Nobel Turing Challenge
Capability benchmark for autonomous, independently validated discovery.
Scientific Program
AI as a Scientist The final frontier of discovery.
Nobel Turing Challenge
Science of Science
Continuous Operation
We are rebuilding the scientific method for an era of massive data, computation, and autonomous laboratories. AI systems read, reason, hypothesise, experiment, and validate alongside researchers, turning science into a continuously operating engine.
Program framing
AI as a Scientist is authored and led by Hiroaki Kitano, Ph.D. (Adjunct Professor, Okinawa Institute of Science and Technology; President, The Systems Biology Institute; Chief Technology Fellow, Sony Group Corporation; President & CEO, Sony Computer Science Laboratories). SBI & SBX maintain the program and infrastructure.
- AI systems operate as scientists with high autonomy.
- Closed-loop experimentation and knowledge consolidation.
- Explicit governance, provenance, and evaluation.
- Human–AI hybrid teams that scale autonomy by tier.
Science of Science
Formalising how breakthroughs are produced, governed, and replicated.
Continuous Operation
Always-on discovery pipelines spanning domains and modalities.
01
Vision
Vision
From pre-industrial craft to continuously operating science
Discovery today is still limited: hypotheses are selected narrowly, iteration cycles take months, and literature volumes exceed human cognition. AI as a Scientist establishes systems that operate without interruption, exploring vast hypothesis spaces, running the experiments they design, and consolidating knowledge with complete provenance.
“We need AI scientists that can perform high-impact research highly autonomously… ‘AI as a Scientist’ may be the only viable path to the future of scientific progress.” — Hiroaki Kitano
Why this matters
- 01 Millions of papers each year bury critical findings deep in the long tail.
- 02 Complex biological and engineered systems are highly interconnected, exploding the hypothesis space.
- 03 Truly unbiased exploration demands autonomy well beyond traditional analytical tools.
02
Approach
Approach
Programmatic science instead of monolithic personas
Programmatic science
Multiple interoperable systems act as scientific agents across reading, reasoning, experimentation, and verification. Governance, evaluation, and transparency are built in from the first line of code.
- Human–AI hybrid teams with autonomy that increases by tier.
- Open-domain hypothesis search with auditable traceability.
- Reproducible, citable discovery pipelines.
Monolithic “AI scientist” (not our goal)
One-shot personas that mimic a single researcher tend to be narrow, under-specified, and weakly accountable. The future of science requires process, not characters.
- Persona over process.
- Domain-bound demos.
- Opaque provenance.
Evidence
Signals pointing to autonomous discovery
03
Roadmap
Roadmap
The Nobel Turing Challenge anchors first-generation goals
The Nobel Turing Challenge sets the bar: AI systems must repeatedly produce independently validated discoveries at the highest level. Success is measured in capability and cadence, not trophies. This framing guides how we design autonomy, assurance, and community engagement.
Principles
- A Science of science — formalise the discovery process.
- B Open hypothesis spaces — avoid pre-narrowed topics.
- C Continuous operation — discovery as an always-on process.
04
System
System properties
What defines an AI scientist
Core properties
- End-to-end autonomy across read → plan → experiment → analyse → share.
- Full auditability and provenance for reproducibility.
- Generalisation across domains and modalities.
Evaluation
- Novelty, correctness, and independent replication.
- Community acceptance and downstream impact.
- Long-run cadence of significant results.
05
Method
Method
The end-to-end discovery loop
01
Knowledge extraction
Normalise literature, data, and code into machine-actionable graphs and models with explicit uncertainty.
02
Hypothesis generation
Enumerate and rank hypotheses using symbolic and statistical methods, maximising expected information value.
03
Robotic experimentation
Automated labs execute protocols around the clock, with results streaming directly back to models.
04
Verification & consolidation
Independent replication, adversarial tests, and aggregation into validated knowledge with full provenance.
06
Autonomy
Autonomy tiers
Scaling capability and measuring success
Autonomy tiers
- T1 — human-in-the-loop hypothesis triage.
- T2 — autonomous design with supervised execution.
- T3 — end-to-end autonomy backed by audit and veto gates.
- T4 — multi-domain autonomy coupled with self-evaluation.
Success metrics
- Scientific: reproducibility, replication rate, novelty beyond training priors, cross-domain robustness.
- Impact: citations, downstream adoption, clinical or industrial translation, cadence of high-impact findings.
07
Cases
Case studies
Deployments across research and talent programs
Digital Twins in Oncology
Multi-component drug design studies leverage Garuda and SBML models to balance robustness trade-offs in cancer systems.
Explore research programsTalent & Education
SBI sponsors IIT Delhi iGEM teams and hosts AI for Science meetings in Tokyo to cultivate the next generation of scientist–AI collaborations.
Team announcement
08
References
References
Foundation literature and resources
- Kitano, H. (2021). Nobel Turing Challenge: creating the engine for scientific discovery. npj Systems Biology & Applications.
- Kitano, H. (2016). Artificial Intelligence to Win the Nobel Prize and Beyond: Creating the Engine for Scientific Discovery. AI Magazine.
- National Academies (2023). AI for Scientific Discovery — proceedings outlining the 2050 objective.
- Nobel Turing Challenge — publications & background; Nobel Turing Challenge — official site.
09
Maintained
Maintained by
The Systems Biology Institute (SBI) and SBX maintain AI as a Scientist and its supporting infrastructure. © 2025 AI as a Scientist.
10
Network
Collaborative network
Institutions powering AI as a Scientist
RIKEN
Advanced biosystems research center
AIST
National Institute of Advanced Industrial Science and Technology
MIT CSAIL
Computational biology and AI lab exchange
RIKEN BDR
Biosystems Dynamics Research collaboration
RIKEN-EMBL
International systems biology alliance
JST Moonshot
Moonshot Goal 9 digital twin consortium
11
Engage