What does a CEO look like?
What comes to mind when you think of a secretary?
Researchers posed these questions to an AI program, and the images it produced equated CEOs with white men, and secretaries with women. That’s discriminatory.
AI-ENABLED DISCRIMINATION
And this wasn’t an isolated incident. Stories abound of stereotypes in large language models (LLMs) in education, technology and other fields. LLMs pose significant hurdles to algorithmic fairness, with recent versions still exhibiting biases and perpetuating stereotypes against women and girls, UNESCO found. Meanwhile, laws that were set up in New York City to fight discrimination in hiring are mostly being ignored (subscription required).
No wonder many of us in the C-suite are having such a hard time trusting AI, even as the technology proves itself as an incredibly useful business tool for raising productivity, efficiency and revenue. This skittishness strikes to the heart of the fear of AI: essentially, that AI is going to facilitate discrimination. And even if it isn’t designed to do that, it will, inherently, because humans are flawed, and humans are the ones programming it.
CRMS AND ALGORITHMIC REDLINING
There’s a saying in Human Resources: A Candidate Relationship Management (CRM) tool is only as good as the data you feed it. The same holds true for an AI system, particularly an AI-enabled CRM system.
As the name implies, CRMs help companies engage and nurture talent and maintain a healthy balance between headcount and vacancies. The system leverages the information provided to it from various departments. That means that AI systems have the potential to discriminate if fed the wrong data.
When that happens, we need to call it what it is: algorithmic redlining. Because it is people, not technology, that must be the last line of defense in deciding between right and wrong. Business leaders must recognize algorithmic redlining for what it is, and take the following steps to prevent it.
RADICAL TRANSPARENCY AND ACCOUNTABILITY
We need to shed some light on how AI CRM systems make their recommendations. Companies should be clear and open about their AI’s output process. This means documenting every step and making this information publicly accessible. Once this becomes standard, companies will have no choice but to address any biases upfront instead of ignoring them.
ETHICAL AUDITS BY EXTERNAL WATCHDOGS
And internal audits alone aren’t enough. Regular external audits are also needed, conducted by independent third parties with the credibility and expertise to objectively examine AI systems and help companies avoid any biases or unethical practices.
External oversight ensures that companies can’t hide behind complex algorithms and also demonstrates the ethical lengths a company will go to to ensure transparency and fairness while protecting valuable data and creating the best possible outcomes.
AI ETHICS BOARDS WITH TEETH
Many companies will soon have AI ethics boards (if they don’t already), but in my experience, they often lack real power. It’s time to change that.
These boards should have the authority to veto AI deployments that don’t meet ethical standards and require action for non-compliance. This will ensure ethical considerations are a fundamental part of AI development and deployment.
Staff the panel with a mix of people who are well-versed in the technologies. One company that stands out among the rest is ThermoFisher Scientific (full disclosure: a client of my company). Its AI Leadership Committee is headed by experts in various disciplines, and there’s an entire section on the company’s approach to AI in its Corporate Social Responsibility report.
PUBLIC ACCOUNTABILITY AND LEGAL LIABILITY
Companies should be prepared to face public scrutiny and legal consequences for their AI systems’ actions. Many are not.
This means being ready to answer tough questions from regulators, the media, and the public about how your AI systems operate and the recommendations they make. It also means being prepared for lawsuits if your AI systems cause harm. This level of accountability should force companies to take AI ethics seriously.
ASK QUESTIONS
Still iffy on AI? Work with your HR tech providers to understand how AI is being ethically applied.
THE ONLY THING WE HAVE TO FEAR IS…
By recognizing and preventing algorithmic redlining, adopting radical transparency, undergoing external ethical audits, empowering AI ethics boards, and accepting public accountability, corporate decision-makers can mitigate the risks associated with rogue AI behaviors.
But that’s not the most worrisome liability to organizations. In my view, the most worrisome would be choosing not to use AI. Companies that avoid AI will ultimately be left behind and put out of business. Mark my words.
In the banking sector alone, AI is reducing human errors, catching suspicious activity and reducing annual costs. And in healthcare, GE HealthCare (another client of my company) and Amazon are teaming up to help doctors improve diagnoses using AI. Amazon’s cloud computing services and generative AI models will help doctors more quickly and accurately diagnose ailments.
So, I recommend starting out by applying AI in one or two small but powerful use cases, such as email. Using GenAI to craft messages is a great way to demonstrate how it works while building people’s confidence in its reliability. Apple recently unveiled plans to use GenAI for writing tools that can proofread iPhone users’ emails, notes and texts. Google’s Chrome browser, meanwhile, has implemented a feature that writes or refines text based on webpage content.
FINAL THOUGHTS
AI doesn’t need enterprise-level IP to work well. It functions just fine with the data it has already been trained on.
So by all means, take AI for a test drive. See what it can do for your organization. But it’s time to stop treating it as a black box and start holding it—and its buyers—accountable for its outcomes.
AI can yield both prosperity and prejudice; the two are not mutually exclusive.