Publications · Academic reports
Each research question the CoLab takes on is being written up as an academic report. Titles are working drafts; links go live here as each report is published.
01 / Evacuation
In preparationHow can AI inform evacuation decisions?
A write-up of the evacuation-risk models, information index, and behavioral simulations, and what they imply for who reaches safety.
02 / Cultural heritage
In preparationHow can technology support the ethical return of cultural artifacts?
A report on pairing verifiable credentials with grounded, cited provenance to support repatriation, attribution, and ethical stewardship.
03 / Traceability
In preparationHow can ethical claims in supply chains be made verifiable?
A report on structural forced-labor risk and interoperable trace records that let buyers and regulators verify claims without exposing supplier data.
04 / Diplomacy
● PublishedHow can AI help practitioners rehearse high-stakes diplomacy?
A report on AI-mediated environments where negotiators rehearse against agents that model historical context, incentives, and red lines.
05 / Sustainability
● PublishedMay 2025
How large is AI's environmental footprint — and how can it be reduced?
A CoLab report on AI's energy use across training and inference, the data-center and hardware toll, and the mitigation strategies, regulations, and policies that could bend the curve.
06 / Research
● PublishedOctober 2025
How can AI help researchers formulate rigorous research questions?
A survey of how AI supports researchers across the question-formulation workflow — finding gaps, generating candidate questions, summarizing the state of the art, and flagging contradictions — with guardrails against hallucinated citations and a red-team verification model. Ships with a reusable Copilot 'Researcher' prompt and a journal-credibility rubric.
07 / Evacuation
● PublishedJuly 2026
Is leaving actually safer than staying, and how would anyone know?
A plain-language report on a decision support prototype that scores 104 active crises on two risks at once, keeping danger and feasibility deliberately apart so that operational difficulty cannot quietly cancel out a legal obligation. Prepared as masters research at the NYU Center for Global Affairs.
08 / Evacuation
● Published2026
How dangerous is it to stay, and is leaving actually possible?
A plain-language report on a prototype that keeps two questions apart: how dangerous it is for civilians to remain, and whether an organised evacuation is possible at all. Separating them matters legally, because the duty to protect arises from danger, not from operational convenience.
09 / Evacuation
● Published2026
What does it actually cost to move a besieged population, and what does staying cost?
A report on a framework that scores seven weighted risk dimensions, estimates the cost of evacuating against the cost of staying, and finds the point at which those two curves cross. Calibrated against 31 documented historical operations.
10 / Evacuation
● Published2026
How much does poor field information degrade an evacuation decision?
A report on a simulator that models what happens when planners act on incomplete intelligence. All data is synthetic and reproducible from a seed, so the mechanism rather than any particular place is what is on display.
11 / Evacuation
● Published2026
How do information and demographics decide who actually leaves?
A report on an agent-based model of household evacuation behaviour, tracing how warnings spread through a community and how age, mobility, and trust shape who reaches safety and who does not.
12 / Evacuation
● PublishedJuly 2026
Could a daily severity measure have shown the danger in Mariupol as it unfolded?
A retrospective report scoring 77 days of the siege on a daily severity measure built from open sources. Conduct descriptions are attributed to their sources rather than adjudicated, and contested figures are left contested.
13 / Cultural heritage
● PublishedJuly 2026
Can an artwork's ownership history be checked automatically, and how far should that be trusted?
A report on an assistant that assembles a provenance timeline, flags looting and valuation risk, and issues a signed record. Candid about how much of the demonstration runs on fixtures rather than live sources.
14 / Cultural heritage
● PublishedJuly 2026
What can be established about an object's history from free public sources alone?
A report on a tool that queries seven public sources and scores confidence with a fixed, inspectable algorithm, deliberately flagging what cannot be verified rather than filling the gaps.
15 / Cultural heritage
● Published2026
What is worth recording about a visit to a work of art?
A report on a digital souvenir passport that records attendance at art experiences. It performs no provenance or authenticity checking, and the report is explicit that nothing it records is evidence of ownership or lawful origin.
16 / Traceability
● Published2026
Where are the conditions that make forced labour likely, before any case is detected?
A report on an index scoring 184 countries across 11 domains and 43 indicators, measuring the conditions that enable forced labour rather than counting cases. Ships with uncertainty bands and is meant to be read in tiers, not as a league table.
17 / Diplomacy
● PublishedJuly 2026
Can practitioners rehearse a multi-party negotiation against AI agents?
A report on a simulator in which AI delegations negotiate a live crisis, producing a structured record of positions, concessions, and tactics. Across every Monte Carlo trial run, no comprehensive settlement was reached.
18 / Disaster response
● Published2026
How fast can building damage be assessed after a disaster, and with how few labels?
A plain-language report on a satellite damage-assessment platform built by the Microsoft AI for Good Lab, used in 31 field deployments. The software is Microsoft's; only this report is the CoLab's.