Official Warns AI Could Heighten Bias

A senior European official has warned that using Artificial Intelligence (AI) to decide whether to offer a mortgage could lead to discrimination. Margrethe Vestager said this was a more likely risk than dramatic predictions of AI leading to human distinction.

Vestager is the European Commissioner for Competition. That's a key position as not only is she responsible for competition and antitrust issues across the 27 countries in the European Union, but her work affects businesses around the world that operate in Europe, including tech companies.

Talking to the BBC she warned that AI can increase discrimination, even though it's promoted as an objective technology. (Source: bbc.co.uk)

AI Plays By Own Rules

The problem is that AI is different to ordinary computerized systems that crunch data and give a verdict, for example whether to offer somebody a mortgage at a particular rate or what premium an insurance customer should pay.

These systems work "objectively" by following a fixed set of rules developed by humans. In this case, the computing is simply there to speed up the process and avoid errors. AI technology will more commonly look at data and figure out its own set of rules. For example, it might look at characteristics of previous mortgage holders alongside whether or not they made repayments on time.

Vestager argues that the problem with such technologies is that the data itself may include discrimination or bias. For example, if people with particular characteristics may have been charged higher mortgage rates through bias. Those higher rates may have led to more foreclosures among people with those characteristics, effectively turning the bias into a self-fulfilling prophecy.

One of the key issues highlighted by Vestager is the reliance on data that may already contain biases or discriminatory patterns. AI algorithms learn from historical data, and if that data reflects societal biases or discriminatory practices, there is a risk that the AI systems will perpetuate and amplify those biases in their decision-making processes. This can have significant consequences, particularly in areas such as mortgage lending, where access to housing and financial opportunities can have far-reaching implications for individuals and communities.

Regulation In The Cards

To address these concerns, policymakers and technologists need to prioritize fairness, accountability, and transparency in AI systems. This involves careful examination of the data used to train AI algorithms, identification of potential biases, and mitigation strategies to ensure that discriminatory patterns are not reinforced. It also requires ongoing monitoring and evaluation of AI systems to identify and rectify any unintended discriminatory outcomes.

Regulation can play a crucial role in establishing guidelines and standards for the responsible development and deployment of AI. By implementing clear rules and safeguards, policymakers can help mitigate the risk of discrimination and ensure that AI systems are aligned with societal values and legal frameworks. However, striking the right balance between regulation and innovation remains a complex challenge, as overly burdensome regulations could stifle technological advancement and limit the potential benefits of AI.

Collaboration between policymakers, industry stakeholders, researchers, and advocacy groups is essential to foster a multidisciplinary approach to AI governance. This collaboration can help identify and address potential biases, develop best practices, and establish accountability mechanisms that promote fairness and transparency. Furthermore, efforts should be made to increase diversity and inclusivity in AI development and deployment teams to reduce the risk of unintentional biases and foster a broader perspective in system design.

Politicians in Europe recently backed legislation on the development of AI, though some argue that regulation of AI isn't even possible, let alone desirable. (Source: washingtonpost.com)

The current proposals would mean different uses of AI could be regulated differently, with potential harm to the public the main focus.

What's Your Opinion?

Do you share the fears about AI and discrimination? Is regulating AI development a good idea? Are there particular uses of AI where regulation is more important?