Get Your Weekly Digest

Receive the best podcast summaries and insights delivered to your inbox every week.

We respect your privacy. Unsubscribe anytime.

TED Talks Daily

How AI is unearthing hidden scientific knowledge | Sara Beery

with Sarah Beery

Published November 25, 2025
View Show Notes

About This Episode

Ecologist and AI researcher Sarah Beery explains how vast ecological databases like iNaturalist contain far more information than simple species sightings, including individual identification, species interactions, vegetation, and food webs. She describes how her team at MIT built an AI-powered system called Inquire that lets scientists search millions of images using natural language queries to rapidly extract research-ready datasets, dramatically accelerating ecological discovery. The talk closes with a call for widespread citizen participation in data collection to help build a more complete, actionable picture of life on Earth and support conservation in the face of the biodiversity crisis.

Topics Covered

Artificial intelligence Tracking and wildlife Climate change and agriculture Environmental restoration Data-driven decision-making Innovation strategy Biodiversity loss Ecological data Citizen science Wildlife conservation

Disclaimer: We provide independent summaries of podcasts and are not affiliated with or endorsed in any way by any podcast or creator. All podcast names and content are the property of their respective owners. The views and opinions expressed within the podcasts belong solely to the original hosts and guests and do not reflect the views or positions of Summapod.

Quick Takeaways

  • Most of Earth's species and their ecological relationships remain unknown, leaving us effectively "seeing only a fifth" of nature while trying to protect it.
  • Existing platforms like iNaturalist contain hundreds of millions of labeled images with rich, untapped information about individuals, interactions, habitats, and food webs.
  • Traditional manual data curation and conventional supervised AI approaches are too slow because they require thousands of labeled examples for every new scientific question.
  • Beery's team created Inquire, an AI system that lets ecologists turn natural-language questions into rapid searches across hundreds of millions of images without coding or pre-labeled training sets.
  • Using Inquire, researchers replicated in about three hours a bird diet study that previously required 1,560 hours of manual effort, with nearly identical results.
  • Similar discovery systems could be extended to bioacoustic recordings, satellite data, aerial video, and animal GPS trajectories to reveal cross-dataset ecological patterns.
  • By unlocking hidden knowledge in existing data, scientists can better identify knowledge gaps and focus future data collection where it is most needed for conservation.
  • Citizen scientists play a crucial role, as every uploaded photo, sound, or observation becomes a piece of the global biodiversity puzzle.

Podcast Notes

Podcast and talk introduction

Host introduction and show framing

Elise Hugh introduces TED Talks Daily[2:13]
• She says the podcast brings new ideas to spark curiosity every day
Introduction of the talk's central question[2:29]
• Elise asks what if we could map every living species on Earth and use that knowledge to protect the planet
Brief description of the speaker and her work[1:36]
• Host describes Sarah Beery as an ecologist and AI researcher
• She explains that Beery and her team at MIT are building tools that let scientists ask questions directly to vast ecological databases
• These tools aim to unlock hidden insights from millions of images and recordings
• Elise frames Beery's idea as AI becoming a powerful ally in understanding and saving the natural world

Framing the biodiversity knowledge crisis

Medical analogy for our limited view of nature

Doctor analogy to illustrate incomplete knowledge[2:53]
• Beery asks the audience to imagine being a doctor trying to save a patient's life while only seeing a fifth of their body
• She questions how one could prescribe medicine or perform surgery under such constraints
Connecting the analogy to ecology[3:47]
• Beery states that this limited visibility is analogous to our situation with nature across the planet
• She emphasizes that we need to act now to protect ecosystems under threat
• Despite urgency, she notes there is still so much we do not know about life on Earth

Beery's background and research focus

Dual identity as AI researcher and ecologist[3:23]
• Beery identifies herself as both an AI researcher and an ecologist
Role at MIT[3:23]
• She explains that as a professor at MIT she leads a research group
• Her group develops methods to help us learn more about the natural world
Vision for AI in ecology[3:47]
• Beery says she sees a future where AI can help exponentially increase our ecological knowledge across species and ecosystems
• She argues that to reach this future, we must change how we use AI in ecology
Need for new kinds of AI methods[3:47]
• Beery calls for AI methods that are flexible and interactive
• She stresses that scientists must be able to use these methods to discover knowledge hidden in ecological data

Scale of unknown biodiversity and interconnected risks

How many species we know and don't know

Estimated total number of species[4:00]
• Scientists estimate that there are 10 million species sharing the planet with humans
Observed versus unobserved species[4:12]
• Beery states that we have only ever observed 2 million of those species
• She highlights that 8 million species, or 80% of Earth's biodiversity, remain unknown

Need for deeper species-level knowledge

Beyond existence: key ecological questions[3:39]
• Beery notes that knowing a species exists is not enough to protect it
• She lists critical questions: Where does it live? What does it eat? Does it migrate? How far?
Why deeper knowledge matters for risk assessment[4:29]
• She explains that this deeper species knowledge requires more than a single observation
• Such knowledge is necessary to understand what puts species at risk

Example of cascading ecological effects

Insect population crashes and bird diets[4:41]
• Beery poses a hypothetical: what if insect populations crash across North America?
• She notes that this scenario is not just hypothetical, saying "we know this is currently happening"
• She asks what this means for birds that eat insects, and which birds will be most at risk versus able to adapt to other food sources
Food chain implications[5:00]
• Beery extends the question to predators higher in the food chain that eat birds
• She emphasizes that everything is interconnected, and threats to one species or group can ripple outward
• These ripples can trigger the complete collapse of an ecosystem as we know it

Multiple drivers of species threat

Environmental pressures[5:11]
• Beery lists key pressures: shrinking habitats, rising temperatures, disappearing food and water sources
• She mentions natural disasters like fire causing large-scale death and displacement
Role of invasive species[5:28]
• Invasive species move in and out-compete native species for resources

Escalating extinction rates and scientific response

Magnitude of current extinctions[4:46]
• Beery says extinction rates are now 100 to 1,000 times higher than expected based on past data
Efforts to understand and respond[5:46]
• She notes that scientists, policymakers, and community members worldwide are racing to understand what is causing the loss
• They are asking which factors most contribute and what actions can stop it
Tragic timing of many species discoveries[5:56]
• Beery remarks that it can feel like we are discovering species just in time to write their obituaries

Case study: Tapanuli orangutan

Discovery and conservation status[6:02]
• Beery cites the Tapanuli orangutan, discovered in 2017
• She notes it is one of only three species of orangutan on Earth
• It was critically endangered before we even knew it existed
Limits of traditional data collection[6:12]
• Beery says traditional forms of data collection are too slow to keep up with the current crisis

Ecological data as an untapped gold mine

Existing large ecological databases

Good news: vast, underused data resources[6:21]
• Beery says we are sitting on vast databases of ecological knowledge and have barely scratched the surface
Focus on iNaturalist[6:34]
• She introduces iNaturalist as one example platform
• She explains that 300 million images have been uploaded to iNaturalist by passionate volunteers
• In every image, the community has identified a species
• This species occurrence data has already been transformative for science

Hidden information within a single image

Example of a labeled Grant's zebra image[6:57]
• Beery shows an image labeled "Grant's Zebra" in iNaturalist
• She notes it is clearly evidence that Grant's zebra were sighted in that place and time
• She argues the image shows much more than just species presence
Individual identification and movement[7:11]
• There are three Grant's zebra in the image
• Beery says they can be identified at the individual level based on their unique stripe patterns
• By identifying individuals, scientists can monitor how species move across the planet and study social networks, growth, and health
• She adds that such data can even be used to estimate full population size
Community interactions and habitat context[6:39]
• The zebra are coexisting with a herd of wildebeest in the image
• Beery points out an oxpecker, a bird that eats ticks and helps reduce the spread of disease
• She notes that the background vegetation can be identified and its type and coverage characterized
• From vegetation, scientists can estimate biomass and locally stored carbon
• She says they can look at what animals are eating in the image to build a stronger knowledge of the local food chain

Scale and diversity of ecological data sources

Multiplying insights across massive datasets[8:08]
• Beery invites the audience to take the knowledge in that one image and multiply it by 300 million images in iNaturalist
Other major ecological databases[7:55]
• She adds millions of bioacoustic recordings in Xenocanto
• She mentions tens of millions of camera trap images in Wildlife Insights
• She references thousands of hours of deep sea footage in FathomNet
Challenge: accessing the knowledge[8:13]
• Beery describes us as sitting on an "ecological gold mine"
• The main problem is efficiently accessing the knowledge within these massive datasets

Limits of human and traditional AI analysis

Human capacity constraints

Time required to manually review images[8:24]
• Beery estimates that if it takes about one second to look at every image, a person would need to work full time for 40 years to review all iNaturalist images

Current supervised AI workflow in ecology

Training task-specific models[8:40]
• Beery describes an ecologist interested in bird diets wanting to find examples of birds eating insects in the database
• They can train an AI model to help by collecting hundreds or thousands of example images to teach the model what to look for
• Once trained, the model is a powerful tool that can quickly find new examples of birds eating insects
Remaining bottleneck: data labeling for each new question[9:34]
• Beery argues that having to collect hundreds or thousands of examples every time we want to look for something new is still too slow

Reimagining scientific discovery with question-driven AI

Scientific curiosity as the starting point

Role of questions in discovery[9:19]
• Beery says scientific discovery begins with scientific curiosity and with asking questions about the world
• She gives examples of such questions: How far can a Grant's zebra migrate? What plants grow back after a forest fire? Do birds eat insects during the winter?

Desired interface: directly asking databases

Imagining question-answerable ecological databases[9:34]
• Beery asks the audience to imagine being able to directly ask questions to ecological databases and receive answers
• She presents this as an alternative to manually labeling examples for each new query

Introduction to the Inquire system

Overall goal and usability[9:49]
• Beery says her team at MIT has been working toward this vision
• They have developed a system that helps ecologists find answers in the data without collecting any examples to teach an AI model or writing any code
Technical core: linking images and language[10:01]
• Under the hood, they develop AI models that learn and understand similarities between images and scientific language
• This learned similarity is what allows users to "just ask" questions of the data

How Inquire works in practice

Experiment design through search terms

Translating questions into queries[9:19]
• Beery explains that an ecologist first designs an experiment by taking a scientific question and breaking it into search terms
• One such term might be "bird eating insect"

Massively parallel semantic search

Comparing queries to millions of images[10:30]
• Inquire takes the search term and directly compares it to all 300 million images within seconds
Efficiency versus generative AI[10:42]
• The system is engineered to operate quickly and efficiently, making it truly interactive
• Beery notes that it requires far less computational power than a generative AI approach like ChatGPT

Human verification and data export

Prioritizing likely matches[10:53]
• Once images are sorted by relevance to the query, scientists can focus on data most likely to be relevant
• They can quickly verify true matches rather than scanning every image
Creating research-ready datasets[11:04]
• After verification, scientists have human-verified examples that they can export and analyze directly

Case study: bird diet research accelerated by Inquire

Discovering diverse diet evidence

Range of food types observed[11:36]
• Beery describes a collaborator who used Inquire and found thousands of examples of birds eating insects
• They also found examples of birds eating seeds, fruit, nuts, carrion, nectar, and plants

Seasonal diet comparisons

Differences between summer and winter diets[11:18]
• Researchers analyzed differences in species diets between summer and winter using the discovered data
• They found that some birds, like American robins, do eat insects in winter but far less than in summer
• They also found that species like American tree sparrow, which are highly dependent on insects in summer, do not eat them at all in winter

Dramatic time savings versus manual work

Comparison of effort and results[11:56]
• The entire process, from question to answer, took about three hours using Inquire
• Beery notes that another team spent 1,560 hours manually curating data to perform a similar study
• When comparing Inquire's results to the manual study, there was an almost perfect match
• Beery says this shows we can quickly access hidden knowledge in these datasets

Broader applications and creative uses of Inquire

User creativity and open-ended exploration

Inspiration from collaborators[12:09]
• Beery says she has been inspired by the creativity of scientists using the system
• She emphasizes the flexible ways people have explored many different questions

Example research questions enabled

Post-fire forest regeneration[12:22]
• Scientists have used the system to look at how forests regenerate after fire
Urban versus rural mortality[12:29]
• Others have investigated differences in species mortality between urban and rural areas
Phenology and climate change[12:35]
• Researchers have also examined how flowering events are changing in relation to a changing climate
Open-ended possibilities[12:41]
• Beery states that the possibilities are truly endless
• Because the system is open-ended, any scientist can ask the questions they are interested in

Extending the paradigm to other data types and integrated systems

Applying discovery-driven AI beyond images

Other ecological data modalities[13:05]
• Beery notes they have shown this approach works for images but could imagine similar systems for other data types
• She lists bioacoustic recordings, aerial video, satellite data, and GPS trajectories from animal collars as examples
• She says this could be applied to any ecological data type one can think of

Interrelation of diverse data sources

Shared focus on life on Earth[13:17]
• Beery points out that all these types of data are innately interrelated because they look at the same thing: life on Earth
• They provide complementary but distinct perspectives
Future integrated discovery systems[13:21]
• She imagines future systems that help scientists quickly discover hidden connections between all these data sources

Impact on conservation strategy and resource allocation

Role of AI tools in addressing, not solving, the crisis

Clarifying AI's contribution[13:34]
• Beery acknowledges that AI systems like Inquire alone cannot solve the global nature crisis
• However, they can help maximize the value of data already collected

Identifying knowledge gaps and targeting data collection

Strategic use of limited resources[13:46]
• By fully leveraging existing data, scientists can better understand what knowledge gaps remain
• They can then strategically use resources to collect new data to fill those gaps

Reducing time and cost from data to action

Faster information for conservation decisions[13:52]
• Overall, Beery says this approach reduces the time and cost of deriving information that supports conservation actions
• She gives examples like understanding how to ensure that food and habitat resources are available when species need them most
• Key periods include migration through an area, breeding or rearing young, and recovering from natural disasters like fire

Call to action and shared responsibility

A unique moment of risk and capability

Dual realities of crisis and tools[14:16]
• Beery says we stand at a unique point in history
• On one hand, there is an unprecedented biodiversity crisis
• On the other, we have unprecedented tools to address it

People power and AI tools

Human willingness to contribute[14:22]
• Beery notes that millions of people around the world are eager to contribute to nature conservation and scientific discovery
AI's scaling role[14:35]
• She says AI tools enable scientists to find patterns in all that data at scales impossible for humans alone

Conservation's future in data and participation

Where the future of conservation resides[14:46]
• Beery argues that the future of conservation is not only in remote rainforests or deep ocean trenches
• She says the future of conservation is hiding in our ecological databases, both existing and those yet to be collected

Citizen science and individual action

Everyone can contribute data[14:57]
• Beery directly addresses the audience, saying everyone can contribute
• She encourages people to collect data and upload it to platforms like iNaturalist
Each observation as part of a larger puzzle[15:09]
• She states that every photo uploaded, every sound recorded, and every observation shared is a piece of the puzzle

Closing message of urgency and collaboration

Acting now with AI as a tool[15:15]
• Beery reiterates that we know we need to act now to save nature under threat
• She says that together, with scientific AI tools in our toolbox, we can help by building the complete picture of life on Earth
• She ends her talk with a thank you

Post-talk credits and context

Event and partnership details

TED Countdown and Bezos Earth Fund[15:36]
• Elise Hugh notes that this was Sarah Beery at a TED Countdown event in New York City
• She says the event was in partnership with the Bezos Earth Fund in 2025

TED curation and production credits

Information on TED curation[15:37]
• Listeners curious about TED's curation are directed to TED.com slash curation guidelines
Production team acknowledgments[15:49]
• Elise states that TED Talks Daily is part of the TED Audio Collective
• She notes that the talk was fact-checked by the TED Research Team
• She credits producers and editors Martha Estefanos, Oliver Friedman, Brian Green, Lucy Little, and Tansika Sangmarnivong
• She mentions the episode was mixed by Christopher Fasey-Bogan
• Additional support came from Emma Taubner and Daniela Balarezo
• Elise signs off by saying she will be back tomorrow with a fresh idea and thanks listeners for listening

Lessons Learned

Actionable insights and wisdom you can apply to your business, career, and personal life.

1

Before collecting more data, fully leverage the information already available by using tools and methods that can extract deeper, hidden insights from existing datasets.

Reflection Questions:

  • • What datasets, reports, or logs do I already have access to that I rarely analyze in depth?
  • • How could I use better search, visualization, or AI tools to uncover patterns in information I already possess?
  • • What is one existing data source I will commit to mining more thoroughly over the next month before gathering new data?
2

Design tools that let domain experts work in their own language-turning their natural questions directly into queries-so they can explore and iterate without relying on technical intermediaries.

Reflection Questions:

  • • Where in my work do I depend on others to translate my questions into technical tasks or analyses?
  • • How might enabling non-technical experts to ask questions directly change the speed and quality of decisions in my organization?
  • • What is one process or tool I could redesign so that subject-matter experts can interact with it more intuitively and independently?
3

Use automation to eliminate low-value, repetitive work (like manual data curation) so that human effort can focus on verification, interpretation, and higher-level reasoning.

Reflection Questions:

  • • Which repetitive tasks in my daily work could be handled by software or simple automation instead of by me?
  • • How would reallocating the hours I spend on rote tasks to analysis and strategy change my results over the next quarter?
  • • What is one specific repetitive workflow I will map out this week and explore automating or streamlining?
4

Think in interconnected systems: changes to one component (like insect populations) can cascade through entire networks, so decisions should account for upstream and downstream effects.

Reflection Questions:

  • • In a current project or problem, what are the upstream inputs and downstream consequences I might be overlooking?
  • • How could mapping the system around my work-suppliers, customers, teams, dependencies-improve the quality of my decisions?
  • • What is one important decision I'm making now where I should pause and explicitly consider second- and third-order effects before acting?
5

Broad participation, even via small individual contributions, can create powerful collective datasets that unlock insights and solutions no single expert or institution could generate alone.

Reflection Questions:

  • • Where could I contribute small but consistent inputs (data, feedback, observations, ideas) that would meaningfully support a larger effort or community?
  • • How might I better tap into crowdsourced knowledge or contributions to advance a challenge I'm facing?
  • • What is one community, platform, or project I can start contributing to this month, knowing my small inputs compound with others over time?

Episode Summary - Notes by Tatum

How AI is unearthing hidden scientific knowledge | Sara Beery
0:00 0:00