Artificial Intelligence: Responsible AI and the path to long-term growth

This draft regulation would no longer permit the use of unacceptable-risk systems within the EU; subject high-risk systems to various requirements, including human oversight, transparency, cybersecurity, risk management, data quality, monitoring, and mandatory reporting; and it would require transparency within the use of limited and minimal-risk systems, so individuals are aware that they are interacting with AI technologies.

The regulation would not only apply to companies and AI providers within the EU but would have extraterritorial jurisdiction over any AI system providing output within the EU. Companies in breach could also face penalties of up to €30m or 6% of their global revenue – whichever is higher – making fines more substantial than those imposed by the GDPR.

As governments and regulators across the globe increasingly consider the governance and regulation of AI, we believe the risks could become very material for companies from a legal and financial perspective.

Responsible approaches to AI

Beyond regulation, the European Commission has produced Ethics Guidelines for Trustworthy AI¹¹  which can serve as a useful framework for companies to implement Responsible AI practices. These guidelines seek to support AI that is lawful, ethical and robust in the face of their potential harm.

They are built around four ethical principles which bring concrete requirements for companies:

• Respect for human autonomy: Ensuring respect for the freedom and autonomy of human beings; the allocation of functions between humans and AI systems follows human-centric design principles and leaves meaningful opportunity for human choice; human oversight over work processes in AI systems is maintained
• Prevention of harm: AI systems should neither cause nor exacerbate harm or otherwise adversely affect human beings; attention must also be paid to situations where AI systems could cause or exacerbate adverse impacts due to asymmetries of power or information; human dignity as well as mental and physical integrity is protected
• Fairness: Ensuring equal and just distribution of benefits and costs, and that individuals and groups are free from unfair bias, discrimination, and stigmatisation; the use of AI systems does not lead to people being deceived or unjustifiably impaired in their freedom of choice; the ability to contest and seek effective redress against decisions made by AI systems and by the humans operating them is preserved
• Explicability: Processes should be transparent, the capabilities and purpose of AI systems openly communicated, and decisions explainable to those directly and indirectly affected; “black box” algorithms are avoided

At AXA IM, our approach to Responsible AI draws on a similar structure and requires four clear components. We believe corporate AI systems should be fair, ethical, and transparent while remaining reliant on human oversight – both from a technical and high-level governance perspective.

We think that companies will be able to mitigate AI risks and deliver sustainable value creation if they practice Responsible AI policies – benefitting from the deployment of AI systems whilst building trust and improving customer loyalty.

The bottom line is that Responsible AI is not only about avoiding risks but ensuring that these technologies can be harnessed to the advantage of people and businesses alike.

Assessing how Responsible AI is practiced

We have looked at a range of studies and benchmark/assessment tools which deal with how companies address and practice Responsible AI – and they ultimately reveal a large gap in progress.

The 2021 McKinsey survey,¹² for instance, highlights what we see as a lack of awareness amongst the organisations they consulted towards the various risks associated with AI. We think the graphic below is a good illustration of how companies may be underestimating the risks.

Similarly, the 2022 Big Tech Scorecard by Ranking Digital Rights (RDR),^{13  14} an independent research programme, graded tech companies relatively poorly in terms of algorithmic transparency – based on indicators which follow a similar framework to those discussed earlier, including elements relating to human rights policy in the development of algorithmic systems, risk assessments, transparency and privacy.

None of the major tech companies assessed achieved a score above 22%, according to RDR’s methodology, with some of the biggest global firms near the bottom of the ranking.

The Digital Inclusion Benchmark from the World Benchmarking Alliance,¹⁵ meanwhile, ranked companies on “inclusive and ethical research and development” and sought to capture “the degree to which company practices reflect the participation of diverse groups and consideration of ethics in product development”. Our assessment of benchmark scores for a range of companies found underwhelming performance, with just 3% of the 77 we studied gaining a top score.

We think these results help demonstrate the need for responsible investors, such as AXA IM, to engage companies in discussions over Responsible AI and indicate how we might put that into practice. As part of our responsible technology research and engagement programme, we focus on firms with significant investment plans in AI development.

Our key engagement recommendations urge companies to:

• Align with the key components of AXA IM’s approach to Responsible AI – fair, ethical, transparent, human oversight
• Increase disclosure and transparency around AI system development
• Ensure oversight of Responsible AI by their board and senior executives

We are convinced that mitigating the risks associated with AI systems – and addressing regulatory considerations – are closely related to the ability of companies to deliver long-term value creation with these technologies. We also believe that AI systems can better provide long-term and sustainable opportunities when Responsible AI is practiced.