By Prasanthi Vasanthakumar, Institute of Corporate Directors

Today, artificial intelligence (AI) can speak, write, play games and read minds, making what was once impossible, possible. A study by McKinsey Global Institute notes that early adopters of AI with strong digital capabilities and proactive strategies will enjoy higher profit margins and outperform the competition.
With the ability to decode data far quicker than is humanly possible, AI can help companies solve problems and make better decisions. But it also has a sinister side, where cyberattacks, discrimination and data privacy can magnify existing ethical issues and create new problems.

To help boards navigate the ethical challenges and risks of this extraordinary technology, we spoke to Niraj Bhargava, ICD.D, CEO and Co-founder, NuEnergy.ai and Nicole Janssen, Co-founder and Co-CEO, AltaML.
For Bhargava, who is the National Academic Director and Lead Instructor of the ICD’s new course, Board Oversight of AI, there’s no question that supporting ethics with AI fits squarely within the duties of a board member. “Our Duty of Care naturally includes ethics, and with AI, its intelligence and abilities to affect decision-making must also be carefully considered,” he says. “One of the biggest challenges facing directors is recognizing that all organizations have AI, not just those that are technologically advanced.”

The risks of AI

Understanding how AI is used and the resulting ethical considerations should be a key priority for boards. EY has identified six key areas of AI risk,1 of which at least three have ethical implications – algorithmic bias, cyber-risk and reputational risks.

Discriminatory data

Due to their underlying hand-made algorithms, AI applications are still prone to human error and bias. For example, skewed data can unfairly exclude certain groups in medicine. “If the images used during a diagnosis model do not include black or brown skin, then the effectiveness of diagnosing those diseases for individuals with a different skin colour would be far less effective,” says Janssen.

Even when AI is explicitly used to make ethical choices, things can go awry. An organization may turn to AI for non-discriminatory hiring, but if it programs the application with performance criteria based on past top performers, any racial, gender or ethnic biases in previous recruiting practices will seep in.

Unknown logic

AI programs can be a “black box,” where the logic behind decisions is not transparent. Sometimes, even programmers can be in the dark. This unknowability poses significant risks to companies when AI leads to an outcome that is undesirable or inexplicable.

“These types of decisions can raise ethical concerns and create fear,” says Janssen. She notes that some solutions in this area focus on transparency around process, including where the data originated and which parts of it are used in the model. “It’s important to work on explainability, so that end users who are impacted by the AI model can fully understand why decisions are made,” she says.

Hacker haven

AI systems can also be an easy target for cybercriminals who wish to steal personal data or confidential information about an organization. If the program’s logic is a mystery, the company may not know the system has been infiltrated and is malfunctioning. Risk can vary, from something that can be relatively innocuous such as implanted bias to grave danger like a hijacked robot.2

Reputational damage

The fallout from a cyberattack is only one reputational risk posed by AI. As this technology uses enormous amounts of sensitive data to make important decisions about individuals in wide-ranging areas such as credit, education, employment and health care, biases, errors, compromised privacy and unethical use can damage the organization’s reputation.

Ethical oversight

Both Bhargava and Janssen emphasize the importance of proper and persistent oversight to ensure ethical considerations are etched into the AI process.

Leaders need to understand that AI models have to be continually improved. “You can't create a model, continue working with it and expect it to just operate appropriately because you may be adding biased data to it,” says Janssen. “Continual improvement and monitoring are essential.”

“AI is not like most software that may only require a one-time purchase or build consideration,” adds Bhargava. “It is trained on data that needs to be governed, as AI can drift to unintended consequences after deployment. Because AI is dynamic, it needs dynamic oversight.”

Organizations should consider proactively setting up technical and organizational guardrails to guide, monitor and assess how AI is used by all stakeholders. For example, a technical guardrail is the use of explainable AI, where the logic behind decision-making is transparent. Many tools for managing risk exist, such as an AI ethics code of conduct that can help ensure compliance with corporate values, a whistleblowing mechanism that allows employees to raise concerns and a process by which the board is informed about AI-related risks and issues. 3

AI isn’t going away

While COVID-19 has brought many economies to a grinding halt, it has fast-tracked the pursuit of artificial intelligence. Janssen points to a McKinsey survey that finds the pandemic has accelerated AI digitization by three to seven years, depending on the industry.

“However, urgency does not replace good governance and the requirement to ask the right questions – it emphasizes it,” says Bhargava. “This is true for fast-emerging AI. Today, we are governing machines, as well as people, strategies, financials, and societal needs, which are all impacted by AI.“

Indeed, governance of AI is practical and necessary across organizations, but not just for advanced AI. Machine learning and other AI technologies are already commonplace, and require strong oversight.

Because AI is ubiquitous, boards will continue to wrestle with existing and emerging ethical issues. From determining their role in addressing the greater impact of AI on society as a whole, including displaced workers, to building and sustaining user trust in the technology, directors will have a lot to process.

Sources and Further Reading

Compiled by Heather Wilson, Director, Research Services
1 EY. Why AI is both a risk and a way to manage risk (2018)
2 International Corporate Governance Network.  Artificial intelligence and board effectiveness (2020)
3 Deloitte. AI ethics: A business imperative for boards and C-suites (2019)
Forbes.  Why board directors and CEOs must become AI literate to lead forward (2020)
Harvard Business Review.  What boards need to know about AI (2019)
Slate.  Why not appoint an algorithm to your corporate board? (2019)
World Economic Forum.  Empowering AI leadership: an oversight toolkit for boards of directors (2020)

Originally published in the Director Lens Communique by the Institute of Corporate Directors.

Home

The board’s role in ethical AI