November 20, 2024

AI 101: "Black box" anxiety

GenAI can’t always explain itself—and here’s why that’s not so scary.
MarTech
TABLE OF CONTENTS

People often talk about AI as a “black box”—a technology that is making important decisions, often without being able to fully “explain itself.” Understandably, in a certain light, this can be anxiety-inducing.

When people refer to AI as a “black box,” they’re often referring to the lack of transparency in how these systems produce results. You feed the system data, and it gives you actionable outputs, but the exact process of how it gets there? That’s harder to explain.

The key issue is that even when AI seems to work—when its predictions or insights are accurate—people often don’t understand why it works.

One of the central concerns with the black box nature of AI is the training data itself: Where did it come from? What  kind of biases and errors exist in that data? How do I know I can trust what’s coming out of this thing?

Beyond biases, there’s the issue of accountability. Who actually takes the fall if something goes wrong here? It’s tough to place the blame when using a system like that.

While professionals might understand the broader mechanics of AI, even we may not know exactly how it got from inputs to outputs in a given scenario. But we know how it operates, in a broad sense.

For non-technical users, AI might feel like an incomprehensible magic trick. For professionals, it’s more like a tool we can use effectively, even if we don’t fully understand its inner workings.

This disparity in understanding between experts and laymen does feed public anxiety. But think of it this way: The human brain is arguably the ultimate “black box.” We take in a bunch of information, and then we end up trusting our gut intuition. We’re OK with that when we do it because we just trust ourselves.

However, this same implicit trust doesn’t necessarily extend to others—or to machines. Is this person, or software, worthy of taking information from? The issue isn’t so much the opacity itself but the familiarity—or lack thereof—with the entity producing the decision.

To build trust, companies can focus on transparency and can welcome regulation, but it’s not always as simple as that. While there’s growing talk of AI regulation, in a global context, that’s always going to be only as good as the weakest link. If everyone’s working on it around the world, and one country has heavy regulations on AI, all that means is they’re going to be behind. This creates a challenge for consistent global standards.

Another partial solution is leaning into familiarity and education. The fear is often simply of the novel or unknown. People see something that they don’t understand, and that gives them anxiety. If an LLM is acting “like a human,” people can understandably start to feel that it’s real—it’s perceiving, it’s thinking. But all the GenAI is doing here is literally just mimicking. That’s it. The results can be impressive, but it’s not magic.

While certain fears, like automation replacing jobs, are legitimate, greater transparency can help reduce other less plausible worries. AI should always be part of a system, rather than the final authority—that’s why you’ll always hear AI experts stressing the importance of having a “human in the loop.”

We also have to keep in mind the context we’re talking about. With martech, there are certainly stakes involved—but they’re not as high as they would be when integrating AI into fields like medicine, where there could be clear life-or-death implications.

Ultimately, the black box of AI reflects broader questions about how we trust and interact with technology. AI only knows what it’s been trained on. It takes all that information and it’s going to do its best to create an answer based on the inputs that were asked of it.

Understanding that—and setting guardrails—can make AI feel less like an alien technology and more like a tool we can work with.

Louis Criso

Louis Criso is the Head of AI Solution Development at Stagwell Marketing Cloud.

Take five minutes to elevate your marketing POV
Twice monthly, get the latest from Into the Cloud in your inbox.
Related articles
Coca-Cola’s risky AI video bet
Online reactions to a new holiday campaign were harsh—but did they miss the point?
Media
AI 101: "Black box" anxiety
GenAI can’t always explain itself—and here’s why that’s not so scary.
MarTech