# AI scientific discoveries: the landmark results and what held up

> Where AI changed science, from AlphaFold's Nobel Prize to AI-found antibiotics, crystals, weather and maths, and what labs and proofs actually confirmed.

- URL: https://computese.com/ai-enabled-scientific-discoveries/
- Author: Duong Quan Nguyen, CEO, Computese
- Published: 2024-08-17
- Updated: 2026-09-25
- Topics: AI & automation

## In short
- AI's clearest scientific result is protein structure prediction: AlphaFold 2 was competitive with experimental accuracy in most CASP14 targets in 2020, and its creators shared the 2024 Nobel Prize in Chemistry.
- In drug and materials research, AI mostly shrinks the search: millions of candidates scored by a model, a few dozen made and tested. Halicin, abaucin and GNoME's crystals all followed that pattern.
- What held up is what a blind test, a lab or a proof confirmed. A-Lab's 2026 correction, the critiques of GNoME and data leakage found in 294 papers show why predictions need independent checks.
- AI helps science where there is a large labelled dataset, a cheap simulator or checker, and a search space too big for people. Organizations with their own data can use the same pattern, with the same validation.

AI-enabled scientific discovery is research in which a machine learning model predicts, searches or proposes what experiments alone would take years to find. The landmark results are protein structure prediction (AlphaFold, recognized by the 2024 Nobel Prize in Chemistry), antibiotic candidates such as halicin and abaucin, predicted crystals, faster weather forecasts and new mathematical constructions.

This guide takes each example from its primary paper or lab announcement, with its date and what was actually validated: by a blind test, a lab, an animal model or a proof. It then covers why AI works in these fields, where it falls short, and what the pattern means for an organization sitting on years of its own data. Related posts are under [AI](https://computese.com/category/ai/).

## What counts as an AI-enabled scientific discovery

Whether it is called AI-enabled discovery, AI for science or AI for scientific discovery, AI plays one of three roles in the results below. It **predicts** a property that would otherwise need an experiment or an expensive simulation: a protein's shape, a crystal's stability, next week's weather. It **searches** a space far too large to test, ranking millions of molecules or programs so that people test a handful. Or it **proposes** something new: a molecule, a proof, a hypothesis. In every case the model is a tool that scientists built, trained and checked.

Two of the 2024 Nobel Prizes involved AI, and both went to people. The [Nobel Prize in Physics](https://www.nobelprize.org/prizes/physics/2024/press-release/) went to John Hopfield and Geoffrey Hinton for foundational work that enables machine learning with artificial neural networks. The [Nobel Prize in Chemistry](https://www.nobelprize.org/prizes/chemistry/2024/press-release/), announced on 9 October 2024, went half to David Baker "for computational protein design" and half jointly to Demis Hassabis and John Jumper of Google DeepMind "for protein structure prediction".

| Field             | AI system         | Published                         | How the result was checked                           |
| ----------------- | ----------------- | --------------------------------- | ---------------------------------------------------- |
| Protein structure | AlphaFold 2       | Nature, July 2021                 | Blind test (CASP14) on structures not yet public     |
| Antibiotics       | Halicin model     | Cell, February 2020               | Lab assays and mouse infection models                |
| Antibiotics       | Abaucin model     | Nature Chemical Biology, May 2023 | Lab assays and a mouse wound model                   |
| Materials         | GNoME             | Nature, November 2023             | DFT calculations; 736 structures made by other labs  |
| Weather           | GraphCast         | Science, November 2023            | Scored against ECMWF's HRES on 1,380 targets         |
| Weather           | ECMWF AIFS        | Operational, 25 February 2025     | Runs beside ECMWF's physics-based model              |
| Mathematics       | FunSearch         | Nature, December 2023             | Every construction checked by a program              |
| Mathematics       | Gemini Deep Think | IMO, July 2025                    | Proofs graded and certified by olympiad coordinators |

## Protein structure: AlphaFold and the 2024 Nobel Prize

A protein is a chain of amino acids that folds into a three-dimensional shape, and the shape decides what it does. Predicting that shape from the sequence alone had been an open problem for more than 50 years. Experiments such as X-ray crystallography, NMR and cryo-electron microscopy can determine it, but slowly: when the [AlphaFold 2 paper](https://www.nature.com/articles/s41586-021-03819-2) appeared in Nature on 15 July 2021, experiments had determined the structures of around 100,000 unique proteins, a small fraction of the billions of known sequences, and one structure could take months to years of work.

AlphaFold 2 was tested blind. At CASP14 (May to July 2020), the assessment that scores predictions against recently solved structures not yet made public, its median backbone error was 0.96 Å, against 2.8 Å for the next best method; a carbon atom is about 1.4 Å wide. The network learned from structures in the Protein Data Bank (PDB) released up to 30 April 2018, and its accuracy was confirmed again on PDB structures deposited after that date. Every prediction also carries a per-residue confidence score, pLDDT, which tells a researcher which parts to trust.

![A straight chain of grey beads passes through a small neural network box and comes out folded into a compact orange ribbon, while a shelf of already-folded grey ribbons feeds the network from behind.](https://computese.com/images/blog/ai-enabled-scientific-discoveries/folding.af920bd910-1536.webp)

*AlphaFold learned the fold from structures that experiments had already solved; its predictions extend that library rather than replace it.*

The scale followed. According to the Nobel committee, AlphaFold 2 has been used to predict the structure of virtually all of the 200 million proteins researchers have identified, by more than two million people in 190 countries. A prediction takes GPU minutes to GPU hours, depending on the protein's length. It works alongside experiments rather than instead of them: the paper notes that its predictions were already helping experimentalists interpret cryo-electron microscopy maps and solve crystal structures by molecular replacement.

[AlphaFold 3](https://www.nature.com/articles/s41586-024-07487-w), published in Nature on 8 May 2024, extends prediction from single proteins to complexes of proteins, DNA and RNA, small molecules, ions and modified residues, using a diffusion-based architecture. Its authors state its limits plainly. It predicts static structures like those in the PDB, not how molecules move in solution. Like other generative models, it can hallucinate plausible-looking structure in disordered regions. On one benchmark, 4.4% of its outputs still violated chirality. The paper was published without code, only pseudocode and a non-commercial web server; Google DeepMind has since released the [inference code on GitHub](https://github.com/google-deepmind/alphafold3), with the model parameters available only from Google under separate terms of use.

## Antibiotics: halicin, abaucin and generative design

Antibiotic discovery is a search problem. Chemical libraries hold millions of molecules, few of them kill bacteria, and resistance keeps eroding the drugs we have. Virtual screening with a trained model is the AI answer: score everything, test the top few.

In February 2020, the Collins Lab at MIT published [a deep learning approach to antibiotic discovery in Cell](https://pmc.ncbi.nlm.nih.gov/articles/PMC8349178/). The team trained a neural network on 2,335 molecules measured for whether they stop _E. coli_ from growing, then ran it over the Broad Institute's Drug Repurposing Hub. Its pick, a compound previously investigated as a diabetes treatment, was renamed halicin. It killed a wide range of pathogens in the lab, including _Mycobacterium tuberculosis_, and treated _C. difficile_ and pan-resistant _Acinetobacter baumannii_ infections in mice. The model then scored more than 107 million molecules from the ZINC15 database; of 23 top predictions tested, eight were antibacterial and structurally distant from known antibiotics. According to [MIT's announcement](https://news.mit.edu/2020/artificial-intelligence-identifies-new-antibiotic-0220), that screen took three days.

![A huge grid of small molecule symbols pours into a funnel fitted with a network symbol. Only a few vials come out, and one orange vial clears a ring of bacteria in a petri dish.](https://computese.com/images/blog/ai-enabled-scientific-discoveries/screening.9af2be09a5-1536.webp)

*The model's job is to shrink a library of millions to a list short enough to test; the petri dish and the mouse model still decide.*

Three later studies refined the method:

- **Abaucin** ([Nature Chemical Biology, 25 May 2023](https://www.nature.com/articles/s41589-023-01349-8)), with the halicin paper's first author, Jonathan Stokes, as senior author. The team measured about 7,500 molecules against _A. baumannii_, a hospital pathogen that is often multidrug-resistant, trained a network on that data and used it to find structurally new candidates. Abaucin acts on _A. baumannii_ specifically (narrow-spectrum), disrupts lipoprotein trafficking through a mechanism involving the protein LolE, and controlled an infection in a mouse wound model.
- **An explainable structural class** ([Nature, 20 December 2023](https://www.nature.com/articles/s41586-023-06887-8)). The Collins Lab measured antibiotic activity and human-cell toxicity for 39,312 compounds, predicted both for 12,076,365 compounds with graph neural networks, and extracted the chemical substructures behind the predictions. Of 283 compounds tested, one structural class was selective against MRSA and vancomycin-resistant enterococci and reduced infection in mouse models.
- **Generative design** ([MIT News, 14 August 2025](https://news.mit.edu/2025/using-generative-ai-researchers-design-compounds-kill-drug-resistant-bacteria-0814), on a paper in Cell). Instead of screening existing libraries, generative models designed more than 36 million possible compounds. For gonorrhea, the screen left about 1,000; the team picked 80 and found that only two could be synthesized. One of them, NG1, was effective against _Neisseria gonorrhoeae_ in a dish and in a mouse model of drug-resistant infection. For MRSA, 22 of about 90 shortlisted designs were made, six were strongly active, and the best, DN1, cleared a MRSA skin infection in mice.

> [!NOTE]
> Every compound above was validated in lab dishes and animals, not in people. A mouse result is where drug development starts. As of the August 2025 announcement, the nonprofit Phare Bio was modifying NG1 and DN1 for further testing.

## Materials: GNoME's 2.2 million crystals, and what was made

New inorganic crystals underpin batteries, chips and solar cells, and a crystal is only useful if it is stable enough to exist. Databases such as the Materials Project compute stability with density functional theory (DFT), a quantum-mechanical approximation that is accurate but expensive. By the GNoME authors' own recalculation, those efforts had identified about 48,000 computationally stable crystals.

[GNoME](https://www.nature.com/articles/s41586-023-06735-9), published by Google DeepMind in Nature on 29 November 2023, used graph neural networks to filter candidate structures, ran DFT on the survivors, and fed every result back into training: active learning, which the paper calls a data flywheel. The result was 2.2 million structures below the previous convex hull, the boundary of lowest energies that defines which compounds are stable. Because new compounds compete with each other, 381,000 new entries remain on the updated hull, for 421,000 stable crystals in total. The paper also reports that 736 of its stable structures had already been made independently by other labs.

![A network box proposes crystal lattices, a server running a physics simulation checks each one and sends the results back into the network, and one orange crystal leaves the loop for a robot arm and furnace.](https://computese.com/images/blog/ai-enabled-scientific-discoveries/loop.1d6826b1b5-1536.webp)

*Simulation checks each prediction and feeds it back as training data; only a lab that makes the crystal shows it exists.*

"Discovered" here has a precise meaning: predicted stable by DFT, not made in a lab or shown to be useful. In April 2024, chemists Anthony Cheetham and Ram Seshadri examined a small random subset of the predictions in [Chemistry of Materials](https://pmc.ncbi.nlm.nih.gov/articles/PMC11044265/) and found scant evidence of compounds that are at once novel, credible and useful, though they called the methods promising.

A companion paper, published the same day, described the [A-Lab](https://www.nature.com/articles/s41586-023-06734-w) at Lawrence Berkeley National Laboratory: a self-driving laboratory that plans solid-state syntheses with models trained on the literature, runs them with robots and reads the X-ray diffraction patterns with machine learning. [Berkeley Lab announced](https://newscenter.lbl.gov/2023/11/29/google-deepmind-new-compounds-materials-project/) 41 new compounds from 58 attempted targets in 17 days. In March 2024, chemists at Princeton and University College London argued in [PRX Energy](https://journals.aps.org/prxenergy/abstract/10.1103/PRXEnergy.3.011002) that errors in the analysis meant no new materials had been discovered in that work. In January 2026, the authors published a [correction](https://www.nature.com/articles/s41586-025-09992-y): a manual re-analysis confirmed 36 of the 40 reported successes and found four inconclusive; one compound was dropped because it had been in the training data; and "novel" was clarified to mean new to the prediction platform, not necessarily new to science. The paper now reports 36 compounds from 57 targets. The methods behind these results, from property-predicting networks to MatterGen's generative design, are covered in [AI in materials science](https://computese.com/revolutionizing-material-science-with-ai/).

## Weather forecasting: GraphCast and ECMWF's AIFS

Weather is the clearest case of AI as a surrogate model for a simulation. Traditional numerical weather prediction solves the physics of the atmosphere on supercomputers; according to [Google DeepMind](https://deepmind.google/blog/graphcast-ai-model-for-faster-and-more-accurate-global-weather-forecasting/), a 10-day forecast from ECMWF's HRES can take hours on a supercomputer with hundreds of machines. GraphCast learned from the past instead: it was trained on four decades of ECMWF's ERA5 reanalysis and steps the global weather forward six hours at a time. [Published in Science on 14 November 2023](https://europepmc.org/article/MED/37962497), it predicts hundreds of variables 10 days ahead at 0.25° resolution in under a minute, and it outperformed the most accurate operational deterministic systems on 90% of 1,380 verification targets.

The stronger signal came from a forecast agency. On 25 February 2025, [ECMWF made its own AI model, AIFS, operational](https://www.ecmwf.int/en/about/media-centre/news/2025/ecmwfs-ai-forecasts-become-operational), running side by side with its physics-based Integrated Forecasting System (IFS). ECMWF reports gains of up to 20% for tropical cyclone tracks and about 1,000 times less energy per forecast. The AI does not start from nothing: AIFS uses the same initial conditions as IFS, built from a short-range forecast and around 60 million quality-controlled observations, and its grid spacing is 28 km against 9 km for IFS. The observing system and the physics-based analysis remain the foundation every AI forecast starts from. How far AI reaches into climate projections, and where it stops, is covered in our guide to [AI for climate research](https://computese.com/ai-for-climate-research/).

## Mathematics and algorithms: results a program can check

Mathematics has an advantage over lab science: a construction or a proof can be checked exactly, so a model can propose freely and a checker can reject its mistakes.

- **FunSearch** ([Nature, 14 December 2023](https://www.nature.com/articles/s41586-023-06924-6)) pairs a large language model with an automated evaluator in an evolutionary loop: the model writes programs, the evaluator scores them, the best survive. It found new constructions of large cap sets, a problem in extremal combinatorics, including the largest improvement in 20 years to the asymptotic lower bound, and better heuristics for online bin packing.
- **AlphaGeometry** ([Nature, 17 January 2024](https://www.nature.com/articles/s41586-023-06747-5)) solved 25 of 30 recent olympiad geometry problems, against 10 for the previous best method, after training on 100 million synthetic theorems and proofs.
- **AlphaProof and AlphaGeometry 2** scored 28 of 42 points at the [2024 International Mathematical Olympiad](https://deepmind.google/blog/ai-solves-imo-problems-at-silver-medal-level/), silver-medal standard. AlphaProof works by formal verification in the language Lean, so its proofs are machine-checked, but the problems were translated into Lean by hand and some took up to three days.
- **Gemini Deep Think**: an advanced version scored 35 of 42 at the [2025 olympiad](https://deepmind.google/blog/advanced-version-of-gemini-with-deep-think-officially-achieves-gold-medal-standard-at-the-international-mathematical-olympiad/), gold-medal standard, working in natural language within the 4.5-hour limit, and its solutions were graded and certified by IMO coordinators. Our [Gemini explainer](https://computese.com/google-unveils-gemini-the-most-advanced-and-versatile-ai-model-yet/) covers the model family.
- **AlphaEvolve** ([Google DeepMind, 14 May 2025](https://deepmind.google/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/)) found a way to multiply 4×4 complex-valued matrices with 48 scalar multiplications, improving on Strassen's 1969 algorithm in that setting. Applied to more than 50 open problems, it matched the best known solutions in about 75% of cases and improved on them in about 20%.

## Other fields: fusion control and AI research assistants

AI has also reached physics experiments and hypothesis generation, again with a real-world check at the end.

In February 2022, [a Nature paper from DeepMind and EPFL's Swiss Plasma Center](https://www.nature.com/articles/s41586-021-04301-9) described a deep reinforcement learning controller for the magnetic coils of the TCV research tokamak. The controller learned by interacting with a tokamak simulator, then ran directly on the hardware, holding a range of plasma shapes, including two separate plasmas in the vessel at once.

In February 2025, Google introduced an [AI co-scientist](https://research.google/blog/accelerating-scientific-breakthroughs-with-an-ai-co-scientist/), a multi-agent system built with Gemini 2.0 that generates hypotheses and research proposals, with experts in the loop. Drug repurposing candidates it proposed for acute myeloid leukemia were then shown to reduce the viability of cancer cell lines in lab experiments. Asked why a family of bacterial genetic elements turns up across many bacterial species, a question an Imperial College London group had already answered by experiment but not yet published, it independently proposed the same mechanism.

## What makes AI useful in science

The same conditions recur in every result above. Where they are missing, AI rarely delivers.

| Ingredient                        | What it gives the model                           | Examples from this guide                                         |
| --------------------------------- | ------------------------------------------------- | ---------------------------------------------------------------- |
| A large, labelled dataset         | Examples of the answer, measured before the model | PDB structures, ERA5 reanalysis, 39,312 measured compounds       |
| A simulator or checker            | A cheap way to reject wrong predictions           | DFT for GNoME, a tokamak simulator, Lean, FunSearch's evaluator  |
| A search space too big for people | A reason to rank rather than test everything      | 107 million molecules, 12 million compounds, 36 million designs  |
| A real-world test at the end      | Proof that the prediction holds                   | Growth assays, X-ray diffraction, mouse models, olympiad graders |

**Data** comes first: AlphaFold stands on structures solved by experiment, GraphCast on four decades of reanalysis, the antibiotic models on compounds measured in the lab. A model learns only what someone recorded, consistently and with the right label.

**Simulation** makes the loop affordable: a trained model answers in minutes what took hours of supercomputing or months at the bench, so it can filter candidates in front of the expensive method, as GNoME did for DFT.

**Search** is where the payoff is: ranking millions of candidates and sending the best few to the lab turns a fixed budget of experiments into a better-aimed one.

## The limits: lab validation, reproducibility and hype

### A prediction is not a discovery until something real confirms it

Each step from model to reality loses candidates. A compound that is stable in a calculation is not necessarily one a lab can make, and one a lab can make is not necessarily useful: that is the core of the GNoME critique. Of 80 generated antibiotic designs checked with synthesis vendors, only two could be made. A-Lab's count fell from 41 compounds to 36 after chemists re-read the diffraction data. And a mouse is not a patient.

### Reproducibility needs held-out data and published code

In an August 2023 survey in [Patterns](https://pmc.ncbi.nlm.nih.gov/articles/PMC10499856/), Sayash Kapoor and Arvind Narayanan of Princeton found data leakage in 17 fields that had adopted machine learning, affecting 294 papers and in some cases producing wildly over-optimistic conclusions. In their own re-analysis of civil war prediction, complex models that had seemed far better than logistic regression were not, once the leakage was fixed. They propose "model info sheets", which researchers fill in to detect and prevent each type of leakage they catalogue, including temporal leakage, where test data comes from before the training data. The AlphaFold 2 evaluation on structures released after its training cutoff is the pattern to copy; a paper published without code, as AlphaFold 3 was, is harder for others to check.

### Hype travels faster than peer review

A lab announcement is written to be read; the paper is written to be checked. [GNoME's announcement](https://deepmind.google/blog/millions-of-new-materials-discovered-with-deep-learning/) presented 380,000 stable materials with the potential to power future technologies, while what the paper established was stability computed by DFT. Preprints have not been peer reviewed at all: in May 2025, [MIT asked arXiv to withdraw](https://economics.mit.edu/news/assuring-accurate-research-record) a November 2024 economics preprint, "Artificial Intelligence, Scientific Discovery, and Product Innovation", after an internal review left it with no confidence in the data's provenance, reliability or validity. Peer review is not the end of checking either: A-Lab passed review at Nature and was corrected more than two years later, after outside chemists examined its data. That correction is the system working.

> [!WARNING]
> A preprint has not been peer reviewed, and a press release is not the paper. Before repeating an AI discovery claim, find the paper and check what was measured, not only what was predicted.

## How to check an AI discovery claim

A short routine catches most overstatements:

1. **Find the primary source.** Note the journal or preprint server and the date, and read the abstract, not the headline.
2. **Separate predicted from measured.** Computed stable, synthesized, active in a dish, effective in mice and effective in people are five different claims.
3. **Check that the test was blind.** CASP14 used unreleased structures; AlphaFold 2 was re-tested on structures released after its training data. A random split of time-ordered data can leak.
4. **Ask "better than what?"** GraphCast was compared with HRES, AlphaGeometry with the previous best method. A result with no baseline is a demonstration, not evidence.
5. **Look for independent replication and critique.** GNoME's 736 structures made by other labs count for more than its 2.2 million predictions.
6. **Check for corrections, code and data.** The article page lists corrections; the methods section says whether code and data are available.

## What AI in science means for organizations with data

The pattern that worked in these labs transfers to any organization with years of measurements: test results, quality records, sensor logs, transactions or support tickets. The lessons are the same, at a smaller scale.

1. **Treat historical data as the asset.** AlphaFold learned from structures experiments had already solved, and abaucin's model from about 7,500 measurements its team made. If your records are inconsistent, missing labels or scattered across systems, fix that before choosing a model.
2. **Test the way the scientists did.** Hold back the most recent data for testing, as AlphaFold's evaluation did, and compare the model with the method you use today.
3. **Put a cheap check between the model and the decision.** A simulation, a rule, a small experiment or a person should confirm what the model proposes.
4. **Use AI to rank and narrow, not to decide.** The halicin screen scored more than 107 million molecules and tested 23. The value came from the 23 tests.
5. **Record what was used.** Which data, which model version and which results, so an outcome can be reproduced or corrected later.

If your data lives in spreadsheets and disconnected systems, the first step is the platform under it: our [data platform service](https://computese.com/services/data-platform/) builds ingestion and a governed warehouse or lakehouse, engineered with tests, lineage and access control. When a model step is ready for a real workflow, our [AI and automation service](https://computese.com/services/ai-automation/) runs an evaluation set from your real cases before every change, and nothing changes in your systems until a rule or a person approves it.

## Key terms
- **AlphaFold**: Google DeepMind's AI system that predicts a protein's 3D structure from its amino acid sequence. AlphaFold 2 (2020) led to the 2024 Nobel Prize in Chemistry; AlphaFold 3 (2024) also models complexes with DNA, RNA and small molecules.
- **CASP**: The Critical Assessment of protein Structure Prediction, a blind test held every two years in which methods predict structures that have been solved but not yet made public.
- **Virtual screening**: Scoring a large library of molecules with a model, so that only the most promising few are made or bought and tested in the lab.
- **Density functional theory (DFT)**: A quantum-mechanical approximation used to compute the energy, and so the stability, of a material. Accurate enough to check predictions, too slow to run on every candidate.
- **Convex hull**: In materials science, the boundary formed by the lowest-energy compounds across a chemical system. A predicted crystal on it is considered thermodynamically stable; one above it should decompose.
- **Active learning**: A loop in which a model chooses the candidates to test, the test results are added to its training data, and the next round starts from the improved model.
- **Self-driving laboratory**: A lab where software plans experiments, robots run them and models read the results, such as Berkeley Lab's A-Lab for inorganic powders.
- **Surrogate model**: A fast learned model that stands in for an expensive calculation or simulation. AI weather models play this role for numerical weather prediction.
- **Data leakage**: Information from the test data slipping into training, so a model looks more accurate than it is. A leading cause of machine learning results in science that fail to reproduce.
- **Formal verification**: Writing a proof in a language such as Lean so that software checks every step. AlphaProof's solutions at the 2024 International Mathematical Olympiad were written this way.

## Common questions

### What are some examples of AI scientific discoveries?

The best documented are AlphaFold's protein structure predictions (Nature, 2021), the antibiotic candidates halicin (2020) and abaucin (2023), GNoME's 381,000 predicted stable crystals (2023), the GraphCast and ECMWF AIFS weather models, and new constructions and olympiad-level proofs in mathematics from FunSearch, AlphaProof and Gemini Deep Think.

### Did an AI win the Nobel Prize?

No, people did, for work on AI. The 2024 Nobel Prize in Chemistry went half to Demis Hassabis and John Jumper for protein structure prediction with AlphaFold and half to David Baker for computational protein design. The 2024 Nobel Prize in Physics went to John Hopfield and Geoffrey Hinton for foundational work on artificial neural networks.

### Has AI discovered a new antibiotic?

AI has found antibiotic candidates that work in lab dishes and in mice: halicin (2020), abaucin (2023), a structural class active against MRSA (2023) and the generatively designed NG1 and DN1 (2025). The published work stops at animal models. As of MIT's August 2025 announcement, a nonprofit was modifying NG1 and DN1 for further testing.

### Can AI make scientific discoveries on its own?

Not in any of the cases here. Robotic labs and AI research assistants automate parts of the work, but A-Lab's 2026 correction came from chemists re-examining its X-ray data, and the 2025 olympiad proofs were graded by human coordinators. Today's systems predict, rank and propose; experiments, proofs and peer review decide.

### Are AI results in science reproducible?

Not automatically. A 2023 survey in Patterns found data leakage in 294 papers across 17 fields that use machine learning, sometimes with wildly over-optimistic conclusions. Look for a test set kept apart from training, a comparison with an established method, published code and independent replication.

## Sources
1. [Press release: The Nobel Prize in Chemistry 2024](https://www.nobelprize.org/prizes/chemistry/2024/press-release/), NobelPrize.org
2. [Press release: The Nobel Prize in Physics 2024](https://www.nobelprize.org/prizes/physics/2024/press-release/), NobelPrize.org
3. [Highly accurate protein structure prediction with AlphaFold](https://www.nature.com/articles/s41586-021-03819-2), Nature
4. [Accurate structure prediction of biomolecular interactions with AlphaFold 3](https://www.nature.com/articles/s41586-024-07487-w), Nature
5. [AlphaFold 3 inference code and model parameter terms](https://github.com/google-deepmind/alphafold3), Google DeepMind on GitHub
6. [A deep learning approach to antibiotic discovery (Cell, 2020)](https://pmc.ncbi.nlm.nih.gov/articles/PMC8349178/), PubMed Central
7. [Artificial intelligence yields new antibiotic](https://news.mit.edu/2020/artificial-intelligence-identifies-new-antibiotic-0220), MIT News
8. [Deep learning-guided discovery of an antibiotic targeting Acinetobacter baumannii](https://www.nature.com/articles/s41589-023-01349-8), Nature Chemical Biology
9. [Discovery of a structural class of antibiotics with explainable deep learning](https://www.nature.com/articles/s41586-023-06887-8), Nature
10. [Using generative AI, researchers design compounds that can kill drug-resistant bacteria](https://news.mit.edu/2025/using-generative-ai-researchers-design-compounds-kill-drug-resistant-bacteria-0814), MIT News
11. [Scaling deep learning for materials discovery](https://www.nature.com/articles/s41586-023-06735-9), Nature
12. [Millions of new materials discovered with deep learning](https://deepmind.google/blog/millions-of-new-materials-discovered-with-deep-learning/), Google DeepMind
13. [Artificial Intelligence Driving Materials Discovery? Perspective on the Article: Scaling Deep Learning for Materials Discovery](https://pmc.ncbi.nlm.nih.gov/articles/PMC11044265/), Chemistry of Materials (PubMed Central)
14. [An autonomous laboratory for the accelerated synthesis of inorganic materials](https://www.nature.com/articles/s41586-023-06734-w), Nature
15. [Google DeepMind Adds Nearly 400,000 New Compounds to Berkeley Lab's Materials Project](https://newscenter.lbl.gov/2023/11/29/google-deepmind-new-compounds-materials-project/), Berkeley Lab
16. [Challenges in High-Throughput Inorganic Materials Prediction and Autonomous Synthesis](https://journals.aps.org/prxenergy/abstract/10.1103/PRXEnergy.3.011002), PRX Energy
17. [Author Correction: An autonomous laboratory for the accelerated synthesis of inorganic materials](https://www.nature.com/articles/s41586-025-09992-y), Nature
18. [Learning skillful medium-range global weather forecasting (Science, 2023)](https://europepmc.org/article/MED/37962497), Europe PMC
19. [GraphCast: AI model for faster and more accurate global weather forecasting](https://deepmind.google/blog/graphcast-ai-model-for-faster-and-more-accurate-global-weather-forecasting/), Google DeepMind
20. [ECMWF's AI forecasts become operational](https://www.ecmwf.int/en/about/media-centre/news/2025/ecmwfs-ai-forecasts-become-operational), ECMWF
21. [Mathematical discoveries from program search with large language models](https://www.nature.com/articles/s41586-023-06924-6), Nature
22. [Solving olympiad geometry without human demonstrations](https://www.nature.com/articles/s41586-023-06747-5), Nature
23. [AI achieves silver-medal standard solving International Mathematical Olympiad problems](https://deepmind.google/blog/ai-solves-imo-problems-at-silver-medal-level/), Google DeepMind
24. [Advanced version of Gemini with Deep Think officially achieves gold-medal standard at the International Mathematical Olympiad](https://deepmind.google/blog/advanced-version-of-gemini-with-deep-think-officially-achieves-gold-medal-standard-at-the-international-mathematical-olympiad/), Google DeepMind
25. [AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms](https://deepmind.google/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/), Google DeepMind
26. [Magnetic control of tokamak plasmas through deep reinforcement learning](https://www.nature.com/articles/s41586-021-04301-9), Nature
27. [Accelerating scientific breakthroughs with an AI co-scientist](https://research.google/blog/accelerating-scientific-breakthroughs-with-an-ai-co-scientist/), Google Research
28. [Leakage and the reproducibility crisis in machine-learning-based science (Patterns, 2023)](https://pmc.ncbi.nlm.nih.gov/articles/PMC10499856/), PubMed Central
29. [Assuring an accurate research record](https://economics.mit.edu/news/assuring-accurate-research-record), MIT Department of Economics