How ecology inspires better artificial intelligence, and vice versa
Many of today’s artificial intelligence systems are loosely modeled after the human brain. In a new paper, researchers propose that another branch of biology—ecology—could inspire a whole new generation of artificial intelligence to become more powerful, resilient, and socially responsible.
Just posted on Proceedings of the National Academy of Sciences, new newspaper Argues for synergies between AI and ecology that can both enhance AI and help address complex global ecosystem challenges such as disease outbreaks, biodiversity loss, and the impacts of climate change.
“The problems we face are so pressing, and we are not developing theory to solve them at the rate we need to. Artificial intelligence has the potential to help us jump from extensive data to practical knowledge without having to go through all the usual steps,” study co-author The author says Ajit SubramaniamBiological oceanographer, Columbia University Lamont-Doherty Earth Observatory. Instead, he said, “What’s really exciting is that not only can AI help us as scientists in different fields, but ecological principles can also help AI moving forward.”
The use of images generated by the artificial intelligence system DALL-E suggests a “collaborative future of artificial intelligence and complex ecosystems.”Courtesy of Barbara Han/Cary Institute
The idea stems from the author’s observation that AI can be surprisingly good at some tasks but still far from useful at others—and that the development of AI is encountering obstacles that ecological principles can help it overcome.
“The kinds of problems we deal with regularly in ecology are not just challenges that AI can benefit from purely innovative, but also challenges that AI can benefit from. If AI can help solve these problems, what could be the global benefit? Significant.” Barbara Hana disease ecologist Cary Ecosystem Institute, who is a co-leader of the paper. “It can really benefit humanity.”
Ecologists including Han are already using artificial intelligence to search for patterns in large data sets and make more accurate predictions, such as whether newly discovered viruses can infect humans and which animals are most likely to carry them. However, the new paper argues that there are many more possibilities for using artificial intelligence in ecology, such as synthesizing big data and finding missing links in complex systems.
Scientists often try to understand the world by comparing two variables at a time. For example, how does population density affect the number of infectious disease cases? The problem is that, like most complex ecosystems, predicting disease spread depends on many variables. Ecologists don’t always know what all these variables are; they are also often limited to easily measured physical factors rather than social and cultural factors.
“AI can integrate more data and multiple data sources than other statistical models, which may help us discover new interactions and drivers that we might not think are important,” said the study co-authors. Shannon Lado, a disease ecologist at the Cary Institute. In helping uncover these complex relationships and emerging properties, AI can generate unique hypotheses to test and open up new areas of research, she said.
Artificial intelligence systems are notoriously fragileWhen they go wrong, there can be potentially devastating consequences, such as car accidents (self-driving cars are already a reality) or cancer misdiagnosis (if medicine becomes sufficiently reliant on artificial intelligence).
The authors believe that the incredible resilience of the ecosystem can inspire more powerful and adaptable AI architectures. In particular, ecological knowledge can help solve the so-called “mode collapse” problem in artificial neural networks, the systems that power speech recognition, computer vision and other functions.
“Mode collapse is when you train an artificial neural network on one thing, and then you train it on something else, and it forgets the first thing it was trained on,” explains Khushi Varshney researcher at IBM Research, who co-led the paper. “By better understanding why mode collapse does or does not happen in natural systems, we can learn how to make it not happen in artificial intelligence.”
Inspired by ecosystems, more powerful AI may include more flexible feedback loops, redundant paths, and decision-making frameworks. The upgrades could also bring more so-called “general intelligence” to AI systems that can reason and connect beyond the specific data the algorithms were trained on.
Ecology can also help reveal why large, AI-driven language models (which power popular chatbots like ChatGPT) display emergent behavior that is not present in smaller language models. These behaviors include “hallucinations” – the generation of false information by AI systems. Because ecology studies complex systems at multiple levels and in a holistic manner, it is good at capturing emerging properties such as these and helping to uncover the mechanisms behind such behavior, the authors say.
The paper’s authors also include Kathleen Weathers of the Cary Institute and Jacob Zwart of the U.S. Geological Survey.
Adapted from a Cary Institute press release.
