ChatGPT: Mitigating the legal risks

"Large language models, like ChatGPT, need a large bank of training data to be able to provide valuable results. That training data is likely to engage intellectual property rights, most likely copyright.  If protected materials are used as training data, then the creator of the AI should clear this with the owner first. In the UK, a proposed exemption to permit general use of materials for text and data mining was being considered, but it was recently stated in a House of Commons debate that the UK government won’t be proceeding with this approach.

In ChatGPT’s case, OpenAI has elected (for now) to train the model only on pre-2021 data and not connect it to the internet, a decision which may partially mitigate this risk. On a practical level, it can be difficult for a rights holder to establish a breach by an AI tool, depending on the output. Other AI models have had issues - Getty Images launched a claim against Stability AI for purportedly training its AI using Getty Image’s data.

Once ChatGPT is commercially integrated into Bing, rights holders may be rightly concerned that this could result in a reduction in click-through rate. This will depend on how the technology is implemented, and is a recurring trend in the industry as search engines seek to keep users on-page (such as via Google’s ‘Featured Snippets’)."

‍

Andrew's comments were published in Press Gazette, 9 March 2023, and can be found here.