In today's competitive business environment, artificial intelligence, including generative AI, has become integral for many organizations, helping them streamline operations and, in some cases, improve efficiency.
But as outlined in this year’s Global Risks Report — prepared by the World Economic Forum with the support of Marsh McLennan and others — this technology does not come without risks, including bias and discrimination risks. And as AI models become more globally adopted and their applications expand, it is essential for companies and business leaders to take proactive steps to address the risks of bias and discrimination.
Understanding the potential for AI bias
No matter how sophisticated generative AI models are becoming, it is important to remember that they operate on algorithms and training data created by humans, which means that any inherent human biases can be transferred to AI. Any systematic favoritism or discrimination reflected in the data used by AI could, in turn, result in biased outcomes, which can have significant consequences for both businesses and society as a whole.
For example, biased AI models could perpetuate existing societal prejudices and lead to unfair treatment, discrimination, and exclusion. This can harm not only people and communities, but also the bottom line of companies, as biased AI tools used in recruitment, employee development, performance reviews, and more could potentially lead to inefficiencies, missed opportunities, or even litigation.
Addressing AI bias and promoting equitable decision-making
To promote equitable decision-making in the digital age, organizations and business leaders should consider a number of actions, including:
- Establish governing principles and foster diverse, multidisciplinary AI oversight teams: It is critical for organizations to design and implement principles that should be followed — whether by humans or AI. It is also essential to develop oversight mechanisms to ensure these principles are adhered to. For example, establishing such principles can help colleagues using AI as a tool during the recruitment process to assess whether they are promoting equitable decision-making. Assembling an AI governance team with individuals from different backgrounds, experiences, and perspectives can help identify and address any AI biases. A diverse, multidisciplinary, cross-enterprise group of AI users and stakeholders also can help identify biases that may have been overlooked during model development or that start manifesting when applied to new use cases.
- Conduct regular audits and testing: Ongoing monitoring and feedback loops can support continuous improvement of AI systems. When possible, carefully assess the data used in the initial training of AI models to identify if it is unrepresentative or discriminatory. When this is not feasible, test the outputs to better understand how the model is functioning and assess any potential bias. Ongoing analysis of the outcomes of AI algorithms can help identify if they disproportionately impact certain groups or exhibit biased behavior. Regular audits and testing can provide valuable insights into the performance of generative AI systems and help identify areas for improvement. It is also important to conduct due diligence on AI model vendors and developers, asking questions on data selection and processes.
- Promote collaboration and industry standards: As the prevalence of generative AI use intensifies, it is important for organizations and industry stakeholders to provide feedback and collaborate in establishing industry-wide standards for both the use of AI and especially to address biases in the technology. By working together, companies can promote consistent and ethical practices across an industry.
Addressing bias starts from the development process, underscoring the importance of establishing ethical guidelines and frameworks. Transparent AI systems can provide a clearer understanding of the factors considered in decision-making, enabling better scrutiny and more effective identification of biases. Developing and implementing ethical guidelines and frameworks can address multiple issues, such as fairness, accountability, and transparency. These guidelines can serve as a roadmap for addressing biases and provide a system for resolving ethical dilemmas that may arise in AI development and deployment. Further, by embracing transparency, companies can build trust with stakeholders and improve accountability in their AI systems.
The role of insurance
Organizations and businesses should also review their insurance coverage with their broker or insurance advisor in case they are accused of AI-related bias. Employment practices liability insurance will typically cover AI-related discrimination claims. However, there may be exclusions related to the Biometric Information Privacy Act (BIPA) or other privacy related laws, which could have implications for an AI-related discrimination claim, such as the use of facial recognition tools.
Taking action against AI bias
As AI technology progresses, it is vital for organizations and leaders to actively address biases in AI systems and generative AI models. By being aware of and understanding the root causes, detecting and measuring outcomes, and implementing mitigation strategies, companies can work towards AI systems that are fairer, more transparent, and more accountable. This not only can help companies better harness the full potential of AI, but also advance inclusivity and equity in a changing world.