AITB No. 03 Accounting for Data Acquisitions for AI Model Training
Issue: How should entities account for the costs associated with acquiring datasets specifically for training AI models?
Background: Datasets are integral to training AI models, making them functionally efficient. With the surge in AI applications, there's a parallel increase in the procurement of relevant datasets, often at significant costs.
Guidance:
- Capitalization vs. Expensing: If a dataset is acquired for a specific AI project with future economic benefits, the cost should be capitalized. However, if the data is for general training purposes without direct ties to a specific project, it should be expensed.
- Amortization of Capitalized Data Costs: Capitalized costs related to datasets should be amortized over the expected period of benefit, considering the rapid obsolescence of data in certain industries.
- Treatment of Data Updates: Periodic updates or additions to existing datasets, if significant in cost, should be evaluated for capitalization based on their expected contribution to future economic benefits.
- Impairment Considerations: Given the dynamic nature of data relevance, entities should regularly assess capitalized data assets for impairment.
Examples:
- Company C acquires a dataset for $2M, specifically for a new AI project expected to last 5 years. They would capitalize the $2M and amortize it over 5 years.
Note: This is a fictional representation and does not represent any real-world standard for AI. The development of such standards would involve extensive consultations with experts, stakeholders, and the public. Fictional representations simplify complex AI concepts, stimulate discussion, envision future scenarios, highlight ethical considerations, encourage creativity, bridge knowledge gaps, and set benchmarks for debate in fields like accounting.