I once wrote an article titled “Do Your Metrics Matter?” It was inspired by the joy of sitting in an airplane, at the gate, going nowhere fast for close to 2 hours. All the while the airline app was proclaiming an on-time departure. Having ample time to discuss this perceived error with the crew, I learned that the closed aircraft door equaled departure. Departed, it turned out, was not an indication of a flight in motion but a process gone awry (a circumstance that may persist to this day, for what it’s worth). It was particularly discordant given the strident urgency of the flight crew urging folks to quickly take their seats to—you guessed it—ensure an on-time departure. Departed, indeed.
It’s no secret that what we attend to matters. What gets spotlighted gets governed. The real question is: What are we paying attention to and to what end? I would argue that critically assessing if your metrics matter is an evergreen exercise. To which, in the context of AI, a new variant has arisen: Do your principles matter?
Are your corporate value statements driving discrete behaviors and positively influencing decision making? Or are they performative? One can argue that such concerns have always been core to how businesses are run and products are developed.
This doesn’t negate the fact that AI systems, by virtue of where and how they are deployed, have raised the stakes—or at least our sights—on issues of ethics, trust, and risk.
In response, most organizations engaging in AI have codified numerous AI principles. The unfortunate (albeit obvious) truth is that an organization needs to focus on all of them, all the time. This does not mean that everyone needs to focus on everything, even part of the time. But knowing what needs attention by whom requires attention in and of itself. This will become even more critical as regulations such as the AI EU Act come into force.
You may be thinking, “Of course defining who does what is important. It is why companies organize around functions: sales, marketing, product management, and so on.” The difficulty when it comes to values, principles, and other overarching guidance is there are no clean delineations.
Responsibilities cannot be cleanly distributed: transparency to marketing, security to IT, explainability to R&D, risk mitigation to legal. One cannot (or should not) declare that human centricity is the primary domain of customer service while security belongs to IT alone. It’s a complicated many-to-many mishmash.
It is tempting to avoid the complexity and the hard work of defining this web of interrelated actions either by proselytizing singular value statements (i.e., we put customers first) or providing loose priorities (i.e., ease of access but not at the expense of security) that employees are expected to natively interpret and apply. It would be nice to believe this will promote the broadest scale coverage and adoption. However, experience shows that this is not the case.
Without structure to guide debate and resolve inevitable conflicts, those with the most prominent voices or convictions about specific issues will prevail. More seemingly esoteric (i.e., ethical or social questions) or less-understood principles fall by the wayside. If there isn’t a well-understood connection between a rote decision or task with the professed principle, it will be overlooked entirely. This is also a sneaky way of outsourcing accountability to people who may have the least authority to initiate or ensure action.
Moreover, in practice, not all principles are equally relevant for every function, as not all functions can meaningfully impact the realization of every principle. Yes, principles are important as broad directional guideposts. No, principles do not need to be role- or task-specific; this way lies madness. But the practical impact of a given principle in the course of any given employee’s day job does vary. So clearly defining what individual employees are most able to affect matters.
To govern effectively, one must clearly focus both the organization’s and the individual employee’s attention. Principles create a vision for how an organization wants to show up in the world.
But it is a paper exercise without clarity into how underlying business and technology practices realize these objectives. Clearly mapping how principles show up in functional roles and activities, where positive tensions are expected to arise, and how such tensions will be resolved is required. Doing this well requires dedicated time and attention. Simply put, principles are easy. Practice is hard.