New research could empower people without AI expertise to help create trustworthy AI applications

In recent years, the rapid development of artificial intelligence (AI) has led to concerns about the fairness, transparency, and trustworthiness of automated decision-making systems. These systems are increasingly being used in critical areas such as healthcare, finance, and law enforcement, yet the majority of AI development and evaluation is carried out by experts with specialized knowledge. A new study from researchers at UK universities challenges this status quo by proposing that involving people without AI expertise in the development and evaluation of AI applications could lead to better, fairer, and more trustworthy outcomes.

The research, which will be presented at a major international computing conference, focuses on a concept known as "participatory AI auditing." This approach involves members of the public in the evaluation of real-world AI applications, allowing them to assess the potential impacts of these systems on society. The study's authors argue that by incorporating diverse perspectives and experiences, AI applications can be improved in ways that would not be apparent to developers alone.

To test this idea, the researchers enlisted a group of volunteers from the public to evaluate two real-world AI applications. The volunteers were asked to consider how these systems might affect different demographic groups and to identify potential biases or unfair treatment. Their insights were then compared to those of AI experts, revealing that the public's input often highlighted issues that experts had overlooked.

One of the key findings of the study is that people without AI expertise are often better equipped to detect biases and unfairness in AI systems. This is because they bring a unique vantage point, having experienced the effects of such systems firsthand. For example, a volunteer might notice that an AI system used for credit scoring appears to disproportionately disadvantage a particular ethnic group, an issue that might not be immediately apparent to the developers.

The researchers suggest that participatory AI auditing could become a standard practice in the development of AI applications. By involving the public in the evaluation process, developers can gain valuable insights into the real-world impacts of their systems, leading to more equitable and trustworthy outcomes. This approach also has the potential to increase public trust in AI, as it demonstrates that the needs and concerns of ordinary people are being taken seriously.

However, the study also acknowledges the challenges associated with participatory AI auditing. One concern is that the public may not have the necessary understanding of AI to evaluate its performance accurately. To address this, the researchers propose that educational resources be developed to help people understand the basics of AI and its potential impacts. Additionally, the study emphasizes the importance of ensuring that the public's input is taken seriously and acted upon by AI developers and policymakers.

In conclusion, the research from UK universities presents a compelling case for involving people without AI expertise in the development and evaluation of AI applications. By leveraging the diverse perspectives and experiences of the public, AI systems can be improved to be more fair, transparent, and trustworthy. As AI continues to play an increasingly important role in society, participatory AI auditing offers a promising approach to ensuring that these systems benefit everyone and do not perpetuate existing inequalities. The study's findings are likely to spark further debate and discussion about the role of the public in shaping the future of AI, and could pave the way for more inclusive and equitable AI development practices.