Birdwatchers’ casual notes could tell scientists something they have struggled to measure: which animals help, avoid or compete with one another. Researchers led by the University of Michigan have used AI to extract records of species interactions from the comments people add to wildlife observations. The idea is surprisingly straightforward. People already describe what they see in the natural world, but much of that useful detail sits inside ordinary sentences rather than a neatly organised scientific database. Language models may help turn those sentences into evidence that researchers can check and study.
The work, described in a University of Michigan report and published in a peer-reviewed ecology paper in October 2026, focuses on information from participatory science platforms. Birdwatchers and other volunteers often record more than a species name or a location. They describe behaviour: a bird chasing another, an insect visiting a flower or several animals feeding together. Those additional observations can matter because ecosystems are built from relationships, not simply lists of organisms present in the same place.
Birdwatchers’ Notes Can Reveal Hidden Relationships
Imagine somebody standing near a lake who notices one species of goose feeding alongside another. Their online record might identify the bird and then include a sentence describing how the flock behaved. That comment could contain evidence of an interaction, but it would be difficult for a researcher to find and classify if it were buried among huge numbers of unrelated reports. A language model trained to interpret natural language can help sort potentially relevant passages from the rest.
The researchers were interested in interactions between species rather than simply identifying the species themselves. That distinction may sound minor, but it changes the ecological question. Knowing that two birds live in an area does not establish whether they compete, co-operate or merely happen to share the same habitat. Understanding how organisms affect one another can help scientists investigate why populations change and how relationships may shift as environments change.
The team's study used two case studies involving bird-to-bird interactions and interactions between plants and pollinators. Its approach works with written observations in different languages and extracts information such as the species involved and the apparent type of interaction. The authors report that the workflow could categorise these observations with high accuracy in their tests. That claim relates to the data and validation procedures they used, not to a promise that any model will correctly interpret every wildlife comment it encounters.
A Nature Diary Is Not the Same as a Scientific Measurement
Citizen science already provides an extraordinary resource for ecology. Volunteers contribute sightings from places and times that professional field teams could not cover continuously. Yet the most useful part of a note may be phrased casually. Someone might write that a bird chased another away from a feeder, or that an insect moved between flowers. Turning that into a structured record requires identifying the actors, the behaviour and the context without adding meaning that the observer did not supply.
This is where AI can make a difference in workload. A researcher might otherwise need to read a large collection of comments manually and label each relevant example. Automated extraction can identify promising records for review and support analysis across wider areas. But the person writing the original note may not have followed a fixed protocol. Their description could be ambiguous, mistaken or missing important context. A model can organise such information, but it cannot transform an uncertain observation into a certain one by writing it in a clean table.
The published research abstract in the US National Library of Medicine's PubMed database describes a workflow for finding interactions in multilingual citizen science comments. The authors say that appropriate validation, expert review and privacy safeguards remain necessary. This is a vital part of the story. Useful automated classification is not an invitation to remove specialists from the process. It makes previously difficult material easier to inspect, while leaving scientific interpretation and data quality as responsibilities that need care.
There is a further problem of sampling. People are more likely to report striking behaviour than ordinary inactivity, and popular locations may receive far more attention than remote ones. Some species attract keen observers while others go unnoticed. A database made from comments therefore reflects the habits of both wildlife and the people watching it. Researchers must account for that when drawing conclusions. A model could be excellent at reading the notes yet still produce a distorted picture if the underlying observations are not representative.
Why Species Interactions Matter for Conservation
An ecosystem is a network of relationships. Pollinators visit plants, predators influence prey and birds may compete over resources or share opportunities. If a species moves into a new area, it may encounter different companions and rivals. Changes in temperature, habitat or land use can alter those relationships in ways that cannot be understood from a simple map of where each organism has been seen. Interaction data can help researchers ask more detailed questions about those changes.
This is why scattered field notes become valuable. A comment describing an unusual feeding relationship can point towards something worthy of further investigation. When many independently observed interactions are assembled, scientists may be able to examine patterns across geography or time. That does not mean the AI has proved the cause of a population change. It provides a richer set of candidate observations with which researchers can test explanations or plan additional fieldwork.
The same distinction helps separate this study from LiveAIWire's coverage of BirdNET and birdsong recognition. A system that identifies a bird from its call answers 'what species might this be?' The Michigan project asks 'what did these species do together?' These are complementary forms of information, but one cannot replace the other. Recognising an animal does not automatically reveal whether it was nesting, fighting, sharing food or responding to another creature.
Researchers have likewise used computer vision to recognise individual animals from distinctive markings, as discussed in LiveAIWire's article about identifying jaguars. Such projects illustrate how automated tools can reduce the effort needed to organise observations. The Michigan study instead begins with human language. It makes use of the fact that a volunteer may already have done the interpretive work of noticing and describing a meaningful event. The model's role is to find and organise those descriptions without silently inventing the missing detail.
The People Who Write the Notes Still Matter
It would be easy to describe the system as AI discovering nature on its own. That would be misleading. The original observations come from people who spent time outdoors, recognised something interesting and chose to write it down. The researchers add scientific questions, validation methods and interpretation. The model assists with processing a collection that would otherwise be cumbersome. The project therefore illustrates how AI can increase the usefulness of human effort rather than make that effort unnecessary.
There are also reasons to protect some of that information. Precise locations for rare or vulnerable species may create problems if they become too easy to discover. People contributing observations may have expectations about how their text and location details will be reused. An extraction pipeline could concentrate information that was previously difficult to search. The researchers explicitly note the importance of privacy for users and sensitive species, which should remain part of any wider deployment.
A good conservation workflow would preserve the source context of each observation. That means recording what the original comment said, when and where it was contributed if permitted, which species were identified and how confidently the interaction was classified. Researchers need the ability to correct mistakes and revisit ambiguous examples. If a model simply exports polished claims without their evidence, the final dataset may look more trustworthy than it is. Provenance is as important in ecological AI as it is in reporting or historical research.
What Wildlife Volunteers Can Contribute
For the individual nature enthusiast, the finding makes careful notes feel worthwhile. A sighting that includes behaviour, context and a clear distinction between what was observed and what was guessed may offer more information than a species name alone. The study does not prescribe a new universal format for volunteers, and nobody should feel obliged to turn a walk into a formal survey. It does suggest that plain language can have scientific value when researchers know how to examine it responsibly.
There is a balance to strike. More detailed notes may be useful, but inaccurate certainty is not. An observer who writes 'the two birds appeared to compete for food' has preserved an interpretation as an interpretation. That can be better than asserting an unobserved motive. Research tools, whether human or automated, work best when the original record keeps that distinction. A model should not silently strengthen a cautious note into a definitive claim about animal behaviour.
Some environmental AI projects begin with imagery or recordings. LiveAIWire's report on detecting orcas with AI illustrates a different source of observations gathered in the field. The Michigan approach shows that another valuable resource is already sitting in text that volunteers have written. No new sensor is required to create those historical notes. The work lies in extracting, validating and organising their meaning. That makes the project accessible even to readers who are not interested in how the underlying language model was built.
The surprising point is that the raw material was there all along. Nature lovers were recording fragments of ecological stories, often as a personal aside to a sighting. AI may help scientists read those fragments at a scale that was previously difficult. The strength of the idea depends on respecting both halves: the richness of what humans observe and the discipline needed to check what a machine extracts. If those remain in balance, an ordinary field note could contribute to a much clearer picture of how life interacts.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.