The adage “garbage in, garbage out” (GIGO) has been around for nearly 70 years, and it’s never been more relevant thanks to the intensifying need for information governance and the emergence of generative and agentic AI as potential disruptors.
Regardless of your mandate (e.g., regulatory compliance, privacy protection, migration to the cloud), the core principle remains the same: the quality of your output directly reflects the quality of your inputs. AI enthusiasts take notice.
It’s not news that our exploding volumes of data are riddled with junk, duplicates, inconsistencies, and outright errors, and that the accelerating pace of operations means we have less time and fewer resources to filter these out. The result is that we risk ending up with skewed analytics that lead to flawed reports, misguided strategies, and ultimately bad decisions.
AI Can Make it Worse
The rise of AI promises to make things worse because its computational power means we can get ourselves into trouble more quickly and more deeply than when we rely on our own cerebral power.
We already know that the generative AI models that so dominate information conversations these days are only as good as the data they’re trained on. So if that training data is of poor quality, contradictory, or lacking in relevance, the outputs will be subpar at best and dangerously misleading at worst – even as they are produced in the blink of an eye.
This is no less true for agentic AI, which being more process-oriented is taking the conversation to the next level (though whether that is a level up or down is kind of the point of this essay).
Agentic AI systems can autonomously take actions and make decisions based on their understanding of the world – an understanding that, currently at least, has to be taught to them by people who possess that knowledge.
The problem I have is that so many of the places I work with can’t clearly articulate what they do, why they do it, and how they make decisions. If they could, they might actually be able to boost the efficacy of their human workers and get less excited about having AI do it for them. But they can’t, because the exercise of capturing all this it is fraught with biases, limitations of scope, and sometimes outright falsehoods.
It seems to me that feeding instruction sets that contain these flaws into a process system that is designed to perform independently is less than a great idea because those flaws will inevitably affect how the system perceives and interacts with the world. And if you thought running a business using garbage information is problematic, wait ’til you try it with garbage process logic.
Success […] Requires Constant Effort, Vigilance and Reevaluation (Mark Twain)
The lesson here is clear: in our information-drenched, AI-powered world, we must be vigilant about the quality of the information we feed into our systems. Only by carefully curating, rigorously governing, and institutionalizing a commitment to truth and accuracy can we keep garbage from going in and thereby from flowing out.