Could AI Make Science Move Faster?

Introduction

AlphaFold is the clearest early sign that AI can speed up science: it turned protein structure prediction from a slow, specialist bottleneck into a resource that millions of researchers can query. But the bigger AI bloom question is what happens next. The strongest optimistic case is not merely that AI gives scientists clever tools; it is that it could become a stack of discovery accelerators across biology, materials, climate, energy and medicine, compressing years of search into weeks or days while widening who can participate.

That case is now plausible enough to take seriously, but not strong enough to treat as guaranteed. AI has already produced striking scientific outputs: more than 200 million predicted protein structures in the AlphaFold Database, millions of proposed crystals from Google DeepMind’s GNoME, faster weather models such as GraphCast and GenCast, AI-guided fusion plasma control, and new systems that generate hypotheses or help interpret genetic variation. Yet the hard part remains moving from prediction to proof: making materials, testing medicines, validating mechanisms, scaling lab automation, and ensuring the gains serve broad human flourishing rather than becoming concentrated behind private platforms. Nature[AlphaFold]WikipediaAlpha FoldAlpha Fold[Google DeepMind]deepmind.googlemillions of new materials discovered with deep learningGoogle DeepMindMillions of new materials discovered with deep learning29 Nov 2023 — AI tool GNoME finds 2.2 million new crystals, includi…

AlphaFold showed what acceleration looks like

Before AlphaFold, knowing a protein’s three-dimensional shape often meant expensive and difficult experimental work using methods such as X-ray crystallography or cryo-electron microscopy. Those methods remain vital, but they do not scale easily to the full variety of proteins found across life. AlphaFold changed the practical tempo of this field by predicting protein structures from amino-acid sequences with accuracy that, in many cases, is competitive with experimental structures. The original Nature paper described a system that directly predicts atomic coordinates from sequence information and homologous sequence alignments, addressing a problem that had shaped biology for decades.[Nature]nature.comOpen source on nature.com.

The most important feature of AlphaFold was not only that a model worked. It was that the result became infrastructure. The AlphaFold Protein Structure Database, developed by Google DeepMind and EMBL-EBI, now gives open access to more than 200 million predicted protein structures. A 2024 database paper described AlphaFold DB as a large digital library with more than 214 million entries, which means a researcher can often begin a project with a plausible structural model rather than spending months trying to obtain one.[AlphaFold]WikipediaAlpha FoldAlpha Fold

That is why AlphaFold matters for the AI bloom thesis. It offers a concrete example of intelligence becoming more abundant inside science. A bottleneck that used to depend on scarce expert labour, specialised equipment and uncertain experimental luck became partly searchable, shareable and reusable. The gain is not that all of structural biology is solved. It is that many more researchers can ask structural questions earlier, cheaper and at larger scale.

AlphaFold 3 widened the signal by moving beyond single protein structures towards biomolecular interactions involving proteins, nucleic acids, small molecules, ions and modified residues. This matters because biology is not just a collection of isolated shapes: drugs bind to proteins, proteins interact with DNA and RNA, and cellular systems depend on complexes. But this also shows the pattern of the field: each AI step opens new scientific opportunities while revealing new limits. AlphaFold 3 still needs experimental validation, and researchers have noted concerns about access, transparency and limits in particular classes of molecular interaction. Nature[EMBL-EBI]ebi.ac.ukSource details in endnotes.

The real prize is a discovery stack, not a single model

AlphaFold is sometimes treated as if it were the whole story of AI in science. A better way to see it is as one layer in a larger discovery stack. Scientific acceleration happens when several kinds of AI tool reinforce one another: models that read literature, models that propose hypotheses, models that simulate systems, models that design molecules or materials, robotics that run experiments, and analysis systems that decide what to test next.

This is the shift from “AI gives an answer” to “AI changes the cycle time of research”. In a traditional research loop, scientists survey prior work, form hypotheses, design experiments, collect data, analyse results, revise the question and repeat. AI can potentially speed up several of those steps at once. Google’s AI co-scientist, for example, is described as a multi-agent system built on Gemini 2.0 to help scientists generate hypotheses and research proposals. Reuters reported early testing with institutions including Stanford University and Imperial College London, while the research description frames the system as a virtual collaborator rather than a replacement for human scientists.[Google Research]research.googleSource details in endnotes.

This kind of system is more speculative than AlphaFold because it operates closer to scientific reasoning itself. A protein structure prediction can be compared against experimental structures. A research hypothesis is harder to score: it may be novel but wrong, plausible but unimportant, or useful only after years of refinement. Still, the direction is important. If AI systems can reliably read vast literatures, connect distant findings, suggest experiments and prioritise what to test, the bottleneck shifts from having ideas to proving which ideas survive contact with the world.

That is where laboratory automation becomes central. An AI that proposes a thousand experiments is not enough if only a handful can be run. Autonomous and semi-autonomous labs aim to close the loop by letting software plan experiments, robots execute them, instruments collect data, and models update the next round. A 2025 perspective on AI and robotics in natural science laboratories highlighted the promise of accelerated discovery in life sciences and materials, but also stressed challenges around robust autonomy, reproducibility, standardisation, human roles and ethics.[arXiv]arxiv.orgSource details in endnotes.

Materials discovery is the clearest “beyond biology” test

Materials research is one of the strongest examples of AI moving beyond AlphaFold-like biology. Modern abundance depends on materials: better batteries, catalysts, solar cells, carbon-capture systems, semiconductors, superconductors, fertiliser processes and construction materials. A society with cheap clean energy and resilient infrastructure would not be built from software alone; it would require new physical capabilities.

Google DeepMind’s GNoME is a striking signal here. In 2023, DeepMind reported that GNoME had identified 2.2 million new crystal structures, including about 380,000 stable materials judged promising enough for further study. The Nature paper on scaling deep learning for materials discovery described this as enabling previously impossible modelling capabilities, while also noting unresolved problems around phase transitions, dynamic stability and synthesizability.[Google DeepMind]deepmind.googlemillions of new materials discovered with deep learningGoogle DeepMindMillions of new materials discovered with deep learning29 Nov 2023 — AI tool GNoME finds 2.2 million new crystals, includi…

The key word is “synthesizability”. A predicted material does not automatically become a usable battery or catalyst. It may be stable in a calculation but hard to make, unstable under real operating conditions, too expensive, toxic, brittle, rare, or simply not useful. This is the same realism test that runs through AI science: prediction is valuable, but the world gets the final vote.

The A-Lab at Lawrence Berkeley National Laboratory showed both promise and caution in one case. The Nature paper introduced an autonomous laboratory for solid-state synthesis that combined computation, literature-derived knowledge, machine learning, active learning and robotics. Over 17 days, it reported realising 41 novel compounds from 58 targets. Berkeley Lab’s own coverage emphasised the combination of AI-generated targets, automated synthesis and the Materials Project database. Nature[PubMed]pubmed.ncbi.nlm.nih.govSource details in endnotes.

Yet the A-Lab case also attracted scrutiny. Chemistry World reported that a later analysis raised doubts about the autonomous lab’s claimed materials discoveries. This does not make the whole field unimportant; it makes the lesson sharper. AI-for-science claims must be judged by reproducible experimental proof, not by impressive numbers of candidates generated on a computer.[Chemistry World]chemistryworld.comSource details in endnotes.

Microsoft’s MatterGen points to another direction: not just screening vast lists of possible materials, but generating candidate inorganic materials under desired property constraints. The Nature paper says MatterGen produces stable and diverse inorganic materials and can be fine-tuned towards a range of property constraints; Microsoft describes it as a move towards generative AI-assisted materials design. This is closer to “inverse design”: start with the material properties society needs, then search for structures that might deliver them.[Microsoft]microsoft.comOpen source on microsoft.com.

Biology is becoming more searchable, but not simple

AlphaFold made protein structure more accessible, but biological acceleration now extends into genetics, variant interpretation, drug discovery and synthetic biology. The common pattern is that AI turns enormous biological search spaces into ranked maps. That can save time, reveal candidates humans would miss, and help researchers prioritise scarce experimental work.

AlphaMissense is a good example. Missense variants are single-letter genetic changes that alter the amino acid sequence of a protein. Many are harmless, but some contribute to disease, and the number of possible variants is too large for manual interpretation. Google DeepMind reported that AlphaMissense predicted the pathogenicity of all 71 million possible human missense variants, classifying 89% of them as likely benign or likely pathogenic. The underlying research presented AlphaMissense as an adaptation of AlphaFold fine-tuned on human and primate variant frequency data, combining structural context and evolutionary conservation.[Google DeepMind]deepmind.googlemillions of new materials discovered with deep learningGoogle DeepMindMillions of new materials discovered with deep learning29 Nov 2023 — AI tool GNoME finds 2.2 million new crystals, includi…

This is scientifically important, but it is not a clinical magic wand. Variant-effect predictions can help researchers and clinicians prioritise which mutations to study, but they do not replace patient evidence, family histories, laboratory assays or clinical judgement. The acceleration comes from narrowing the haystack, not abolishing the need for verification.

A newer and broader signal comes from genomic foundation models such as Evo 2. The Nature paper describes Evo 2 as a biological foundation model trained on around 9 trillion DNA base pairs across all domains of life, with a long context window and single-nucleotide resolution. It reports that the model can predict functional impacts of genetic variation, including non-coding pathogenic mutations and clinically significant BRCA1 variants, without task-specific fine-tuning. The Arc Institute frames Evo 2 as part of an open and collaborative approach to biological modelling.[Nature]nature.comOpen source on nature.com.

For the AI bloom thesis, this matters because biology is one of the deepest constraints on human flourishing: disease, ageing, food production, environmental resilience and biotechnology all depend on understanding living systems. But it also raises safety and governance questions. More powerful biological design tools could help medicine and climate adaptation, while also increasing the importance of biosecurity, access control, privacy and responsible publication norms. Recent work on DNA foundation model embeddings, for example, warns that genomic representations can leak sequence information under reconstruction attacks, showing that even seemingly technical design choices can have privacy consequences.[arXiv]arxiv.orgSource details in endnotes.

Climate, weather and energy show the broader pattern

Scientific acceleration beyond AlphaFold is not confined to lab biology. Climate and energy systems involve complex dynamics, sparse measurements, physical constraints and high-stakes decisions. AI is beginning to help here too, although again the strongest evidence is for specific tasks rather than sweeping replacement of existing science.

Weather forecasting is a vivid case because forecast speed and accuracy have immediate social value. Google DeepMind’s GraphCast predicts global weather up to 10 days ahead and was reported to outperform the European Centre for Medium-Range Weather Forecasts’ high-resolution deterministic system on 90% of 1,380 verification targets, while producing forecasts much faster than traditional supercomputer-based models.[Google DeepMind]deepmind.googlemillions of new materials discovered with deep learningGoogle DeepMindMillions of new materials discovered with deep learning29 Nov 2023 — AI tool GNoME finds 2.2 million new crystals, includi…

GenCast extends the picture from a single forecast to probabilistic forecasting, where the model generates a range of possible weather futures. The Nature paper describes GenCast as a machine-learning weather model with greater skill and speed than ECMWF’s ENS ensemble forecast, generating 15-day global forecasts for more than 80 variables in about eight minutes. This matters for renewable energy planning, storm preparation and risk management because many decisions depend not only on the most likely weather, but on the range of plausible outcomes.[Nature]nature.comOpen source on nature.com.

The caveat is that weather success does not automatically solve climate projection. Weather models can be trained and tested against vast records of past atmospheric states. Climate change asks harder questions about generalisation under future conditions, rare extremes and coupled Earth systems. A 2025 Nature Communications review of AI for extreme weather and climate events stressed the need for accurate, transparent and reliable models, highlighting problems of limited data, real-time integration and trust. A 2026 Science Advances study also found that physics-based models still outperformed AI models for record-breaking heat, cold and wind in many cases.[Nature]nature.comOpen source on nature.com.

Fusion research shows a different kind of acceleration: AI as a control system for difficult physical experiments. DeepMind and the Swiss Plasma Center demonstrated reinforcement learning for magnetic control of tokamak plasmas, one of the hardest real-world control problems to which reinforcement learning had been applied. Later work used AI control to reduce the likelihood of tearing instabilities in the DIII-D tokamak, the largest magnetic fusion facility in the United States. These are not proofs of commercial fusion, but they show AI helping researchers manage complex physical systems that matter for clean energy.[Nature]nature.comOpen source on nature.com.

Why faster discovery still needs real-world proof

The central misunderstanding about AI science is to confuse a generated candidate with a discovery. A predicted protein structure, proposed molecule, candidate material, simulated experiment or ranked hypothesis can be extremely valuable. But science earns its authority through validation: measurement, replication, causal explanation and practical performance under messy conditions.

This creates several bottlenecks that AI does not automatically remove.

The experiment bottleneck. Many claims must still be tested in wet labs, materials labs, field trials or clinical studies. AI can prioritise what to test, but physical experiments take time, money, equipment, safety review and skilled labour.

The data bottleneck. AI models learn from available data. If that data is biased towards well-studied organisms, familiar materials, successful publications or easy measurements, models may reproduce those blind spots. GNoME’s materials work, for example, still faces challenges around whether predicted materials can be made and whether they will be useful in applications.[Nature]nature.comOpen source on nature.com.

The reliability bottleneck. Models may be impressive on benchmarks while failing in rare, out-of-distribution or high-stakes cases. This is especially important in medicine, climate extremes and safety-critical energy systems. The record-breaking weather findings are a reminder that strong average performance can hide weaknesses exactly where society most needs reliability.[Science]science.orgSource details in endnotes.

The interpretation bottleneck. Science is not only prediction; it is understanding. A model may predict that a material, variant or molecule is promising without giving researchers a satisfying causal account. That can still be useful, but black-box acceleration may be fragile if scientists cannot explain, debug or generalise the result.

The access bottleneck. AlphaFold’s open database became powerful partly because it was widely usable. By contrast, debates around AlphaFold 3’s access and code availability showed that AI science can become less democratic if frontier models are locked behind restricted servers or commercial terms.[AlphaFold]WikipediaAlpha FoldAlpha Fold

What would count as a genuine AI bloom signal?

The AI bloom case becomes much stronger if scientific acceleration produces durable improvements in health, energy, climate resilience and material abundance, not just impressive papers. A useful test is whether AI shortens the full path from idea to working intervention.

In biology, that might mean AI-designed candidates moving through laboratory validation, animal studies, clinical trials and affordable deployment. In materials, it means not only predicted crystals, but synthesised materials that outperform existing options in batteries, catalysts, carbon capture, desalination, semiconductors or construction. In climate and weather, it means better warnings, better adaptation decisions, improved renewable energy forecasting and models that remain trustworthy under extreme conditions. In fusion and other energy systems, it means AI helping researchers run more informative experiments, control unstable systems and reduce the cost of iteration.

The strongest near-term evidence will probably not look like one dramatic “AI discovers everything” moment. It will look like repeated compression of scientific loops: more candidate ideas per researcher, better prioritisation, faster simulations, cheaper preliminary screening, more automated experiments, cleaner data, and quicker rejection of dead ends. Over decades, that could matter enormously. A civilisation that can test more good ideas faster has a better chance of curing diseases, extending healthy life, decarbonising industry and preparing for long-term risks.

But there is a distributional question inside the optimism. Scientific acceleration helps humanity bloom only if its benefits are widely usable. If AI discovery is controlled by a few firms, hidden behind proprietary systems, or channelled mainly into luxury markets and military competition, it may deepen inequality and concentration of power. Open databases, public research infrastructure, shared benchmarks, reproducible methods, safety standards and well-funded public-interest science are not decorative additions to the bloom case; they are part of what would make the case true.

The grounded verdict

AI has already accelerated parts of science beyond what would have seemed realistic a decade ago. AlphaFold turned protein structures into a global digital resource. GNoME and MatterGen show how AI can search and generate possible materials. AlphaMissense and Evo 2 show how biological sequence data can become more interpretable and designable. GraphCast and GenCast show that learned models can compete with world-leading weather systems on many tasks. AI-guided plasma control shows that machine learning can help manage difficult physical experiments.

The leap from these examples to an AI-enabled human bloom is still an inference, not a settled conclusion. The evidence supports a serious, conditional optimism: AI can reduce important knowledge bottlenecks, but the largest gains depend on experimental validation, trustworthy deployment, open scientific infrastructure, safety-aware governance and broad access. Scientific acceleration beyond AlphaFold is therefore not a side story in the AI future. It is one of the main tests of whether advanced AI becomes merely another productivity technology, or a force that helps civilisation discover, heal and build at a far larger scale.

Amazon book picks

Further Reading

Books and field guides related to Could AI Make Science Move Faster?. Use these as the next step if you want deeper reading beyond the article.

Book

Human Compatible

By Stuart Russell

Explores how powerful AI could advance humanity if aligned with human values.

See on Amazon

Book

The Kill Chain

By Christian Brose

Directly addresses AI-enabled military intelligence and decision advantage.

See on Amazon

Book

The Coming Wave

By Mustafa Suleyman

Directly addresses whether advanced AI can deliver broad civilisational benefits while remaining governable.

See on Amazon

Book

Army of None

By Paul Scharre

Explains AI systems in military operations and analysis.

See on Amazon

Browse more on Amazon:Human CompatibleThe Kill ChainThe Coming Wave

As an Amazon Associate I earn from qualifying purchases.

eBay marketplace picks

Marketplace Samples

Example marketplace items related to this page. Use the search link to explore similar finds on eBay.

Shop location

UsingUSA USA USAUKCanadaAustraliaIreland

Example eBay listing

Vintage Science Lab Illustration Chemistry New Art Print Poster Picture Cc4630

SearcheBay.co.uk: science lab poster

Browse similar oneBay.co.uk

Example eBay listing

Science Microscope Lab Che Framed Wall Art Poster Canvas Print Picture

SearcheBay.co.uk: science lab poster

Browse similar oneBay.co.uk

Example eBay listing

'Futuristic Science Lab' Wall Art Poster Prints (PP064158)

SearcheBay.co.uk: science lab poster

Browse similar oneBay.co.uk

Example eBay listing

Colorful Science Lab Quote T-Shirt Framed Wall Art Poster Canvas Print Picture

SearcheBay.co.uk: science lab poster

Browse similar oneBay.co.uk

Browse more oneBay.co.uk

Example items shown for inspiration; availability and pricing can change. Branchoria may earn a commission if you purchase through outbound eBay links.

Endnotes

1. Source: deepmind.google
Title: millions of new materials discovered with deep learning
Link:https://deepmind.google/blog/millions-of-new-materials-discovered-with-deep-learning/

2. Source: deepmind.google
Title: graphcast ai model for faster and more accurate global weather forecasting
Link:https://deepmind.google/blog/graphcast-ai-model-for-faster-and-more-accurate-global-weather-forecasting/

3. Source: nature.com
Link:https://www.nature.com/articles/s41586-024-08252-9

4. Source: nature.com
Link:https://www.nature.com/articles/s41586-021-03819-2

5. Source: nature.com
Link:https://www.nature.com/articles/s41586-024-07487-w

6. Source: ebi.ac.uk
Link:https://www.ebi.ac.uk/training/online/courses/alphafold/alphafold-3-and-alphafold-server/introducing-alphafold-3/how-have-alphafold-3s-predictions-been-validated/

7. Source: reuters.com
Link:https://www.reuters.com/technology/artificial-intelligence/google-develops-ai-co-scientist-aid-researchers-2025-02-19/

8. Source: arxiv.org
Link:https://arxiv.org/abs/2501.06847

9. Source: nature.com
Link:https://www.nature.com/articles/s41586-023-06735-9

10. Source: nature.com
Link:https://www.nature.com/articles/s41586-023-06734-w

11. Source: microsoft.com
Link:https://www.microsoft.com/en-us/research/blog/mattergen-a-new-paradigm-of-materials-design-with-generative-ai/

12. Source: nature.com
Link:https://www.nature.com/articles/s41586-025-08628-5

13. Source: deepmind.google
Title: Google Deep Mind A catalogue of genetic mutations to help pinpoint the cause
Link:https://deepmind.google/blog/a-catalogue-of-genetic-mutations-to-help-pinpoint-the-cause-of-diseases/

14. Source: deepmind.google
Link:https://deepmind.google/research/publications/21083/

15. Source: nature.com
Link:https://www.nature.com/articles/s41586-026-10176-5

16. Source: arxiv.org
Link:https://arxiv.org/abs/2603.06950

17. Source: arxiv.org
Title: arXiv Graph Cast: Learning skillful medium-range global weather forecasting
Link:https://arxiv.org/abs/2212.12794

18. Source: arxiv.org
Title: arXiv Gen Cast: Diffusion-based ensemble forecasting for medium-range weather
Link:https://arxiv.org/abs/2312.15796

19. Source: nature.com
Link:https://www.nature.com/articles/s41467-025-56573-8

20. Source: nature.com
Link:https://www.nature.com/articles/s41586-021-04301-9

21. Source: nature.com
Link:https://www.nature.com/articles/s41586-024-07024-9

22. Source: nature.com
Link:https://www.nature.com/articles/s41598-025-32583-w

23. Source: nature.com
Link:https://www.nature.com/articles/d41586-025-04100-6

24. Source: nature.com
Link:https://www.nature.com/articles/d42473-025-00164-0

25. Source: nature.com
Link:https://www.nature.com/ncomms/

26. Source: nature.com
Link:https://www.nature.com/articles/s41586-026-10265-5

27. Source: nature.com
Link:https://www.nature.com/articles/s41586-026-10644-y

28. Source: nature.com
Link:https://www.nature.com/articles/s41746-025-01859-w

29. Source: nature.com
Link:https://www.nature.com/articles/s41592-023-02087-4

30. Source: nature.com
Link:https://www.nature.com/articles/s44387-025-00018-6

31. Source: nature.com
Link:https://www.nature.com/articles/s41392-023-01381-z

32. Source: nature.com
Link:https://www.nature.com/articles/s43247-026-03238-z

33. Source: nature.com
Link:https://www.nature.com/articles/d41586-023-03552-y

34. Source: nature.com
Link:https://www.nature.com/articles/s41586-024-07487-w_reference.pdf

35. Source: nature.com
Link:https://www.nature.com/articles/s41586-021-03828-1

36. Source: hai.stanford.edu
Title: how ai is transforming scientific discovery while keeping humans at the center
Link:https://hai.stanford.edu/news/how-ai-is-transforming-scientific-discovery-while-keeping-humans-at-the-center

37. Source: news.stanford.edu
Link:https://news.stanford.edu/stories/2025/02/generative-ai-tool-marks-a-milestone-in-biology-and-accelerates-the-future-of-life-sciences

38. Source: law.berkeley.edu
Title: Millions of new materials discovered with deep learning Google Deep Mind
Link:https://www.law.berkeley.edu/wp-content/uploads/2024/02/Millions-of-new-materials-discovered-with-deep-learning-Google-DeepMind.pdf

39. Source: law.berkeley.edu
Title: Google AI and robots join forces to build new materials
Link:https://www.law.berkeley.edu/wp-content/uploads/2024/02/Google-AI-and-robots-join-forces-to-build-new-materials.pdf

40. Source: deepmind.google
Title: co scientist a multi agent ai partner to accelerate research
Link:https://deepmind.google/blog/co-scientist-a-multi-agent-ai-partner-to-accelerate-research/

41. Source: deepmind.google
Link:https://deepmind.google/science/alphafold/

42. Source: deepmind.google
Title: alphafold five years of impact
Link:https://deepmind.google/blog/alphafold-five-years-of-impact/

43. Source: deepmind.google
Title: accelerating fusion science through learned plasma control
Link:https://deepmind.google/blog/accelerating-fusion-science-through-learned-plasma-control/

44. Source: microsoft.com
Link:https://www.microsoft.com/en-us/research/wp-content/uploads/2025/01/WEF-2025_Leave-Behind_Accelerating-Materials-Design-with-AI.pdf

45. Source: microsoft.com
Title: ai meets materials discovery
Link:https://www.microsoft.com/en-us/research/story/ai-meets-materials-discovery/

46. Source: microsoft.com
Link:https://www.microsoft.com/en-us/research/video/mattergen-a-generative-model-for-materials-design/

47. Source: microsoft.com
Link:https://www.microsoft.com/en-us/research/publication/a-generative-model-for-inorganic-materials-design/

48. Source: embl.org
Title: alphafold using open data and ai to discover the 3d protein universe
Link:https://www.embl.org/news/science/alphafold-using-open-data-and-ai-to-discover-the-3d-protein-universe/

49. Source: embl.org
Title: first complexes alphafold database
Link:https://www.embl.org/news/science-technology/first-complexes-alphafold-database/

50. Source: embl.org
Title: google deepmind partnership renewal
Link:https://www.embl.org/news/science-technology/google-deepmind-partnership-renewal/

51. Source: discovery.researcher.life
Link:https://discovery.researcher.life/article/artificial-intelligence-for-modeling-and-understanding-extreme-weather-and-climate-events/c750c9cb2aa032e48a5c8d8a2326a465

52. Source: arxiv.org
Link:https://arxiv.org/abs/2502.18864

53. Source: arxiv.org
Link:https://arxiv.org/abs/2508.03278

54. Source: medium.com
Link:https://medium.com/%40EduardoLarranaga/millions-of-new-materials-discovered-with-deep-learning-d64c56ab226f

55. Source: alphafold.ebi.ac.uk
Link:https://alphafold.ebi.ac.uk/

56. Source: pubmed.ncbi.nlm.nih.gov
Link:https://pubmed.ncbi.nlm.nih.gov/34265844/

57. Source: pubmed.ncbi.nlm.nih.gov
Link:https://pubmed.ncbi.nlm.nih.gov/37933859/

58. Source: research.google
Link:https://research.google/blog/accelerating-scientific-breakthroughs-with-an-ai-co-scientist/

59. Source: pubmed.ncbi.nlm.nih.gov
Link:https://pubmed.ncbi.nlm.nih.gov/38030721/

60. Source: newscenter.lbl.gov
Title: google deepmind new compounds materials project
Link:https://newscenter.lbl.gov/2023/11/29/google-deepmind-new-compounds-materials-project/

61. Source: chemistryworld.com
Link:https://www.chemistryworld.com/news/new-analysis-raises-doubts-over-autonomous-labs-materials-discoveries/4018791.article

62. Source: science.org
Link:https://www.science.org/doi/10.1126/sciadv.aec1433

63. Source: wired.com
Link:https://www.wired.com/story/an-ai-dreamed-up-380000-new-materials-the-next-challenge-is-making-them

64. Source: pubmed.ncbi.nlm.nih.gov
Link:https://pubmed.ncbi.nlm.nih.gov/41273079/

65. Source: pubmed.ncbi.nlm.nih.gov
Link:https://pubmed.ncbi.nlm.nih.gov/40678974/

66. Source: pubmed.ncbi.nlm.nih.gov
Link:https://pubmed.ncbi.nlm.nih.gov/41781614/

67. Source: pubmed.ncbi.nlm.nih.gov
Link:https://pubmed.ncbi.nlm.nih.gov/38718835/

68. Source: alphafoldserver.com
Link:https://alphafoldserver.com/

69. Source: facebook.com
Link:https://www.facebook.com/microsoftresearch/videos/microsoft-researchers-introduce-mattergen-a-model-that-can-discover-new-material/9579456348740625/

70. Source: snu.elsevierpure.com
Title: alphafold protein structure database in 2024 providing structure
Link:https://snu.elsevierpure.com/en/publications/alphafold-protein-structure-database-in-2024-providing-structure-/

71. Source: alphafold.ebi.ac.uk
Title: ebi.ac.uk About
Link:https://alphafold.ebi.ac.uk/about

72. Source: alphafold.ebi.ac.uk
Title: ebi.ac.uk Downloads
Link:https://alphafold.ebi.ac.uk/download

73. Source: github.com
Link:https://github.com/google-deepmind/alphamissense

74. Source: datascience.uchicago.edu
Title: nature highlights dsi research on ai for extreme weather forecasting
Link:https://datascience.uchicago.edu/insights/nature-highlights-dsi-research-on-ai-for-extreme-weather-forecasting/

75. Source: darkdaily.com
Link:https://www.darkdaily.com/2023/11/15/google-deepmind-says-its-new-artificial-intelligence-tool-can-predict-which-genetic-variants-are-likely-to-cause-disease/

76. Source: scienceopen.com
Link:https://www.scienceopen.com/document?vid=b1f93136-f0f0-4c78-8e45-27bd037e9bb9

77. Source: venturebeat.com
Title: google deepminds materials ai has already discovered 2 2 million new crystals
Link:https://venturebeat.com/ai/google-deepminds-materials-ai-has-already-discovered-2-2-million-new-crystals

78. Source: youtube.com
Title: [Alpha Fold]({{ ‘alpha-fold/’ | relative_url }})
Link:https://www.youtube.com/watch?v=P_fHJIYENdI

79. Source: youtube.com
Link:https://www.youtube.com/watch?v=gwbWHJnHDZU

80. Source: sciencedirect.com
Link:https://www.sciencedirect.com/science/article/pii/S2352940725003981

81. Source: sciencedirect.com
Link:https://www.sciencedirect.com/science/article/pii/S2666389926000061

82. Source: sciencedirect.com
Link:https://www.sciencedirect.com/science/article/pii/S0920379624000140

83. Source: isomorphiclabs.com
Title: alphafold 3 predicts the structure and interactions of all of lifes molecules
Link:https://www.isomorphiclabs.com/articles/alphafold-3-predicts-the-structure-and-interactions-of-all-of-lifes-molecules

84. Source: Wikipedia
Title: Alpha Fold
Link:https://en.wikipedia.org/wiki/AlphaFold

85. Source: uvio.bio
Title: Alpha Fold Supplementary Information
Link:https://www.uvio.bio/alphafold-architecture/AlphaFold-Supplementary-Information.pdf

86. Source: blog.google
Title: google research ai co scientist
Link:https://blog.google/feed/google-research-ai-co-scientist/

87. Source: blog.google
Title: Alpha Fold 3 predicts the structure and interactions of all
Link:https://blog.google/innovation-and-ai/products/google-deepmind-isomorphic-alphafold-3-ai-model/

88. Source: labcritics.com
Link:https://labcritics.com/blog/2026/05/21/google-deepminds-co-scientist-graduates-from-research-demo-to-nature-paper/

89. Source: niso.org
Title: google research offers ai co scientist
Link:https://www.niso.org/niso-io/2025/02/google-research-offers-ai-co-scientist

90. Source: genengnews.com
Title: google deepmind and edison are building the ai scientist
Link:https://www.genengnews.com/topics/artificial-intelligence/google-deepmind-and-edison-are-building-the-ai-scientist/

Additional References

91. Source: youtube.com
Title: Hassabis: AI Brings Breakthrough Science — But Serious Risks Remain
Link:https://www.youtube.com/watch?v=8wivCld-tCA

92. Source: youtube.com
Title: Beyond Alpha Fold: The future of structural biology is likely “dynamic”
Link:https://www.youtube.com/watch?v=h6f0J-rVbk4

93. Source: youtube.com
Title: How AI Is Rewriting Biology & Materials
Link:https://www.youtube.com/watch?v=4sWayGceA7E

94. Source: researchgate.net
Link:https://www.researchgate.net/publication/389277932_Artificial_intelligence_for_modeling_and_understanding_extreme_weather_and_climate_events

95. Source: researchgate.net
Link:https://www.researchgate.net/publication/386190277_Addendum_Accurate_structure_prediction_of_biomolecular_interactions_with_AlphaFold_3

96. Source: linkedin.com
Link:https://www.linkedin.com/posts/jiajiezhang_the-ai-co-scientist-is-here-nature-medicine-activity-7439522413027098624-f1PZ

97. Source: scribd.com
Link:https://www.scribd.com/document/735742297/s41586-024-07024-9

98. Source: facebook.com
Link:https://www.facebook.com/groups/38680135622/posts/10174777485035623/

99. Source: x.com
Link:https://x.com/NaturePortfolio/status/1730247585523273853

100. Source: facebook.com
Link:https://www.facebook.com/NaturePortfolioJournals/posts/a-paper-in-nature-presents-an-autonomous-laboratory-the-a-lab-that-combines-comp/763531545803114/