Public attitudes to the use of AI in DfT consultations and correspondence

Background and summary

Background and summary of methodology The Department for Transport (DfT) is exploring how it can apply artificial intelligence (AI) and machine learning (ML) to improve efficiency and quality across its work. In particular, DfT has been exploring how generative AI could be used in two different scenarios: 1) the analysis of consultation responses and 2) supporting the drafting of responses to external correspondence.

Consequently, DfT commissioned Thinks Insight & Strategy to conduct research with the
public to explore:

1.1 Public knowledge around AI and its various applications.
1.2 Sentiments around the application of AI with regards to the specific use cases.
1.3 Public opinion regarding the benefits and risks involved with the application of AI.
1.4 The best ways to communicate with the public about these use cases.

To achieve these aims Thinks Insight & Strategy undertook a multi-stage research approach including: three deliberative workshops in London, Newport and Glasgow (each with 24 members of the public), 10 in depth interviews with engaged citizens and nine in depth interviews with DfT stakeholders.

Key findings

This research revealed eight key findings:

1.1 Despite low confidence in their knowledge of AI, participants believed that AI will be
an important technology of the future. They therefore believe it is inevitable and appropriate that the UK Government (and consequently DfT) will utilise it for a variety of tasks.

1.2 Participants spontaneously believed that the key benefits of AI use are speed and efficiency and the key risks associated with its use are concerns around job losses, humans becoming overly dependent upon it and a loss of a ‘human touch’.

1.3 Both use cases (consultations and correspondence) were low salience activities for the general public. Whilst these participants were able to articulate views about each use case, in the context of low salience it is unlikely that either proposed use case will elicit strong reactions from the general public outside of a research context.

1.4 Views of the benefits of AI remained consistent when discussing both use cases. These benefits were most compelling when focused on how they will impact the public rather than DfT (e.g. a faster response to correspondence is more compelling than the task taking up less of DfT’s time).

1.5 Views of the most important risks shifted somewhat when discussing specific use cases with participants focusing much more on quality and accuracy of outputs.  There was a consensus that increased speed was only valuable if AI was producing high quality outputs.

1.6 When they were first shown the two use cases, participants tended to be much more comfortable with AI being used internally at the Department to analyse consultation responses rather than for being used to draft external correspondence. This was driven primarily by a concern about AI’s ability to draft empathetic responses to more emotional messages.

1.7 DfT’s proposed mitigations addressed some of the participants’ concerns, meaning that most ended the sessions feeling comfortable with AI being used for consultations. However, some participants wanted more reassurance on the topic of correspondence including whether humans would properly quality check AI outputs.

1.8 In line with their responses to the use cases, participants’ guiding principles for DfT use of AI focus on quality, accountability and transparency.