Guide · AI & data
AI for Good: Where to Find Real-World AI Projects and Challenges
By the FruitPunch research deskUpdated November 15, 20269 min read
In short
AI for good is the use of machine learning on problems such as health, climate, conservation and education, and a 2020 Nature Communications study found AI could help with 134 of the 169 Sustainable Development Goal targets while hindering 59. The main open routes in 2026 are prize challenges on DrivenData (over $5.5 million in prizes offered to date) and Zindi (over $1 million awarded), Kaggle research competitions such as BirdCLEF ($50,000 in 2025), and unpaid routes through DataKind, Omdena chapters, WILDLABS and open source tools like MegaDetector and BirdNET.
This address once hosted FruitPunch AI, an Eindhoven AI for good community. FruitPunch is a new, independent publication and is not that organization. This guide was written from public sources; the history page explains the change.
AI for good means applying machine learning and data science to social and environmental problems, such as disease detection, wildlife monitoring, crop forecasting or disaster response, usually with or for the organization that owns the problem. The potential is broad but not automatic: a 2020 study in Nature Communications found that AI may act as an enabler on 134 Sustainable Development Goal targets (79%), while 59 targets (35%) may experience a negative impact. For people who want to contribute, the most accessible doors in 2026 are open prize challenges (DrivenData, Zindi, Kaggle, AIcrowd, Solafune), volunteer programs (DataKind, Omdena local chapters) and open source conservation and science tools that accept code and data.
Key facts
- DrivenData says its competitions have offered $5,509,000+ in prize money and drawn 276,000 submissions, with partners including NASA and The World Bank (DrivenData).
- Zindi, launched in 2018, connects members across more than 185 countries (Zindi) and reports +1M in prizes awarded (Zindi).
- The BirdCLEF+ 2025 bird and wildlife sound challenge ran on Kaggle with $50,000 in prize money (LifeCLEF).
- The WILDLABS conservation technology community listed 7,700 people in 120 countries in March 2024 (WILDLABS, archived).
- The UN's AI for Good platform, organized by ITU, reports over 37,000 active contributors in more than 180 countries (ITU AI for Good).
- BirdNET, hosted at Cornell, recognizes over 11,000 species by sound and is open source (BirdNET).
What does "AI for good" actually mean?
The term has no single legal definition. In practice it covers three kinds of work:
- Predictive models for mission-driven organizations. A nonprofit, agency or research group has data and a question. DrivenData describes its process as framing a predictive question "that can be solved by the data at hand and has measurable, real-world impact", running a competition, then working with the organization to put the top model to use (DrivenData).
- Tools and datasets for science and conservation. Examples are camera trap classifiers, bioacoustic models and labeled image libraries that field teams reuse.
- Policy, standards and skills. The UN's AI for Good program, organized by ITU with over 50 UN sister agencies, focuses on identifying applications, training and standards rather than hosting coding contests.
The Nature Communications review is a useful reality check. Its authors argue that the fast development of AI "needs to be supported by the necessary regulatory insight and oversight" to support sustainable development (Vinuesa et al., 2020). A good project names the user, the decision the model supports and the harm it could cause.
Which platforms run AI for good challenges today?
The table below compares the main open platforms using what each one publishes about itself. Eligibility rules are set per challenge, so read the rules page before you invest time: government funded prizes often limit who can receive money.
| Platform | Format | Who can join | Prizes | Published scale | Recent impact example |
|---|---|---|---|---|---|
| DrivenData | Competitions of 2 to 3 months, plus open practice problems | Free account; prize eligibility set per competition | Cash prizes on prize competitions; practice ones have none | $5,509,000+ offered, 276,000 submissions | GEMS geothermal fault mapping, $300,000 in prizes |
| Zindi | Challenges built with businesses, NGOs and governments | Free profile; members in 185+ countries | Monetary rewards on many challenges | +1M in prizes awarded | Crop yield, health delivery and climate challenges |
| Kaggle | Hosted competitions, including research and code competitions | Free account; rules per competition | Some have cash prizes | Hosts long running science series such as BirdCLEF | BirdCLEF+ 2025, $50,000 in prizes |
| AIcrowd | Challenges for businesses, universities, agencies and NGOs, often with code submissions | Free account | Set per challenge | "Dozens of challenges" run | Grew out of EPFL's Digital Epidemiology Lab |
| Solafune | Satellite and geospatial data competitions | Online; rules per contest | Cash prizes | Aerosol estimation contest, 2024 | $12,000 pool, $4,000 first prize |
| Omdena | Local chapter projects of 4 to 8 weeks, open source | Beginner friendly; good English required | None; portfolio and project experience | Chapters run locally initiated projects | Crop pest management in Somalia, 5 weeks |
DrivenData
DrivenData hosts competitions where "a global community of data scientists competes to come up with the best statistical model", and models are ranked on data withheld from competitors (DrivenData). Its current list shows the range: the GEMS challenge to identify hidden geologic faults for geothermal energy carries $300,000 in prizes, the NIH backed PREPARE challenge on early prediction of Alzheimer's disease offered $650,000, and the Bureau of Reclamation's Water Supply Forecast Rodeo offered $500,000 (DrivenData competitions). It also keeps practice problems open for a year or more. "Pump it Up", on predicting faulty water pumps in Tanzania, shows 20,191 participants, and "Conser-vision", a camera trap image classification exercise, shows 2,469 (DrivenData competitions). These are the easiest place to start.
Zindi
Zindi calls itself a professional network and AI challenge platform "for data scientists and AI builders in emerging markets", born in Africa and launched in 2018 (Zindi). Challenges are developed with businesses, NGOs and governments and range from crop yields to traffic and climate. Zindi also runs UmojaHack Africa, an inter-university hackathon series, and says that in a recent survey 1 in 5 users reported getting a data science or AI job because of their Zindi profile (Zindi). That figure is self-reported by the platform.
Kaggle
Kaggle is the largest general competition site, and most of its contests are commercial, so filter for research sponsors. The clearest impact series is LifeCLEF's BirdCLEF: the 2025 edition asked participants to identify species of different taxonomic groups from soundscape recordings in Colombia's Middle Magdalena Valley, was held on Kaggle as a code competition and awarded $50,000 in prize money, plus $2,500 for each of the two best working notes papers (LifeCLEF). Writing a working note is a good way to turn a leaderboard result into a citable paper.
AIcrowd
AIcrowd helps "businesses, universities, government agencies or NGOs" run challenges, and many of its challenges are simulation based, such as reinforcement learning environments (AIcrowd FAQ). It began at EPFL, where the Digital Epidemiology Lab ran machine learning challenges for academic conferences. Its challenge tags include conference tracks such as NeurIPS and CVPR, which suits people who want research credit.
Solafune
Solafune is a Tokyo company that develops satellite and geospatial data analysis for agriculture, natural resources and more (JAXA Japanese Space Industry Catalog). It runs public competitions on its platform. Its 2024 aerosol optical depth contest ran from July 8 to November 1, 2024 with a $12,000 total prize pool, $4,000 of it for first place (Solafune press release). It suits people with remote sensing skills.
WILDLABS
WILDLABS is not a challenge platform but the main forum for conservation technology. In March 2024 its homepage described a community of 7,700 people in 120 countries discussing 1,500 topics like biologging, camera traps and machine learning (WILDLABS, archived). It is the place to find field teams that need help, ask about hardware and data, and learn what conservation users actually want. See also our AI for wildlife guide.
How can you volunteer instead of competing?
Competitions reward the best score, not the most useful collaboration. Volunteer routes trade prizes for direct work with an organization.
- DataKind. Volunteers work "directly with both DataKind staff and our social impact partners" on issues such as poverty, healthcare access and climate impacts, and can be located anywhere because many roles are remote (DataKind). You apply by creating a volunteer profile and responding to calls for projects. DataKind also runs DataDive events, marathon style sessions where data scientists and social impact professionals work through partner problems.
- Omdena local chapters. Omdena describes its local chapter challenges as "not a competition or hackathon but a real-world project", beginner friendly, education focused, open source and lasting 4 to 8 weeks (Omdena). Its main site now centers on paid AI delivery for clients, so check that a chapter is active before signing up (Omdena).
- Climate Change AI. CCAI is a nonprofit "composed of volunteers from academia and industry", founded in June 2019, that runs a summer school, workshops and a community platform (Climate Change AI). Its Innovation Grants program funded research, datasets and tools, but CCAI says it has no future calls currently planned (CCAI).
Which open source projects and datasets can you use or contribute to?
Open source projects let you contribute at any time, with no deadline or leaderboard. These are well established:
- MegaDetector. A model that "identifies animals, people, and vehicles in camera trap images", trained on several million images. It finds animals but does not name species; the repository points to SpeciesNet for that step.
- Zamba. DrivenData's Python tool that "uses machine learning and computer vision to automatically detect and classify animals in camera trap images and videos", and lets users train custom models.
- BirdNET. Deep learning for bird and wildlife sound that recognizes over 11,000 species, with an analyzer, apps and lightweight models for edge devices.
- LILA BC. A repository of labeled biology and conservation datasets that hosts over ten million labeled images, free for both data creators and researchers.
- DrivenData practice problems. Open datasets on dengue spread, earthquake damage and flu vaccines with public leaderboards.
For climate and satellite work, see our AI for Earth guide. Health applications are covered in AI for health, food and farming in AI for food, and pandemic era data projects in COVID-19 and AI.
What happened to FruitPunch AI's challenges?
The old page at this address belonged to FruitPunch AI, an Eindhoven based community that ran AI for good challenges with partner organizations. In December 2022 its homepage listed "1500+ AI engineers", "21 challenges" and labs for wildlife, earth and health (FruitPunch AI, archived). This site is an independent guide that shares the domain and is not connected to that organization, and it does not run challenges. The platforms above are where similar work is open today.
How do you pick the right challenge or project?
Use four filters:
- Match your skills to the data type. Tabular data suits beginners (DrivenData practice problems). Audio and camera trap images suit computer vision and signal work (BirdCLEF, Conser-vision). Satellite imagery suits remote sensing (Solafune).
- Check who uses the result. The best briefs name a partner and say how the model will be used, as DrivenData does for each competition. If no one is named, treat it as practice.
- Read the rules before the data. Prize eligibility, team size, external data and code submission rules differ by challenge.
- Decide what you want out of it. Prize money points to DrivenData, Zindi, Kaggle and Solafune. Collaboration and mentoring point to DataKind and Omdena chapters. Lasting contribution points to open source tools such as MegaDetector, Zamba and BirdNET.
Frequently asked questions
Do I need to be an expert to join an AI for good challenge?
No. DrivenData labels some open problems as beginner practice, such as its flu vaccine prediction exercise (DrivenData), and Omdena describes its chapter challenges as beginner friendly while welcoming experts (Omdena).
Are AI for good challenges paid?
Some are. DrivenData prize competitions have offered over $5.5 million in total (DrivenData) and BirdCLEF+ 2025 offered $50,000 (LifeCLEF). Volunteer programs such as DataKind and Omdena chapters do not pay prizes.
Can people outside the US or Europe take part?
Usually yes. Zindi members come from more than 185 countries (Zindi) and DataKind volunteers may be located anywhere (DataKind). Some government funded prizes restrict who can receive prize money, so check each rules page.
What is the difference between a competition and a volunteer project?
A competition ranks individual or team submissions on held out data, as DrivenData describes (DrivenData). A volunteer project, such as a DataKind engagement, puts you on a team working with a partner organization over weeks or months (DataKind).
Is this site the FruitPunch AI community?
No. FruitPunch is an independent rankings and guides site and is not affiliated with FruitPunch AI, which ran challenges at this domain in earlier years (archived homepage).
Spotted an error in this guide? Email editorial@fruitpunch.ai