Was the claim tested empirically, computationally or only argued?
- Evidence rule
- Distinguish generated hypotheses from validated discoveries.
FUURAA™ · 2026 AI Frontier Technology Radar
Scientific Intelligence & Discovery
Assess how AI supports literature reasoning, hypothesis formation, scientific software, simulation, causal inquiry and discovery.
Watch thesis
Scientific value depends on whether an AI-generated idea survives measurement, replication, falsification and expert challenge.
Applicability boundaryFluent scientific writing, benchmark performance or plausible hypotheses are not discoveries without appropriate validation.
This page separates canonical sources, FUURAA original summaries and analysis. Inclusion does not imply partnership, endorsement or approval.
What this lens tracks
Declare the observation boundary before admitting a new development into the evidence chain.
- 01AI scientists, co-scientists and scientific agents
- 02scientific software, simulation and causal methods
- 03hypothesis generation and empirical validation
Three evidence questions
Ask not only what happened, but how far the evidence travels.
Are methods, data, code and negative results available?
- Evidence rule
- Prioritise peer-reviewed work with inspectable methods and artefacts.
What did independent experts or laboratories verify?
- Evidence rule
- Record failed replications, fabricated references and expert corrections.
Verified primary sources
9 directly relevant 2026 source records for this lens.
Each record states its publication date, source organisation, evidence type, FUURAA original summary and known boundary.
13 July 2026 · Original authorsCDFM: Towards a General-Purpose Causal Discovery Foundation Model
Evidence statusPrimary source · Preprint
CDFM treats unknown causal mechanisms as latent variables and pretrains across diverse synthetic structural causal models to pursue general-purpose zero-shot structural inference.
Evidence boundaryPreprint; not yet peer reviewed.
Open canonical source ↗3 July 2026 · npj Clean EnergyQuantifying drivers of photovoltaic power generation at Bhadla using explainable machine learning and causal discoveryEvidence statusPrimary source · Research
The study combines explainable machine learning with causal discovery to distinguish predictive correlations from plausible drivers of solar-power performance.
Open canonical source ↗25 June 2026 · Scientific Reports / Pacific Northwest National Laboratory-led teamAutoLabs: cognitive multi-agent systems with self-correction for autonomous chemical experimentationEvidence statusPrimary source · Research
AutoLabs converts natural-language requests into executable liquid-handler protocols; its benchmarks indicate that modular agents and iterative self-correction can reduce experimental errors.
Open canonical source ↗12 June 2026 · Nature ElectronicsSustainability assessment using multimodal artificial intelligence agentsEvidence statusPrimary source · Research
Multimodal agents retrieve and combine product, material and lifecycle information to estimate environmental impacts, while also exposing the compute and retrieval costs of agentic assessment.
Open canonical source ↗19 May 2026 · Nature / Google Research-led teamAn AI system to help scientists write expert-level empirical softwareEvidence statusPrimary source · Research
ERA combines language models with tree search to create measurable scientific software, reporting results across single-cell analysis, epidemiological forecasting, geospatial analysis and neuroscience.
Open canonical source ↗19 May 2026 · NatureA multi-agent system for automating scientific discoveryEvidence statusPrimary source · Research
A multi-agent system coordinates literature reasoning, hypothesis development, data analysis and experiment planning, showing how specialised agents can support expert-led discovery.
Open canonical source ↗8 May 2026 · Original authorsArrow: A Foundation Model for Causal DiscoveryEvidence statusPrimary source · Preprint
Arrow pretrains a reusable model to infer causal graphs from new tabular datasets without task-specific retraining, while enforcing acyclic graph structure by design.
Evidence boundaryPreprint; not yet peer reviewed.
Open canonical source ↗25 March 2026 · NatureTowards end-to-end automation of AI researchEvidence statusPrimary source · Research
The AI Scientist pipeline connects ideation, coding, experimentation, analysis, manuscript preparation and review, while the study also records uneven quality, fabricated citations and risks to peer-review systems.
Open canonical source ↗22 January 2026 · Nature CommunicationsAssimilative causal inferenceEvidence statusPrimary source · Research
Assimilative causal inference combines Bayesian data assimilation with backward tracing from observed effects to study instantaneous and time-varying causal relationships in complex systems.
Open canonical source ↗FUURAA analysis
AI can compress parts of the research cycle, but the scarce resource remains trustworthy contact with reality. Institutions that preserve negative evidence and reproducibility will create more durable value.
How to use this lensUse these records as a starting point for continuing observation—not as a complete market map, investment advice, certification conclusion or certain forecast. Source updates, version changes and independent replication can change the assessment.
Continue across the nine lenses
Technical progress often crosses model, infrastructure, deployment and governance boundaries.
Models & Multimodal IntelligenceAgents, Tools & Digital InfrastructureRobotics & Embodied IntelligenceChips, Compute, Data & EnergyAutonomous Laboratories & Experiment SystemsScientific Intelligence & DiscoveryBiology, Health & Digital LifeReal Products, Work & DeploymentSafety, Evaluation, Policy & SocietyReturn to the 2026 AI Frontier Technology Radar