Terms you’ll run into when selling data
Plain-English definitions of the terms used when companies sell data to AI labs: de-identification, licensing, PII, training data and more.
A
- AI training data
- AI training data is the data a machine learning model learns from: the examples used to fit or adjust its learnable parameters. It is distinct from validation data, testing data, and the input data a finished system receives when in use.
- Anonymization
- Anonymization is processing data so the people it describes can no longer be identified by anyone using means reasonably likely to be used. Under GDPR and UK GDPR, truly anonymous data is no longer personal data. Pseudonymized data, where identities are only swapped for codes, still is.
D
- Data broker
- A data broker is a business that collects personal information about people it has no direct relationship with and sells, licenses or otherwise shares that information with others. Exact legal definitions vary by law and jurisdiction.
- Data marketplace
- A data marketplace (sometimes called a data exchange) is an online platform that matches those who hold data, usually organizations, with those who want to use it, so data can be licensed, sold or shared on agreed terms. EU law treats data marketplaces as one kind of data intermediation service.
- Data monetization
- Data monetization is turning data and analytics into financial returns. This can happen directly, by selling or licensing data or information products, or indirectly, by using data to improve operations or make products more valuable.
- Data valuation
- Data valuation is estimating what data is worth, usually from market prices for comparable data, the cost of producing it, or the income it can generate. In statistics and accounting it means a money value; in machine learning it often means scoring how much each record helps a model.
- De-identification
- De-identification is any process that removes the link between data and the people it describes, limiting the risk of identifying anyone while keeping the data useful. The legal bar differs: HIPAA and California's CCPA each set their own test, and EU law treats pseudonymised data as still personal.
E
- Exclusive license
- An exclusive license gives one licensee the right to use what is licensed to the exclusion of everyone else, often including the owner. A non-exclusive license lets the owner license the same rights to others.
I
- Indemnification
- Indemnification is a contract promise by one party to compensate the other for specified losses, such as third-party claims. In a data license, the buyer may ask the seller to cover claims that the data infringed rights or broke privacy law.
- IP assignment
- An IP assignment transfers ownership of intellectual property from one party to another, so the buyer becomes the new owner. A license, by contrast, lets someone use the property on agreed terms while the owner keeps title.
O
- Opt-out (AI training)
- An AI training opt-out is a way for a content owner, customer or individual to say their content or data should not be used to train AI models. In EU copyright law it is an express reservation of rights against text and data mining, made by machine-readable means for content published online.
P
- PII (personally identifiable information)
- PII (personally identifiable information) is information that can identify a person, alone (like a name or email address) or combined with other linked data. US federal guidance uses "PII"; EU and UK law use "personal data" and California's CCPA uses "personal information," each with its own legal definition.