How AI is revealing the hidden intelligence of plants and fungi

Machine learning tools are allowing scientists to better understand forms of intelligence that operate on spaces and timescales utterly unlike our own.

By Conor Feehly

If you’ve ever been on a long-haul flight, you’ve likely suffered the consequences of disrupting your circadian bodily rhythm. Usually, for a day or two, your mood will plummet, your body will ache, and your mind won’t be as sharp as normal. To remedy jet lag, you might take steps to recalibrate. An early night. A nourishing meal.

Surprisingly, plants can also experience the discombobulating aspects that we feel with jet lag. Like a wide range of biota, plants use internal biological clocks to synchronize their cellular activities, such as enhancing their photosynthetic capacities, with their local day-night cycles. But when this pattern gets disrupted, perhaps during transportation or an abrupt seasonal change, the plant’s clock falls out of sync with its environment. As a result, its growth might slow to a halt, it will consume fewer nutrients, and its ability to extract energy from sunlight will diminish.

This “plant jet lag” has caught the attention of the Minimal Intelligence Lab (MINT Lab) at the University of Murcia in Spain. Specifically, they are interested in circadian rhythm disruption in common beans: Phaseolus vulgaris. The researchers have been using machine learning tools to map the internal electrophysiological signals that regulate and coordinate a number of crucial functions in tandem with the plants’ light-dark cycles.

The development and deployment of machine learning tools have allowed researchers like those at the MINT lab to gain insight into the behavioral capacities of species that operate on vastly different timescales and spatial dimensions compared to our own. And perhaps more broadly, they are opening a window for us to perceive hidden forms of intelligence that have surrounded us all along.

Across the life sciences, a growing movement is rethinking cognition, intelligence, and communication in living systems long considered to be more like passive, mechanical automatons than dynamic, agential lifeforms. Rather than treating these traits as exclusive to animals with nervous systems, researchers have been investigating mind-like properties in plants, fungi, slime molds, microbial communities, and other distributed intelligences.

Behavior in slow motion

Paco Calvo, who runs the MINT lab, is a leading researcher in the field of plant cognition. In his book Planta Sapiens, he’s argued that the presence of behaviors like memory, decision-making, learning, and communication in plants has forced us to loosen some of our animal-centered assumptions. For instance, the touch-sensitive creeping herb Mimosa pudica will preemptively fold its leaves into a protective defensive position after it has been repeatedly dropped from a height, displaying a trait known as habituation. This plant is incredibly responsive, and we can see in real time its behavioral capacities. Most plant behaviors and the mechanisms that support them, however, are hidden to our unaided eyes. This is where machine learning tools enter the fray.

In an ongoing project in Calvo’s lab, his team is using machine learning-based analysis of the electrophysiological activity of Phaseolus vulgaris when it experiences jet lag. Where animals tend to behave in rapid locomotive ways, where they might quickly move to evade a predator, plant behavior and responses tend to get expressed through growth, electrical activity, chemical signalling, and other mechanisms that act on longer timescales.

“Machine learning can certainly help detect patterns in such high-dimensional data, say, in time-lapse growth trajectories, root architecture, electrophysiological signals, volatile organic compounds, or multi-plant interactions,” he says. “In that sense, these tools can help make visible [the] dynamics that are otherwise too slow, distributed, or complex for us to perceive directly.”

Plants communicate through chemical, electrical, hydraulic, mechanical, and sometimes acoustic or vibrational channels. They release volatile compounds into the air, exude chemicals through their roots, and respond to signals from neighbours, fungi, herbivores, pollinators, pathogens, and their symbionts, Calvo explains. Could the decoding of plant behaviour through machine learning one day open up the possibility to converse with these and other living systems? Researchers at Project CETI, for instance, are using machine learning to listen to and translate communication among sperm whales. The folks at the Earth Species Project are also using AI models to process communicative data from crows, beluga whales, elephants, and other species. But perhaps cats, whales, and other familiar species are the easy case.

While we might not get to the point where we can have a riveting conversation with a plant in the same way we might come to meaningfully interact with other vocalising species, Calvo thinks the metaphor of plant “conversation” can be useful, provided we handle it carefully.

“There is no single universal ‘plant language’ in the way there is no single universal animal language,” Calvo says. “What we find instead are overlapping signalling systems, shaped by evolutionary history, ecological context, and the specific needs of the organisms involved.”

Rather than thinking about the “syntax” of plant communication, like grammar in the human linguistic sense, patterns of plant communication are structured through the timing, intensity, concentration, sequence, combination, and repetition of their various signaling media. “A calcium wave, an electrical signal, or a hormonal cascade may have different downstream effects depending on the tissue, developmental stage, prior history, and environmental conditions of the plant,” says Calvo. The “meaning” of the signal, he says, is not in the molecule itself, but resides in the whole organism-environment system.

Mapping the underground

Another scientist at the forefront of this movement is Toby Kiers, Professor of Evolutionary Biology at Vrije Universiteit Amsterdam, and co-founder of the Society for the Protection of Underground Networks (SPUN). Kiers, who was a recent recipient of the MacArthur ‘Genius’ grant and the Tyler Prize, is a global expert on mycorrhizal fungi—vast underground fungal networks that form mutualistic relationships with more than 80% of all plant species on Earth. These fungi trade vital nutrients, such as phosphorus and nitrogen, with plants in exchange for plant carbon.

“We are developing new technologies in machine learning and robotic imaging that allow us to carefully track the trade systems that are built by fungal networks, and probe how they make decisions,” Kiers says. “These technologies allow us to observe fungal behavior in real-time and develop models to predict fungal decision-making.”

Understanding how a symbiotic fungus processes information across its distributed network (which can be composed of billions of “nodes,” or tips) requires integrating a vast amount of data that, until relatively recently, has not been possible. As Kiers explains, a fungal network can constantly reshape its body to accommodate new nutrient trade routes with other plant species.

“At the very basic level, plants and mycorrhizal fungi use a complex chemical language to communicate,” Kiers says. “This shared language is essential because the fungus needs to be given ‘permission’ to enter into the plant cell and form a structure where nutrient trade can take place.”

Kiers and her collaborators built a custom-designed imaging robot that could track more than 500,000 growing fungal nodes across space and time in living mycorrhizal networks attached to roots. They then measured ~100,000 cytoplasmic flow trajectories along hyphae within the network, toward and away from roots.

“We used machine learning to process these terabytes of data and developed models to test how fungi optimized carbon movement and supply-chain design challenges,” Kiers says.

What these approaches reveal, Calvo reflects, is not that these systems are “like us” in some parochial sense, but rather that life has developed an astonishing array of solutions to deal with the fundamental problems faced by all living systems.

We understand these functions through the lenses of everyday decision-making, remembering our past, and communicating with our peers. However, plants, fungi, and other diverse intelligences on the tree of life have developed their own, distributed, embodied, environmentally embedded processes to build resilience and facilitate adaptation. We have shared a planet with these intelligences for billions of years, and with machine learning tools, we are starting to get to grips with the complexity of how they survive and thrive.

“That is where the deepest continuity lies: not in plants having little brains, but in living systems having to make their way in uncertain worlds,” says Calvo.

---

Mini Philosophy | Starts With A Bang | Big Think Books | Big Think Business