AI violates privacy. AI-generated output can not be explained. AI is biased.
This is all true, and is happening today, and there’s a risk of these issues accelerating as AI adoption grows. Before the lawsuits start flowing and government regulators start cracking down, organizations using AI need to become more proactive and formulate actionable AI ethics policies.
But an effective AI ethics policy requires more than some feel-good statements. It requires actions, built into an AI ethics-aware culture. “An AI ethics statement is a nice start. It’s also the tip of the iceberg, ”says Reid Blackman, who explores AI ethics in his upcoming book, Ethical Machines: Your Concise Guide to Totally Unbiased, Transparent and Respectful AI (Harvard Business Review Press). “Even those who spend a lot of time talking about and even working in the field of AI ethics have a lot of difficulty understanding what to be done. That’s because they’re trying to build structure around something they still find squishy, fuzzy, and subjective. “
Blackman, CEO of Virtue, seeks to turn something as “squishy” as AI ethics into something concrete. He provides guidelines for instilling actionable ethics into AI systems and processes.
Bring clarity to AI standards. “Between ‘never do this’ and ‘always do that’ there is a lot of ‘maybe do this.'” Blackman cautions. “Think about the tough cases before they happen.” The body of these conclusions form instances of “ethical case law” that can lay the foundation for AI ethics standards. “Creating your organization’s AI ethical case law is an extremely powerful tool in articulating the organization’s AI ethical standards and communicating them to relevant stakeholders.”
Increase awareness among everyone in the organization. Make data scientists, engineers and product owners aware of AI ethics issues., Blackman urges. “Everyone in the organization who might develop, procure, or deploy AI needs to be aware. including HR, marketing, strategy and so on. This will not only require training, but also developing a culture in which that training gets uptake. ”
Bake AI ethics into team culture. Teams such as product development “need not only knowledge about the issues and how, in principle, they can address them, but also concrete tools and processes for diligently carrying out ethical risk identification and mitigation,” he states. “
Make sure there are AI experts as part of an AI ethics committee. “While standard processes and practices and tools are helpful, they are only a first line of defense” in a complex undertaking such as AI, Blackman says. To seriously address the ethical, reputational, regulatory, and legal risks, we need experts in the room. It would be unwise and unfair to charge data scientists, engineers, and product developers with the primary responsibility of identifying and mitigating ethical risks of products. ”
Introduce accountability. There needs to be a sense of responsibility for tools and processes introduced in AI. “Staff needs to be held accountable for using these tools and complying with processes and on pain of penalties ranging from lack of bonuses, to inability to be promoted, to firing,” Blackman points out. “Relatedly, we need to financially incentivize taking these issues seriously.”
Measure everything. Metrics, in the form of key performance indicators that tie AI ethics to business goals, is a key form of metrics. “We need to track both the extent to which the organization is adopting new standards and the extent to which meeting those standards identifies and mitigates the risks those standards are aimed at addressing,” Blackman says. Determining quality KPIs for your actual ethical performance needs to be the product of your AI ethics statement and your AI ethical case law.
Gain executive sponsorship. AI ethics needs leadership, in the form of a person “Responsible for overseeing the creation, rollout, and maintenance of an AI risk program,” says Blackman. “Excitement about AI ethics from junior people is a wonderful thing, but there is no such thing as a viable and robust AI ethical risk program without leadership and ownership from the top.”