AI × DISCIPLINES · L0 → L4
How far has AI transformed each discipline?
43 disciplines in 7 groups, rated from L0 (Not yet transformed) to L4 (AI-led) on four dimensions. Data as of 2026-10-22 (Beijing time). How levels are defined ↓
Recent leaps
AI leapa leap for AI itself · Field leapa leap for the discipline
Progress matrix
Rows: groups · columns: levels · click a discipline for details
All disciplines
Life sciences (structural & molecular biology)
L3Systemic transformation↑ Rising AI ×3Field ×5L3 (close to L4): AlphaFold made structure prediction the default step and won the 2024 chemistry Nobel; RFdiffusion reported atomically accurate de novo antibodies (Nature 2025); Evo 2 (Nature, Mar 2026) models and designs genomes across life. Missing: independent adoption surveys and curriculum evidence.
Life & medical sciences · OECD 1.6Genomics & genetics
L3Systemic transformation↑ Rising AI ×1Deep-learning variant calling (DeepVariant) is in standard pipelines; AlphaMissense scores 71M missense variants (Science 2023); AlphaGenome reads 1 Mb of DNA to predict thousands of tracks (Nature, Jan 2026); genome LMs like Evo generate functional sequences.
Vendor claim Mathematical & physical sciences · OECD 1.1Mathematics
L3Systemic transformation↑ Rising ↓ from L4 AI ×7Field ×4Maths has the strongest AI-discovery record: in May 2026 an OpenAI model disproved Erdős’s 1946 unit-distance conjecture (a human-verified write-up by nine mathematicians incl. Alon, Gowers, Sawin), PNAS published AlphaEvolve’s 67-problem study, and a DeepMind agent resolved 9 open Erdős problems in Lean. Broad practitioner adoption and curriculum change have not happened, and openai/math saw withdrawals and push-back, so the composite is L3 (discovery scored 4).
Mathematical & physical sciences · OECD 1.4Chemistry
L3Systemic transformation↑ Rising ↑ from L2 AI ×1Field ×1Raised to L3: the 2024 chemistry Nobel went to AI structure prediction and design; retrosynthesis and ML potentials are routine; in 2026 autonomous platforms discovered new reactions and ligands (Nat. Commun. Jun & Oct 2026) and low-cost self-driving labs appeared. Day-to-day lab work is still largely manual.
Vendor claim Life & medical sciences · OECD 3.2Medical imaging & pathology
L3Systemic transformation→ Steady Field ×1Imaging is medicine's most AI-penetrated branch: radiology dominates FDA's AI device list; the MASAI RCT showed AI-supported mammography safely cut reading workload by ~44% (Lancet Oncology 2023).
Vendor claim Engineering & computing · OECD 1.2Computer science (algorithms & theory)
L3Systemic transformation↑ Rising AI ×2AI has found adopted algorithms (AlphaDev sorting merged into LLVM libc++; AlphaTensor) and improved hardness bounds (AlphaEvolve 2026); The AI Scientist appeared in Nature (Mar 2026).
Vendor claim Engineering & computing · OECD 1.2Software engineering
L3Systemic transformation↑ Rising ↓ from L4Lowered to L3: adoption is the highest of any field (Stack Overflow 2026: over 70% of respondents use AI daily), but Google’s ~75% and Anthropic’s 90%+ code shares are vendor claims and cannot alone justify L4. METR’s RCT found experienced OSS developers 19% slower in early 2025; 2026 follow-ups estimate speed-ups but with design limits, so independent productivity evidence is mixed.
Vendor claim Engineering & computing · OECD 2.2Cybersecurity
L3Systemic transformation↑ Rising AI ×2AI finds real vulnerabilities at scale: Anthropic validated 500+ high-severity bugs with Claude Opus 4.6 (Feb 2026), Microsoft FORGE contributed to 140 Windows CVEs (May–Sep 2026), TitanCA found 203 zero-days; Google saw the first likely AI-developed zero-day exploit in the wild. Close to level 4.
Earth & environment · OECD 1.5Meteorology & climate science
L3Systemic transformation↑ Rising AI ×1Field ×4AI weather models are operational: ECMWF's AIFS went operational in Feb 2025 and v2 on 12 May 2026; NOAA runs AI global models (AIGFS etc.). GraphCast, GenCast and Pangu laid the peer-reviewed groundwork.
Vendor claim Arts · OECD 6.4Music
L2Local breakthroughs↑ RisingAI music moved from lawsuits to licensing: Warner settled with Suno and Udio, Universal with Udio, and Suno launched label-licensed models (2026); IFPI issued global principles for AI recordings on official charts.
Mathematical & physical sciences · OECD 1.1Statistics
L2Local breakthroughs→ SteadyTabular foundation model TabPFN beats tuned gradient boosting on small tabular data (Nature 2025); prediction-powered inference enables valid inference with ML predictions (Science 2023). Core methods are not yet systematically rewritten.
Vendor claim Mathematical & physical sciences · OECD 1.3Physics
L2Local breakthroughs↑ Rising AI ×1Peer-reviewed results in quantum error-correction decoding (AlphaQubit, Nature 2024), autonomous law discovery (SciExplorer, PRX 2026) and particle-model building (Commun. Phys. 2026); not yet the mainstream way physics is done.
Contested Mathematical & physical sciences · OECD 2.5Materials science
L2Local breakthroughs→ Steady AI ×1GNoME predicted 2.2M crystals and MatterGen designs materials generatively (Nature 2023/2025), A-Lab synthesises autonomously; but the A-Lab paper was corrected in Jan 2026 (41/58 → 36/57), so “AI-discovered materials” claims remain contested.
Vendor claim Life & medical sciences · OECD 3.1Neuroscience
L2Local breakthroughs↑ Rising Field ×1AI enabled the adult-fly whole-brain connectome (FlyWire, Nature 2024) and speech-decoding BCIs (Nature 2023); 2026 saw many connectome foundation models, mostly preprints/benchmarks.
Vendor claimContested Life & medical sciences · OECD 3.1Pharmacy & drug discovery
L2Local breakthroughs↑ RisingAI-discovered drugs are in late-stage trials: Insilico's rentosertib Phase 2a in Nature Medicine (2025), now in Phase 3; Takeda's zasocitinib beat Sotyktu head-to-head in Phase 3. No “AI-designed” drug approved as of Oct 2026.
Life & medical sciences · OECD 3.2Clinical medicine
L2Local breakthroughs↑ RisingFDA lists 1,614 AI-enabled devices (4 Sep 2026 update); AMIE out-diagnosed clinicians in simulated consults (Nature 2025). Prospective evidence for LLMs in care is still thin.
Life & medical sciences · OECD 3.3Public health & epidemiology
L2Local breakthroughs→ SteadyEVEscape forecast SARS-CoV-2 escape mutations from pre-pandemic data (Nature 2023); WHO issued LMM governance guidance in 2024. Use is mostly surveillance and analysis support.
Life & medical sciences · OECD 1.6Ecology & biodiversity
L2Local breakthroughs↑ RisingAcoustic (BirdNET) and camera-trap classifiers are routine; Europe's Biodiversa+ pilot deployed ~200 AI sensors at 70 sites, moving from pilots to networks.
Vendor claim Engineering & computing · OECD 2.2Robotics & automation
L2Local breakthroughs↑ RisingVision-language-action models give robots some generalisation (RT-2 2023, π0 2024, Gemini Robotics 2025), but real-world reliability and deployment at scale remain unsolved.
Vendor claim Engineering & computing · OECD 2.7Energy & fusion
L2Local breakthroughs↑ Rising AI ×1Deep RL controls tokamak plasmas (Nature 2022) and AI avoids tearing instabilities (Nature 2024); DeepMind and CFS use TORAX to plan SPARC operations.
Contested Earth & environment · OECD 1.5Earth sciences (seismology, geology, oceans)
L2Local breakthroughs→ SteadyDeep-learning phase picking (PhaseNet, EQTransformer) is standard for building catalogues; in 2026 AI gives aftershock risk in seconds, but individual large-quake prediction remains unreliable.
Earth & environment · OECD 1.5Geography & remote sensing
L2Local breakthroughs↑ RisingGeospatial foundation models are replacing per-task models: NASA/IBM open-sourced Prithvi; DeepMind released AlphaEarth Foundations global embeddings.
Social sciences · OECD 5.2Finance
L2Local breakthroughs→ SteadyML asset pricing (Gu, Kelly & Xiu, RFS 2020) is a mainstream empirical method; quant and risk use is widespread, while LLMs remain assistive.
Social sciences · OECD 5.3Education
L2Local breakthroughs↑ RisingRCT evidence is arriving: a Harvard physics AI tutor beat active-learning classes (Sci. Rep. 2025); a 2026 NBER two-year Khanmigo trial found ~0.06–0.08 SD/year maths gains, with low engagement.
Vendor claim Social sciences · OECD 6.2Linguistics
L2Local breakthroughs→ SteadyNLLB extended neural MT to 200 languages (Nature 2024), helping low-resource documentation; LLMs are now evidence in debates on acquisition and syntax.
Vendor claim Arts · OECD 6.4Visual arts & design
L2Local breakthroughs→ Steady AI ×1AI works have entered markets and prizes: Christie's sold “Portrait of Edmond de Belamy” in 2018, and a Midjourney image won the Colorado State Fair digital-art prize in 2022; design widely uses generative tools.
Vendor claim Arts · OECD 6.4Film & animation
L2Local breakthroughs↑ Rising AI ×1Text-to-video (Sora from 2024) has entered production; the 99th Oscars rules (May 2026) say generative AI use does not disqualify a film, but acting and writing awards require human work.
ContestedVendor claim Engineering & computing · OECD 2.2Electronics & chip design
L2Local breakthroughs↑ RisingRL chip placement (AlphaChip) was used for several TPU generations (Nature 2021) but its reproducibility is disputed (CACM re-evaluation); in 2026 Synopsys pushes agentic EDA, with AMD reporting 25–40% shorter debug cycles (vendor figures).
Vendor claim Engineering & computing · OECD 2.3Aerospace & mechanical engineering
L2Local breakthroughs→ Steady AI ×1Under DARPA ACE, AI flew the X-62A in within-visual-range dogfights against a crewed F-16 (announced 2024) — a local breakthrough; design and manufacturing use is still mostly assistive.
Contested Social sciences · OECD 5.1Psychology
L2Local breakthroughs→ SteadyCentaur, fine-tuned on 10M+ human choices, predicts behaviour better than classic cognitive models (Nature 2025), though evaluations show divergence from humans; AI dialogues durably reduced conspiracy beliefs (Science 2024).
Vendor claim Mathematical & physical sciences · OECD 1.3Astronomy & astrophysics
L2Local breakthroughs→ SteadyML finds exoplanet candidates and supernovae at scale (≈10,000 new TESS candidates, ApJS 2026; NASA ExoMiner++ ≈7,000), though most still need vetting and follow-up.
Humanities · OECD 6.1Archaeology & classics
L2Local breakthroughs↑ Rising Field ×2Ithaca (Nature 2022) and Aeneas (Nature 2025) restore and date inscriptions; the Vesuvius Challenge read carbonised Herculaneum papyri (2024 prize; ~1.5 m of Greek from PHerc. 1667 in 2026); AI helped find 303 new Nazca geoglyphs (PNAS 2024).
Contested Social sciences · OECD 5.8Journalism & communication
L1Assistive↑ RisingReuters Institute DNR 2026: use of AI chatbots for news rose from 7% to 10%; UK newsrooms use AI for transcription, translation and verification, but integration is limited.
Vendor claim Life & medical sciences · OECD 4.1Agricultural sciences
L1Assistive→ SteadyAI is used for genomic selection, phenotype prediction and pest ID; 2026 Nature Communications crop models are benchmark gains, with no field-validated landmark AI breeding result yet.
Engineering & computing · OECD 2.1Architecture & civil engineering
L1Assistive→ SteadyBroad but shallow: RIBA's 2026 report says 74% of practices use AI on some projects, mostly general generative tools; structural engineers keep judgment central. No recognised AI-originated design breakthrough.
Social sciences · OECD 5.2Economics
L1Assistive→ SteadyStays L1 (close to L2): ML causal inference and text methods are common and a 2026 NBER paper uses an LLM workflow to reproduce and extend published work, but a 74-country Nature Human Behaviour study shows simulated LLM respondents are weak proxies for human preferences; no recognised AI-key discovery yet.
Social sciences · OECD 5.4Sociology & political science
L1Assistive→ Steady“Silicon samples” (LLM-simulated respondents, Political Analysis 2023) spark debate but reliability is doubtful (Nat. Hum. Behav. 2026); main uses are text coding and survey support.
Social sciences · OECD 5.5Law
L1Assistive↑ RisingLegal practice uses AI widely, with problems: Charlotin's database lists 2,202 decisions on AI hallucinations, Reuters counts 1,395 US cases; >60% of surveyed federal judges have used AI but only 22.4% weekly.
Humanities · OECD 6.1History
L1Assistive↑ RisingHandwritten text recognition (e.g., Transkribus) speeds up archive transcription; Schmidt Sciences funded a $10M AI-and-humanities programme in 2026, but historical interpretation remains human.
Vendor claim Social sciences · OECD 3.3Sports science
L1Assistive→ Steady ↓ from L2Lowered to L1: AI in sport is mainly tracking, video analysis and officiating aids (e.g. semi-automated offside at the 2022 World Cup); AI-key research is limited to a few tools such as TacticAI (Nat. Commun. 2024).
Vendor claim Humanities · OECD 6.3Religious studies
L0Not yet transformed→ Steady ↓ from L1Lowered to L0: in religious studies AI is mainly an object of study and an ethical topic (Rome Call 2020, encyclical Magnifica humanitas 2026); tool use is sporadic. Composite below 0.75.
Contested Humanities · OECD 6.2Literature
L0Not yet transformed→ Steady ↓ from L1Lowered to L0: isolated cases in creative writing (Akutagawa Prize 2024) and experiments (AI poetry hard to tell apart in blind tests), but no AI-key results in literary studies and no adoption data. Composite below 0.75.
Humanities · OECD 6.3Philosophy
L0Not yet transformed→ Steady ↓ from L1Lowered to L0 (not yet transformed): AI is mainly an object of philosophical study; tool use is limited to sporadic literature-review attempts and LLM experiments in experimental philosophy. Composite score below 0.75.
No discipline matches these filters.
Latest progress
100 items · ● milestone ○ news · headlines in their original language
How levels are defined
L0Not yet transformed
Composite < 0.75: only scattered use or discussion; AI has not changed methods, discovery, practice or teaching.
Reviewed every 180 days · 3 disciplines
L1Assistive
Composite 0.75–1.49: AI is an aid in parts of research or practice, with cases or surveys, but no recognised AI-key result.
Reviewed every 90 days · 8 disciplines
L2Local breakthroughs
Composite 1.5–2.49: at least one peer-reviewed/officially verified AI-key result, or one dimension at 3; routine workflow largely unchanged.
Reviewed every 90 days · 23 disciplines
L3Systemic transformation
Composite 2.5–3.49 (or ≥3.5 with any of method/discovery/adoption below 3): AI is standard infrastructure with multiple independent breakthroughs.
Reviewed every 60 days · 9 disciplines
L4AI-led
Composite ≥ 3.5 and method, discovery and adoption all ≥ 3: AI produces core results at scale, independently verified.
Review cycle: see method · 0 disciplines
How the score works
S = 0.30·method + 0.35·discovery + 0.25·adoption + 0.10·education (education scores 0 when there is no evidence; if empty, the other weights are renormalised)
Research method ×0.3Scientific discovery ×0.35Practitioner adoption ×0.25Education & curriculum ×0.1
- Leap — 解决长期公开难题、首次超越最强人类/现有最佳方法、或改变学科常规做法，且经同行评审或独立专家/官方核验
- Major — 顶刊/顶会成果、重要系统发布、规模化部署或政策，影响明显但局部
- Incremental — 改进、数据集、调查、个案或早期信号
Labels
- Vendor claimOnly the vendor’s own account; figures and claims are not independently verified.
- ContestedPublicly challenged, withdrawn or corrected; the claim is disputed.
- PreprintPreprint (arXiv, bioRxiv …), not yet peer-reviewed.
- To be verifiedDate or details not yet confirmed (e.g. only year or month known).
Method & disclaimer
Levels are computed from four dimension scores; vendor-claimed figures are not independently verified; see impactRule for breakthrough ratings.
- Link checked (HTTP 200)
- Site blocks scripts; DOI title and date checked via Crossref
- Site blocks scripts; checked by hand
- From RSS/API feed; link is the original
- Link not yet checked
News are matched by discipline keywords from the site’s news collection and other feeds; 35 items also appear in AI News. Data as of 2026-10-22; machine-readable: /api/v1/disciplines.json