By Lofred Madzou
“Quality is never an accident. It is always the result of intelligent effort” – John Ruskin
The adoption of artificial intelligence (AI) is gathering pace. Although highest in the tech, telecoms, financial services, and manufacturing sectors. And with a significant level of adoption in emerging markets, the trend has seen an increase in almost every industry, encompassing a range of business sectors from production, through marketing and sales to HR and risk management.
Alongside this trend, companies are broadening their focus to include stakeholders beyond their shareholders. This can be attributed to a variety of factors. The covid-19 pandemic has shone a critical light on longstanding inequalities in the area of employment opportunities, while the global movements against climate change crisis and in favor of social justice have exerted pressure on companies to create long-term, sustainable value with benefits extending not only to customers and employees but also to the wider community and to citizens at large.
Environmental, Social, and Corporate Governance (ESG) pressure on companies is increasing. In recent years, investors have taken to measuring the sustainability impact of their investments using ESG performance indicators, with ESG funds attracting nearly USD120bn in flows in 2021.
The role of AI governance in an ESG proposition
A strong and credible ESG proposition is now a must for companies. With AI having a growing impact on business and society, it is clear that such a proposition needs to include effective AI governance.
Although recent years have seen AI progress in leaps and bounds, challenges of governance still remain. Deploying AI in a rapidly changing environment such as a global pandemic can lead to significant drops in the performance of AI models, since these evolve with data and use. The processes of debugging and maintaining these models at times of underperformance is more complicated than for classic software.
Utilities are one sector in which effective AI governance needs to play a key role when considering their ESG proposition. Influenced by environmental concerns, these companies are increasingly deploying AI solutions for demand forecasting and power grid optimization. A major goal is to reduce electric power greenhouse gas emissions, which account for almost 25 percent of total greenhouse gas emissions worldwide. However, in order for this to be achieved, the AI solutions deployed need to perform as expected.
Another example is the CEO Action for Diversity & Inclusion (D&I) Pledge. The Pledge has been signed by 2000 CEOs, including those of most Fortune 500 companies, and represents the largest CEO-driven business commitment to advance diversity and inclusion within the workplace. To deliver on their commitments, these CEOs need to ensure that AI solutions used to support their workforce decisions are fit for purpose and continuously monitored in order that processes such as recruitment, talent management and employee retention are efficient and effective.
Indeed, recruitment is an area in which it has been observed that, without proper oversight, AI can replicate human bias. In 2015, a study revealed that Google was less likely to show women adverts for high-paid jobs. In 2018 it was reported that Amazon removed an internal AI recruiting tool that showed a bias against female candidates. These episodes have fueled concerns over AI bias in employment, with the NYC Council passing legislation requiring vendors of AI-powered hiring tools to obtain annual third-party “bias audits”. Similarly, the EU AI Act has identified this area as high-risk and requiring procedures of quality management and conformity assessment.
Why AI Quality frameworks are needed
Sound AI Quality frameworks are the solution to making a company more sustainable while maximizing the benefits of AI.
The term “AI Quality” refers to the set of observable attributes of an AI system allowing the assessment over time of the system’s real-world success—that is to say, the value and risk from the AI system to both the organization and the broader society.
There are four key categories at the core of an AI Quality framework:
- The first consideration is model performance. Sound assessment should include considerations about model stability, conceptual soundness, and robustness as well as simply assessing AI systems based on their accuracy on benchmark datasets.
- Second, data quality is key. High-quality data is essential to building trustworthy AI models, but is often underestimated is the effort needed to collect, clean, and process this data can often be underestimated. Better models can also be built by checking for missing data and data representativeness.
- Third, operational compatibility must be considered. AI models are part of larger business and organizational structures and are never used in isolation. Their integration needs to be facilitated through documentation, model function, and collaborative capabilities in order to drive adoption.
- The final, and equally important consideration is societal impact. It should be remembered that AI systems are value-laden. Senior executives need to ensure that their behavior reflects the values of their company, particularly their ESG commitments. This is why observable attributes of societal value and risk are needed, such as transparency, fairness, privacy, and security.
Looking to the future
Companies may consider that the potential adverse impact of AI use on the strength of their ESG proposition is too big a risk. The alternative is to limit their AI use. However, I believe this to be self-defeating in light of the potential of AI for business growth globally and across all sectors.
As the pressure on companies to become more sustainable whilst also ensuring that their AI solutions are robust and trustworthy is increasing, so a variety of processes and tools are becoming available to help companies to design and deploy high-quality AI solutions. AI Quality frameworks offer a way of measuring the real-world impact of deployed AI solutions, which can be extremely beneficial in mitigating ESG risk in AI systems. It is recommended that companies use these tools now to maximize the benefits of AI whilst making themselves more sustainable and thus more attractive to investors.
About the author
Lofred Madzou, Director of Strategy and Business Development at TruEra, is a leading expert in Responsible AI and AI governance and has spent most of his career driving responsible AI in government and corporate settings. In his role, he works with organizations to strengthen their AI governance, prepare for regulatory requirements and emerging guidelines, and establish processes that allow them to use AI in more effective and responsible ways.