Creationism in Crisis - Pigeons Can Solve Complex Problems More Quickly Than Humans

Feral pigeons can solve complex categorization problems that would trouble humans.

‘Dim-witted’ pigeons use the same principles as AI to solve tasks

The feral pigeon, Columba livia domestica. Dimwit or genius?

In yet another casual and unintentional refutation of creationist dogma, two psychology professors, one from Ohio State University and the other from the University of Iowa, have shown that humans were not created as the species with the highest intelligence, which creationists argue shows we were specially created, different to the other animals.

The researchers have shown that pigeons can solve complex problems that would be difficult for humans because pigeons use the same principles as artificial intelligence to crunch through the available options. The difference being that unlike humans who can extrapolate from solving one problem to general problem-solving similar problem, pigeons don't extrapolate from the particular to the general.

It appears then that the natural process of evolution produced a form of AI, long before humans developed machine learning and AI.

As explained in an Ohio State University news release by Jeff Grabmeier:

A new study provides evidence that pigeons tackle some problems just as artificial intelligence would – allowing them to solve difficult tasks that would vex humans.

Previous research had shown pigeons learned how to solve complex categorization tasks that human ways of thinking – like selective attention and explicit rule use – would not be useful in solving.

Researchers had theorized that pigeons used a “brute force” method of solving problems that is similar to what is used in AI models, said Brandon Turner, lead author of the new study and professor of psychology at The Ohio State University.

But this study may have proven it: Turner and a colleague tested a simple AI model to see if it could solve the problems in the way they thought pigeons did – and it worked.

We found really strong evidence that the mechanisms guiding pigeon learning are remarkably similar to the same principles that guide modern machine learning and AI techniques. Our findings suggest that in the pigeon, nature may have found a way to make an incredibly efficient learner that has no ability to generalize or extrapolate like humans would.

Professor Brandon M. Turner, lead author. Professor of Psychology
The Ohio State University, Columbus, OH, USA.

Turner conducted the study with Edward Wasserman, a professor of psychology at the University of Iowa. Their results were published recently in the journal iScience.

In the study, pigeons were shown a stimulus, which could include lines of various widths and angles, concentric rings and sectioned rings. They had to peck a button on the right or left to indicate to which category it belonged to. If they got it correct, they received a food pellet – if they were wrong, they received nothing.

There were four different tasks in the study, some harder than the others. Results showed that, through trial and error, the pigeons improved their ability to make the correct choices in one of the easier experiments from about 55% to 95% of the time. Even in a more difficult scenario, their correct responses improved from 55% to 68%.

Researchers believed the pigeons used what is called associative learning, which is linking two phenomena with each other. For example, it is easy to understand the link between “water” and “wet.” People teach their dogs to link sitting when they are commanded with receiving a treat.

But those associations are relatively easy.

“Associative learning is frequently presumed to be far too primitive and rigid to explain complex visual categorization like what we saw the pigeons do,” Turner said.

But that’s exactly what the researchers found.

The researchers’ AI model tackled the same tasks using just the two simple mechanisms that pigeons were presumed to use: associative learning and error correction. And, like the pigeons, the AI model learned to make the right predictions to significantly increase the number of correct answers.

For humans, the challenge when given tasks like those given to pigeons is that they would try to come up with a rule or rules that could make the task easier.

But in this case, there were no rules that could help make this any easier. That really frustrates humans and they often give up on tasks like this. Pigeons don’t try to make rules. They just use this brute force way of trial and error and associative learning and in some specific types of tasks that helps them perform better than humans.

Professor Brandon M. Turner

What’s interesting, though, is that pigeons use this method of learning that is very similar to AI designed by humans, Turner said.

We celebrate how smart we are that we designed artificial intelligence, at the same time we disparage pigeons as dim-witted animals, but the learning principles that guide the behaviors of these AI machines are pretty similar to what pigeons use.

Professor Brandon M. Turner

The research is published, open access in the online journal, iScience:

Highlights

• Pigeons can solve an exceptionally broad range of visual categorization tasks.
• Some of these tasks seem to require advanced cognitive and attentional processes.
• Yet, computational modeling indicates that pigeons don’t deploy such complex processes.
• A simple associative mechanism may be sufficient to account for the pigeon’s success.

Summary

Never known for its smarts, the pigeon has proven to be a prodigious classifier of complex visual stimuli. What explains its surprising success? Does it possess elaborate executive functions akin to those deployed by humans? Or does it effectively deploy an unheralded, but powerful associative learning mechanism? In a series of experiments, we first confirm that pigeons can learn a variety of category structures – some devised to foil the use of advanced cognitive processes. We then contrive a simple associative learning model to see how effectively the model learns the same tasks given to pigeons. The close fit of the associative model to pigeons’ categorization behavior provides unprecedented support for associative learning as a viable mechanism for mastering complex category structures and for the pigeon’s using this mechanism to adapt to a rich visual world. This model will help guide future neuroscientific research into the biological substrates of visual cognition.

Graphical Abstract

Turner, Brandon M.; Wasserman, Edward A.
The pigeon as a machine: Complex category structures can be acquired by a simple associative model
iScience 26(10). doi: 10.1016/j.isci.2023.107998

Copyright: © 2023 The authors.
Published by Elsevier. Open access.
Reprinted under a Creative Commons Attribution 4.0 International license (CC BY 4.0)

The conclusion then is that intelligence, which has arisen independently in different taxons - mammals, birds and cephalopods (octopuses) - is an evolved trait just like everything else, and in some cases such as these pigeons, it can surpass that of humans. It should not have been beyond the wit of an intelligent designer to give humans this ability too, yet, if you believe the creation myth, it chose not to, so humans have a sub-optimal intelligence - not the perfect creation of a perfect creator, but the result of a utilitarian natural process.

Only by remaining carefully ignorant of this sort of science can creationists maintain the absurdly childish delusion that they are the special creation of a magic man in the sky who holds them in especially high regard.

Incidentally, while researching for this article, I asked ChatGPT3 to explain how AI solves problems:

How does artificial intelligence solve problems more quickly than humans can? Artificial intelligence (AI) can solve certain types of problems more quickly than humans for several reasons:

1. Speed and Efficiency: AI systems, especially when implemented on powerful hardware and parallel processing, can perform computations and data processing much faster than humans. They can process vast amounts of data and perform complex calculations in milliseconds or microseconds.
2. Consistency: AI systems do not get fatigued, distracted, or make mistakes due to boredom, stress, or other human factors. They can maintain a high level of consistency and accuracy in problem-solving tasks.
3. Handling Large Datasets: AI can quickly analyze and draw insights from massive datasets, something that would be incredibly time-consuming for humans. Machine learning algorithms can identify patterns, correlations, and trends that might be impossible for humans to spot in large datasets.
4. Multitasking: AI systems can efficiently multitask, handling multiple threads of computation simultaneously. For example, they can monitor multiple sensors, analyze data from various sources, and make decisions across different tasks concurrently.
5. Rapid Learning: Machine learning models can be trained on vast amounts of data to learn and adapt to new information, making them capable of solving problems faster once they have been trained.
6. Iterative and Repetitive Tasks: AI is particularly suited for tasks that require repetitive calculations or iterations, such as optimization problems, simulations, or numerical analysis. AI algorithms can iteratively refine solutions until they converge to an optimal answer.
7. Complex Algorithms: AI can employ complex algorithms, including search algorithms, optimization techniques, and heuristic methods, to solve problems that involve a large number of variables or complex relationships among them.
8. Automation: AI can automate various processes and decision-making, enabling problem-solving in real-time without the need for human intervention. This is especially useful in domains like robotics, autonomous vehicles, and industrial automation.
9. Pattern Recognition: AI excels at recognizing patterns in data, images, and other types of information. This ability is valuable for problems like image and speech recognition, fraud detection, and predictive analytics.
10. Scalability: AI solutions can be easily scaled up or down to handle a wide range of problem sizes. Whether it's processing a small dataset or analyzing big data, AI can adapt to the requirements of the task.

It's important to note that AI's ability to solve problems more quickly than humans is context-dependent. AI is best suited for problems that can be formulated as algorithms, tasks that require rapid data processing, and problems where the solution can be learned from data or programmed explicitly. However, AI may not excel in domains that require common-sense reasoning, intuition, creativity, or deep understanding of human emotions and social dynamics, which are areas where humans still have the upper hand.

Can AI lean from problem solving by extrapolating from the specific to the general? Yes, artificial intelligence (AI) can learn from problem solving by extrapolating from specific examples to form generalizations. This process is a fundamental aspect of machine learning, specifically supervised learning and reinforcement learning. Here's how it works:

1. Supervised Learning: In supervised learning, AI systems are trained on a labeled dataset consisting of input-output pairs. The AI model learns to map inputs to outputs by generalizing from the examples in the training data. It then uses this learned generalization to make predictions on new, unseen data. For example, a supervised learning model can learn to recognize cats in images by training on a dataset of labeled cat images and non-cat images. It generalizes from this training data to identify cats in new images.
2. Reinforcement Learning: In reinforcement learning, AI agents learn by interacting with an environment and receiving feedback in the form of rewards or penalties. They generalize from specific experiences to make better decisions in the future. Over time, they learn to take actions that maximize their cumulative rewards. For instance, a reinforcement learning agent can learn to play a video game by trying different actions and generalizing from its past experiences to make better moves in similar situations.
3. Transfer Learning: Transfer learning is a technique where AI models are pre-trained on one task and then fine-tuned on another related task. The knowledge gained during pre-training is transferred to the new task. This is a way to generalize knowledge from one specific problem to a more general set of related problems.
4. Inductive Reasoning: AI can use inductive reasoning to generalize from specific examples to broader principles. For instance, it can identify patterns and trends in data to make predictions about future events.
5. Data-driven Generalization: AI systems can generalize from data by identifying common patterns, correlations, and relationships. For example, a natural language processing model can learn grammar rules by analyzing a large corpus of text, allowing it to generalize and generate coherent sentences.

It's important to note that the quality of the generalization depends on the quality and diversity of the training data. If the AI system is exposed to a wide range of relevant examples, it is more likely to make accurate generalizations. However, AI may not always generalize correctly, especially when faced with data that is significantly different from what it has seen during training. This is a common challenge in machine learning, known as the problem of overfitting (learning the training data too well) or underfitting (failing to generalize effectively). Balancing generalization and specificity is a key aspect of AI model training.

---

Thank you for sharing!

Tweet Link