ICO Launches Consultation on the Application of the Accuracy Principle to Generative AI

As part of the ICO’s ongoing efforts to provide clarity on the application of the UK GDPR to artificial intelligence (AI), a consultation has been issued focusing on the application of the accuracy principle to generative AI outputs and the accuracy of training data and model output. This consultation forms part of the ICO’s aim of developing comprehensive guidance on the impact of AI on the UK’s data protection laws. 

The accuracy principle 

Accuracy is a key principle under the GDPR and Article 5(1)(d) mandates that organisations are under an obligation to ensure that all personal data is accurate and current. Similarly, any inaccurate personal data is required to either be rectified. 

Accuracy and AI 

The ICO’s consultation discusses how developers and deployers of generative AI models should adhere to accuracy principle. The consultation recognises that it is a common misconception that AI outputs must be infallibly precise. However, the ICO clarifies that the necessity for statistical accuracy is not absolute but contingent upon the AI model's application.

For AI systems that influence decision-making processes, especially those concerning individual, (e.g. processing customer inquiries), precision is paramount. The repercussions of inaccuracies in such contexts can be significant, and therefore, the ICO insists on a higher standard of statistical accuracy to ensure fairness and accountability.

Conversely, when AI is employed to spark creativity, such as in generating concepts for media or video games, the stringency of accuracy relaxes. Here, the focus shifts to fostering innovation and creativity, where statistical precision, while still important, becomes less critical.

Training data 

The ICO has further recognised that the accuracy of generative AI outputs will be largely influenced by the quality of training data. It is therefore critical that those responsible for the development of AI models, ensure that training data is curated carefully, and in a way that is reflective of the intended purpose of the AI model. As the requirement for accuracy is dependent on the AI model’s application, developers will need to ensure particular precision is applied to training data that will be utilised for decision making processes involving individuals. 

Our views

In order to ensure the compliance with the accuracy principle, developers and deployers of AI should ensure that the is sufficient communication as to the AI model’s intended purpose and technical accuracy of the generative output. Additionally, sufficient information should be provided where there are any hesitations as to the accuracy or reliability of AI output and clear expectations are set out for end users regarding output accuracy.