Researchers have proposed a synergy between AI and ecology, believing that ecology can make AI more resilient and that AI can address global ecological challenges. Mutual advances could provide innovative solutions that bridge the gap between two different fields.Credit: Barbara Han
In a new paper, scientists explain how ecology can inspire better AI and vice versa, calling for convergence and co-evolution.
Modern artificial intelligence platforms often draw inspiration from the structure and function of the human brain. In a recent study, experts found that looking to another branch of biology, ecology, could pave the way for powerful, resilient, and socially responsible AI systems. I suggest that there is.
Recently published in a magazine Proceedings of the National Academy of SciencesThe paper argues that the synergy between AI and ecology could enhance AI and help solve complex global challenges such as disease outbreaks, biodiversity loss, and the effects of climate change. .
The idea came from the observation that AI is amazingly good at certain tasks, but still useless at others. The development of AI also grew out of the observation that ecological principles hit a wall that could not be overcome.
“The kinds of problems we regularly work on in ecology are not only problems where AI could potentially benefit from a pure innovation standpoint, but if AI can help, It’s also the kind of problem that could have huge implications for the global good,” said Barbara, a disease ecologist at the Cary Ecosystem Institute who co-led the paper with Kush Varshney of IBM Research. Han explained. “It could really benefit humanity.”
How AI can help ecology
Ecologists, including Han Chinese, are already using artificial intelligence to search for patterns in large data sets to determine whether new viruses can infect humans and which animals are likely to carry them. We are making more accurate predictions, such as whether the gender is the highest.
But a new paper argues there is much more potential for AI applications in ecology, such as synthesizing big data and finding missing links in complex systems.
Scientists typically 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, as with most complex ecosystems, predicting disease transmission depends on many variables, not just one. It depends, says co-author Shannon Rado, a disease ecologist at the Carey Institute. Ecologists do not always know what all these variables are, are limited to variables that are easily measured (as opposed to social and cultural factors, for example), and are It is difficult to understand how different variables interact.
“Compared to other statistical models, AI can incorporate larger amounts of data and more diverse data sources, which allows us to discover new interactions and drivers that we didn’t think were important. ,” Rado said. “There is a lot of promise in developing AI that can better capture more types of data, such as sociocultural insights that are very difficult to boil down to numbers.”
In helping uncover these complex relationships and new properties, Rado said artificial intelligence could generate and test unique hypotheses, opening up entirely new areas of ecological research.
How ecology improves AI
Artificial intelligence systems are notoriously fragile and can have devastating consequences, such as misdiagnosing cancer or causing car accidents.
The incredible resilience of ecosystems could inspire more robust and adaptive AI architectures, the authors argue. In particular, Varshney said ecological knowledge could help solve the problem of mode collapse in AI systems that power things like artificial neural networks, speech recognition, and computer vision.
“Modal collapse is when you train an artificial neural network on something, and then you train it on something else, and it forgets what it was originally trained on,” he explained. “A better understanding of why modal collapse does or does not occur in nature may help us understand how to prevent it from occurring in AI.”
A more robust AI inspired by ecosystems could include feedback loops, redundant paths, and decision-making frameworks. These flexibility upgrades could also contribute to more “general intelligence” in AI, allowing algorithms to make inferences and connections beyond the specific data on which they were trained.
Ecology can also help us understand why large-scale AI-driven language models that power popular chatbots such as ChatGPT exhibit novel behaviors that don’t exist in smaller language models. These behaviors include “hallucinations” when the AI generates false information. Because ecology examines complex systems in a holistic manner at multiple levels, it is good at capturing such emerging properties and can help reveal the mechanisms behind such behaviors.
Moreover, the future evolution of artificial intelligence depends on fresh ideas. The CEO of OpenAI, the developer of ChatGPT, said that simply making the model bigger will not lead to further progress.
“Other inspirations will be needed, and ecology offers one avenue to new ways of thinking,” Varshney said.
Towards co-evolution
While ecology and artificial intelligence are progressing separately in similar directions, the researchers say closer and more planned collaboration could lead to as-yet-unimagined advances in both fields. There is.
Resilience provides a compelling example of how both sectors can benefit when they work together. When it comes to ecology, advances in AI in measuring, modeling, and predicting natural resilience can help prepare for and respond to climate change. For AI, a clearer understanding of how ecological resilience works will lead to more resilient AI, better modeling and exploration of ecological resilience, and positive Feedback loops can form.
Closer cooperation also promises to foster greater social responsibility in both areas. Ecologists are working to incorporate diverse ways of understanding the world from indigenous and other traditional knowledge systems, and artificial intelligence can help integrate these different ways of thinking. Finding ways to integrate different types of data could improve our understanding of socio-ecological systems, decolonize the field of ecology, and help correct biases in AI systems.
“AI models are built on existing data, and are trained and retrained when they return to existing data,” said co-author Kathleen Weathers, an ecosystem scientist at the Cary Institute. says. “When we have data gaps that exclude women over 60, people of color, and traditional ways of knowing, we are creating models with blind spots that can perpetuate inequities. .”
Achieving convergence between AI and ecology research requires building bridges between these two siled disciplines, which currently use different vocabularies, operate within different scientific cultures, and have different funding sources. It is necessary to bridge. The new paper is just the beginning of this process.
“I hope it at least sparks a lot of conversation,” Han said.
Investing in the convergent evolution of ecology and AI has the potential to generate innovative perspectives and solutions as unimaginably disruptive as recent breakthroughs in chatbots and generative deep learning, say the authors. is writing. “The implications of successful convergence extend beyond advances in ecology and the achievement of artificial general intelligence, and are critical to surviving and thriving in an uncertain future.”
References: “The Synergistic Future of AI and Ecology” by Barbara A. Han, Kush R. Varshney, Shannon Rado, Ajit Subramaniam, Kathleen C. Weathers, and Jacob Zwart, September 11, 2023. Proceedings of the National Academy of Sciences.
DOI: 10.1073/pnas.2220283120
This research was funded by the National Science Foundation (DBI grant 2234580, DEB grant 2200158), the Carey Institute Science Innovation Fund, and the Lamont-Doherty Earth Observatory Climate and Life Fellowship.