Thanks to RightsZone intern Phoebe Norman from Exeter University for putting together the survey and reviewing the results.
To coincide with the recent Rights Together network event on AI in the Rights Workflow, we recently put out a survey to ascertain the extent to which rights professionals were using LLMs (Large Language Models e.g. Chat GPT, Claude, Gemini, Co-Pilot etc.) in their rights work. The survey, whilst only a small sample size, provided some interesting feedback.
Key headline findings
Over half of the sample said that they were experimenting with LLM’s, whilst over a quarter of respondents were philosophically opposed to using any forms of AI in their work. The remainder were sceptical about AI but open to considering using LLMs.
The vast majority of respondents said that their workplace had policies on the use of LLMs with a variety of restrictions reported, from not being able to use LLMs at all through to restricting use of outputs generated by LLMs.
Interestingly, confidence in using LLMs was low, across the board, perhaps not surprising, given that around 70% of participants reported that they have had no training on the use of LLM’s within their role.
Overall usage of LLMs was low, with very few respondents using them more than monthly, and many not using them at all.
The most popular LLM according to the survey was ChatGPT (42.9%) and the most common task was for customer research (60%), followed by data analysis, with respondents noting they had also found it useful for writing up bookfair notes and preparing costings and itineraries for trips.
The main concerns with using LLM in rights work were:
- accuracy of outputs
- lack of rights specific tools
- data security
- confidentiality
Respondents felt that LLMs could potentially be helpful, but only if they could reliably and accurately guide decision making, interrogate contracts, or reduce admin tasks.
Given the very small sample size, we should be cautious about drawing conclusions from this data, but it does provide an interesting snapshot of perceptions and usage of large language models for rights work.