techUK is pleased to host a member-only roundtable with the Information Commissioner’s Office (ICO) on its generative AI and data protection consultation series.
The ICO will present and discuss its initial analysis from responses, and members will be able to provide further input to support the development of the ICO's guidance. The session will follow Chatham House rules, and attendee details (including job titles) will be shared with the ICO in advance.
The ICO is keen to, as much as possible, gather input from technical teams, including product managers, machine learning and NLP engineers, and data scientists. The ICO seeks insights on implementing data protection by design and by default in GenAI development.
Discussion topics include:
- Did your organisation use web-scraped data to train its models? Did your organisation have to? What other data sources did your organisation use and why?
- How is your organisation defining a purpose for the development of its models in practice?
- What is your organisation's experience of determining the purpose of each stage of the generative AI supply chain? How did your organisation go about it and what purposes did it use?
- What technical and organisational measures do developers and deployers (or downstream developers that fine-tune general purpose models) use in practice to allow individuals to exercise their information rights? These rights include the right to be informed, the right of access, and rights to erasure, to rectification, to restriction of processing and objection to processing. What are the benefits and drawbacks of these measures?
- What technical and organisational measures do developers and deployers use to ensure the accuracy of inputs and outputs, where accuracy is required by the purpose for which the model will be used? What are the benefits and drawbacks of these measures?
- How do developers demonstrate the provenance of training data in practice, and communicate that to the public/individuals and downstream deployers? What are the benefits and drawbacks of the techniques used?
- What methods (and their effectiveness) do developers use to identify and remove personal data before training, if they do so?
Audre Verseckaite
Audre joined techUK in July 2023 as a Policy Manager for Data. Previously, she was a Policy Advisor in the Civil Service, where she worked on the Digital Markets, Competition and Consumers Bill at the Department for Science, Innovation and Technology, and at HM Treasury on designing COVID-19 support schemes and delivering the Financial Services and Markets Bill. Before that, Audre worked at a public relations consultancy, advising public and private sector clients on their communications, public relations, and government affairs strategy.
Oliver Alderson
Oliver is the team assistant for the Policy and Public Affairs teams, joining techUK in November of 2023. He assists the teams admistrative support, communication, and event production.