Ithaca raises historians' accuracy from 25 to 72 percent

AI system

System: Ithaca (transformer-based deep neural network for epigraphic restoration and attribution)
Model: Ithaca, a transformer-based neural network.
Release: Exact release not established.
Application developer: DeepMind with Ca' Foscari University of Venice, University of Oxford and Athens University of Economics and Business
Role: Ithaca proposed missing text, dates, and places of origin for ancient inscriptions.
Configuration: Researchers used the 2022 Ithaca system.
Inputs and outputs: Not assessed
Autonomy: Not assessed
Workflow: A large digitized corpus of inscriptions with known provenance and date. A neural network trained on artificially damaged text. Evaluation against held-out inscriptions and with historians in the loop.
Human oversight: Not assessed
Relation to AI: Ithaca's outputs were evaluated against held-out ground truth and in a study of historians working with and without it. The 421 BC dating is a model prediction consistent with evidence established by other means, not an outcome the model caused.
AI contribution: Ithaca's outputs were evaluated against held-out ground truth and in a study of historians working with and without it. The 421 BC dating is a model prediction consistent with evidence established by other means, not an outcome the model caused.

Outcome identifiers: Not assessed
Benefit: raised historians' accuracy from 25 to 72 percent when they worked with it
Category: Improved Task Accuracy
Status: Not assessed
Beneficiary groups: Society.
Beneficiaries: Not assessed
Reach: Not assessed
Distribution and access: Not assessed
Duration: Not assessed
Expected?: Yes
Costs and harms: Not assessed

Studies: Not assessed
Study design: Not assessed
Measurement: Not assessed
Setting and sample: Not assessed
Measures: raised historians' accuracy from 25 to 72 percent when they worked with it
Comparison: historians restoring damaged texts without the tool
Results: Reference: 25%. Result: 72% (historians using Ithaca); 62% for Ithaca alone.
Additional value from AI: Not assessed
Limitations: Not assessed
Denominators: Not assessed
Materials for replication: Not assessed

Participants: Developer: DeepMind with Ca' Foscari University of Venice, University of Oxford and Athens University of Economics and Business. Operator: Research consortium; public interface at ithaca.deepmind.com.
Jurisdictions: Developer: United Kingdom.
Ownership and control: Not assessed
Dependencies: Not assessed
Cross-border connections: Not assessed
Geopolitical implications: Not assessed

Conditions for replication: A verified event of a benefit created or restored would require an epigrapher-accepted restoration or re-dating attributable to the model and documented in a non-vendor source.
Replication attempts: Not assessed
Later findings: 23 July 2025. DeepMind followed Ithaca with Aeneas, a successor model for Latin inscriptions published in Nature on 23 July 2025, extending the approach to a second language and to retrieval of parallel texts.
Availability: Not assessed
Institutional decision: Not assessed
Grounds: Not assessed
Action and result: Not assessed
Success criterion: Not assessed

Assessed benefits: Society.
Depth: Not assessed
Reasons: Society: The sources give accuracy measurements and a prediction that agrees with independent re-dating of Athenian decrees. No discrete, verified downstream result is attributable to Ithaca.
Observed or intended: Not assessed
Rubric: Not assessed
Assessor and date: Not assessed
Alternative assessments: Not assessed
Fill: Independently studied

Kind: Not assessed
Related entries: Not assessed
Field: Humanities / epigraphy. Academic research.
Purpose: Built to restore, geographically attribute and date ancient Greek inscriptions; the reported capability was that purpose.
Dates: Date basis unassessed: 2022 (precision unassessed). Updated: 21 October 2026.
Stage: Not assessed
Version: SHA-256: 302890ae1f7d…

Peer-reviewed study Developer

Assael, Y., Sommerschield, T. et al. (2022). Restoring and attributing ancient texts using deep neural networks. Nature 603, 280–283.
Report Independent author

Roueché, C. (2022). AI minds the gap and fills in missing Greek inscriptions. Nature, News & Views, 9 March 2022. doi:10.1038/d41586-022-00641-2
News report Independent author

Google DeepMind's new AI can help historians understand ancient Latin inscriptions. MIT Technology Review, 23 July 2025.
Press release Developer

Ca' Foscari University of Venice (2022). AI to aid historians in restoring, placing and dating ancient texts. University news.