experience and discriminate against internal states. “Is that not equal introspection?” asked Wasserman.
It is difficult to imagine AI matching pigeons in this particular task. This reminds us that despite AI and animals share an association mechanism, there is more to life than to behaviour or learning. Doves deserve ethical considerations as living creatures. While pigeons can experience and suffer pain, AI chatbots cannot otherwise fool people and believe them, even in some large-scale language models trained in corpus that contain explanations of human suffering and sensory computer science fiction stories.
University of Iowa/Wasserman Lab
“Intensive public and private investment in AI research in recent years has brought about the very technology that forces us to stand up to today's AI sense issues,” two science philosophers wrote. ion 2023 answer These current questions require similar investments in animal cognition and behavioral research. “Indeed, comparative psychologists and animal researchers have been working on questions that suddenly seem urgent for AI.
Such work will bring knowledge not only about technology and animals, but also about ourselves. Most psychologists probably won't go all the way to Sutton by claiming that it's a reward enough to explain most, if not all human behavior, but people often learn from the association, too. In fact, most of Wasserman's undergraduates ultimately succeeded in a recent experiment with Striped Discs, but only after they gave up on searching for rules. They relied on the association like doves, and then they could not briefly explain what they had learned. With plenty of practice, they began to get the feel of the category.
That's another irony about associative learning. For a long time, the most complex form of intelligence, namely cognitive abilities like rule-based learning, can make us human, but we also call it for the easiest tasks, such as sorting objects by color or size. On the other hand, some of the most sophisticated demonstrations of human learning are only learned through experience, not through rules, but through experience, for example, learning to taste differences between grapes.
Learning through experience relies on ancient association mechanisms that we share with pigeons and countless other creatures, from bees to fish. Laboratory pigeons are not only in our computers, but in our brains, in the engines behind humanity's most impressive feat.
Ben Claire is a Berlin-based science and travel writer.
