Using ethical AI to understand audience behaviour and drive informed innovation

Photo Credit: Roy Jones

Andy Crouch explains how the team at Akumen is utilising “ethical AI” to better understand the emotions and behaviours of audiences through their own words.

It’s hard to ignore the rise of AI in 2023. From ChatGPT to Dall-E, artificial intelligence has gone mainstream. But the potential for AI stretches far beyond copywriting and image creation, with tools that have the potential to empower and inform anyone with a digital device and an internet connection.

In the museum space, companies like text analytics specialist, Akumen, is harnessing AI to provide museums and other cultural institutions with the ability to gather meaningful, actionable insights from their visitors and audiences.

Akumen is world-leading in textual data analysis, using software to identify emotions, behavioural drivers, and topics of concern to audiences. In partnership with Cornwall Museums Partnership (CMP), Akumen has developed an AI-driven software module and chatbot specifically to analyse visitor feedback to previously unachievable levels.

Known as Audience-I, the platform takes crowd wisdom and human-centric experience at scale and in real time to extrapolate insights. The algorithm for Audience-I has been created using a large volume of public domain feedback data – including reviews from Tripadvisor – to formulate a rule-based AI that’s continuously refined and improved to maintain robust moral standards.

As Andy Crouch, Director at Akumen, explains, “It’s important to distinguish Audience-I from tools like ChatGPT or Bard. The key point is that it isn’t generative and it isn’t a statistical model. What Audience-I does is distil human experience through an ethical AI solution that’s informed by manual human interpretation.

“Ethical rule-based Artificial Intelligence is becoming more common in all sectors where audiences, visitors, and stakeholder groups provide spoken or written feedback. The traditional approach for gathering feedback to improve, innovate, and justify funding has been limited to analysis of quantitative data. This typically includes standard metrics like visit frequency, age, gender and demographic. But ethical AI gives us the ability to do more than that.”

Audience-I gives organisations the capability to analyse the qualitative data, otherwise known as free or unstructured text, dialogue, or open question data. This is a marked improvement on historical techniques, which are unable to analyse even small datasets beyond generic sentiment analysis – often measured as good, bad or neutral.

Andy Crouch – Photo credit James Craxton

In practice, this enables cultural institutions to collate feedback and use it to drive improvements – be it in the layout of exhibits or even the queueing system at the café. Andy continues,

“The IP for Audience-I is really about turning big data into big but prioritised insights – by structuring unstructured text that we then analyse. Using those insights, it’s then possible to triage responses to find the information that organisations want to act upon. And, critically, these insights can then be put into the hands of people who can do something with it.”

“It is emotions and not sentiment that drives people’s decisions and behaviours. The ability of our AI-enabled software to identify and extract novel indicators in language, such as emotions, themes, and psychometrics, enables new levels of evidence-based insights to be available for organisational decision makers.

“The critical thing to note is that Audience-I is about insight democratisation – without making insights meaningful and actionable, it’s just noise. We know that museums and galleries gain lots of reviews, testimonials and insights from a range of different sources but, without a mechanism to interpret that and make it actionable, it can easily go to waste.

“Our platform is designed to be easy to implement, low friction and low hassle. But it’s powerful.”

What does ethical AI really mean? And why not use ChatGPT?

Asked to elaborate on what is unique about Akumen’s approach and the difference between Audience-I and large language models like ChatGPT, Andy explains,

“Fundamentally, ChatGPT and its rivals are black boxes. They are built using complex algorithms that draw on billions of parameter points from public domain data. While that approach can produce some very impressive results, it is problematic because we can’t open the box and take a look inside at how it’s working.

“Our AI is manually created and designed to be a completely transparent box. That means it’s possible to interrogate those inputs and parameter points when necessary. So, whether it’s identifying biases or updating outmoded concepts, we can manually unpack the box and nuance the rules.

“Our data is built from the ground up, with the ontology – the framework of understanding – completely customisable to the needs of the museum and its visitors.”

Small dataset. No problem

Rich, open data has value even in instances where the number of contributors is relatively small. It doesn’t face the same issues as traditional quantitative surveys. In those instances, a sample size of 10 people may be deemed insufficient at an institution welcoming 100,000 people through the doors each year. But thanks to Audience-I, this isn’t a problem. Andy explains,

“A traditional data scientist will tell you that if you don’t have 1.5 million data points we can’t rely on the data. That’s not a concern for us because the ontology is built in such a way that all rich content has extensive value. By “rich” I mean the volume of emotional hits, sentiment hits and theme hits contained within that data.

“In essence, the idea that one comment is merely anecdotal is turned on its head. We can turn every anecdote into a new stream of reliable, evidence-based, granular data.”

At Andy’s local science museum in Cornwall, Discovering 42, a version of Audience-AI has been installed that gives families the ability to not only type comments into the computer system but also give verbal feedback that can be captured and collated.

Audience-I has also recently been used at the Encounters film, animation and virtual reality festival in Bristol to gather feedback on the event using a simple QR code for accessing the survey.

In other use cases, Andy says that other museum clients are collecting their own survey data directly but that the AI solution can simply be used as a plug-in feature.

Combatting survey fatigue

There’s no doubt that completing surveys can feel like a chore to many people. From workplace forms to tourist questionnaires, gathering feedback can be a dull box ticking exercise.

Yet, one of the benefits of a small chat box format – able to capture both text and speech – is that people can voice their thoughts when and how they choose. And where better to offer this open-ended solution than in creative spaces where the typical audience may not want to have their contributions restricted.

Metrics to inform funding applications

A particular area of interest for Akumen is funding applications, where Audience-I can add value to a project or programme of events. Andy explains,

“Funding bodies obviously have metrics and criteria that are designed to reflect potential or proven return on investment. We can help museums to gather more accurate data over a wider range of categories, whether that’s understanding if an installation makes people feel more educated on sustainability; if an exhibition makes kids more passionate about pursuing science in the future; or if a museum visit simply helps visitors’ mental wellbeing. Those kinds of outputs.

“What is crucial to the arts and culture is generating an emotional response. And where we can help out is by capturing that and feeding it back in a practical way. We’re also conscious that AI shouldn’t be a thing that freaks people out.”

Akumen has been delivering this capability to partners such as the NHS for more than 5 years. And despite the hype around the leading names in generative AI currently, Andy is keen to point out that the ethical, transparent approach Akumen is promoting is important for a number of reasons:

“For any organisation looking to embrace this kind of technology, you need to be sure that the engine driving those recommendations and insights if fit for purpose and bulletproof. This is particularly important in one of our other sectors – healthcare – where the risks associated with errors can be significant. So, being able to inform the AI models is essential to mitigating those risks.

“Museum professionals can think of this as human created and human curated AI – a flexible solution that allows the feedback loop to meet their values and strategy.”

For more information about Audience-I, please contact Andy Crouch via email acrouch@akumen.co.uk or LinkedIn