Risk and Trust Perceptions of the Public of Artificial Intelligence Applications

🔬 Research summary by Connor Wright, our Partnerships Manager.

[Original paper by Keeley Crockett, Matt Garratt, Annabel Latham, Edwin Colyer, Sean Goltz]

Overview: Does the general public trust AI more than those studying a higher education programme in computer science? The report aims to answer this very question, emphasising the importance of civic competence in AI.

Introduction

Is the opinion of the general public on AI different to those studying computer science in higher education? With a survey titled “You, me and “AI”: What’s the risk in giving AI more control?” the paper aims to compare responses on the level of trust and risk that people of the general public and students studying computer science in higher education give. What is for sure is that civic competency in AI is crucial in creating representative technology, something we hold dear to our hearts here at MAIEI.

Key Insights

Civic competency in AI

One of the main slogans of this paper, my TEDxYouth talk and what we do at MAIEI is the importance of civic competency in the AI field. By improving public understanding of AI, we better equip them to fight any misinformation on the subject. One way to do this is to develop online courses, following in the footsteps of the University of Helsinki. By allowing non-experts to become involved in the debate, we enrich and make more representative the AI space.

Nevertheless, the paper points out how some may feel intimidated by courses offered by universities, for they don’t feel they have the right qualifications. Hence, a future focus can be in creating courses specifically designed for the common person.

One of my core beliefs is that everyone can bring something to the AI table, no matter the level of expertise. Such value is clearly demonstrated in the data collated in the paper’s surveys.

The results

One of the main driving forces behind the survey is how previous studies conducted on the general public show varying degrees of knowledge about AI, but they all lack a robust description of the general public. Hence, the paper takes the general public to be those who have no specific knowledge in AI.

The groups of participants were split into Group 1 (the general public) and Group 2 (students of a higher education computer science programme). The groups were then asked questions on 3 different themes: trust, risk and questions on a scale of 0-10. A bird’s eye view of the results are as follows:

Trust

The groups were found to agree on questions such as not trusting an automated message from their boss, but differed on questions as to whether to trust a driverless car that had passed a “digital MOT” (p.g. 4).
In this case, university students were more trusting of the AI involved.

Risk

The students always associated the same if not more risk to different AI applications, especially in terms of following instructions from a recognisable digitised voice.

On a scale from 0-10

There was general parity between the two groups on statements such as “I believe the minority of AI systems are biased”. The only difference came in how students placed less emphasis on AI system decisions being explainable.

Between the lines

While the general public is defined as being without deep knowledge in the field, it is crucial that they are deemed to be a key stakeholder. In this way, their interactions with AI systems must be considered when evaluating an AI model’s performance. As the paper rightly mentions, risk can occur at different points of the AI lifecycle, making system monitoring a vital aspect of a successful AI system. I hold that we cannot view AI systems as able to generalise over the whole population, meaning the practice is critical in ensuring that the system accurately tends to what it is designed to do and the problems this could bring.