The profile of Artificial Intelligence is higher now than it has ever been. As with many innovations, the potential impact on consultation will be very important. The Consultation Institute takes upon itself the role of developing expertise and standards for all things consultation. The sorts of generative AI models developed by companies like OpenAI and Google and made available to the public look like they could transform many areas of consultation.

What is the best way to avoid problems when using AI?

While there has been a lot of talk about the existential threat posed by AI to the future of humanity, the risks with consultation are more limited. These prosaic risks relate mostly to Gunning Principle 4: conscientious consideration. This requires you to have taken people’s views into account when making a decision.

The reason why there is a risk with using AI is that if you let an AI analyse the consultation responses on its own, and the AI misses an argument that requires a response, your consultation will be unlawful. But even if the AI spots all the arguments, it may not understand the significance of each.

There is a broader concern about generative AI, which appears to have a fundamental inability to not make things up. There is an eye opening case from the US, where Robert Mata filed a submission in a personal injury case in which he cited a number of cases. There was, though a problem. None of these cases existed. The AI he used to generate his response ‘hallucinated’ them. Such an hallucination in a consultation report would be enormously problematic.

There is a relatively simple solution to all of this. I like to call it ‘collective intelligence’. It is where we use machine learning to supplement human activity. It is fine to use an AI to assist people doing a task. But you should never rely on the AI alone.

I discuss some of the specific use cases where AI might be able to deliver this ‘collective intelligence’ below.

What if I’m undertaking a consultation?

Many of those who will read this will be responsible for running consultation. Here, there are very many opportunities for the use of AI. In a practical sense, the models of “Generative AI”, which are creating so much attention offer the potential to add enormous power to the ability to analyse qualitative data, free text and even spoken word submissions.

There are a number of effective ways to use a generative AI to analyse your consultation.

1. Prompt engineering

Like any new technology, there will be a steep learning curve on how to use it most effectively. The key to better use of AI is ‘prompt engineering’ – the way in which you can most effectively use the prompt to generate the response that you want. The Consultation Institute will be including a section on this in our “Effective Digital First Consultation” training course.

2. Grouping themes

Generative AI is very effective at analysing large volumes of text and identifying key themes that are present in the text. It can do this rapidly and effectively. It can disambiguate different arguments, even where the conclusion is the same, and give you a feel for why people are objecting. AI could very effectively link back any statements in a consultation response in the metadata of a report to help the consultor show conscious consideration too.

3. Drafting documents

Generative AI is able to very effectively translate text into plainer forms, to make it more accessible, for instance lowering the reading age required for comprehension of a document. You could even use it to convert text to spoken word so it can be digested by people who prefer to absorb information in this way. There are great examples of the use of generative AI for translation into other languages, and this may be useful too, where equality assessments suggest the need to translate a document.

4. Producing rapid responses

Often consultors will want to produce a rapid response to a meeting, to synthesise a range of topics and provide a synopsis of the meeting or other event.

To emphasise the point from above, it would be very dangerous to do these things without thorough oversight from people. You should not allow something that has not been thoroughly reviewed ‘into the wild’.

5. Understanding which responses have been generated using AI

It will be helpful to understand whether AI has been used in the drafting of a response. For this reason, it might be useful to include a tick box, or free text field in which consultees are asked if they have used AI, and what AI tool(s) they have used.

What if I’m responding to a consultation?

One of the main concerns we at tCI have picked up speaking to people is that AI may make is more difficult for people to respond to consultations. While there will be barriers to entry, it may help many people who previously lacked the time or skill to hone long consultation responses.

We think you should be open that you have used AI in your response if that is what you have done, and you should check your response very clearly to make sure it has not hallucinated something that might make your response less credible.

What if I’m worried about the use of AI in understanding my response to a consultation?

We think that all public authorities should publish a statement in conjunction with each consultation specifying what purposes AI will be used for, and how that use will be overseen by humans. For instance, this might mean saying “we will use a generative AI (insert name) to help make the language in the consultation document more accessible, to assist with the analysis of consultation responses and to create images to illustrate the consultation documents and website.” This should include an option for consultees to have their contribution excluded from this analysis.

What the Consultation Institute is doing next?

We are actively considering the use of AI in consultations. As the context is changing so rapidly, this will require ongoing input from you as interested parties. If you have any observations please do get in touch.

Specifically, we are considering:

·   Creating Standards for the use of AI in consultation

·   Developing a training course on how to use AI to deliver better consultation

·   Developing resources for the use of AI in consultation

NB No intelligence artificial or otherwise was used in the writing of this piece.