Human-in-the-Loop: The Critical Balance in AI-Powered Decision Making

AI should not displace human beings, but augment human beings and our ability to execute. This philosophy is central to how we approach AI implementation at Iron Mountain.

A Practical Example: Data Classification at Scale 

Let me give you a very specific use case. Let's say a customer has terabytes of data.  

In years past, tagging and classifying data needed to be done by a human being. We were leveraging human beings to peruse terabytes, perhaps petabytes of data to classify information manually. 

AI can be effectively used to scan that data, pick up relevant information, summarize them into meaningful buckets, and then apply our policy center rules. The AI tags and classifies the data, noting data retention levels, personally identifiable information, PCI data, and HIPAA data. Then it sends this analysis to a human in the loop so that a human being can validate what the AI has provided and review the recommended actions. 

With AI, can turbocharge processes. We can move at a very fast pace because AI is capable of initial analysis, classification, data enrichment, and data tagging that previously required human involvement. This allows human cycles to be spent on tasks that require higher-level analysis and complex decision making. 

Addressing the "Laziness" Concern 

I've been asked whether AI might make humans lazy in their oversight responsibilities. I go back to the invention of a calculator. The introduction of calculators did not make human beings lazy, but it allowed us to use those calculations to enable things like AI and other technological advancements. 

I see AI as another tool in our arsenal—another tool in our toolkit that allows us to be that much more efficient. You're giving a bicycle to someone who is walking. Now they can go faster. Earlier, they may have only been able to travel 4 miles in a day. Now they can do 45 miles in a single day. 

The Quality Question 

The question isn't about the time spent on a decision—it's about whether the right decisions are being made. It doesn't really matter if the person manually grappled with the decision for two hours. Ideally, because the AI is making a suggested decision, it's actually better if humans take 15 minutes for a decision process, because then they can make more decisions and handle a higher volume. 

Human Oversight Remains Critical 

Human in the loop is a very critical element to successful, responsible AI. Human intervention and review are absolutely crucial, especially in areas that require decision-making that have significant customer impact. We cannot understate the importance of human oversight in AI implementations. 

This blog post is based on an original AIIM OnAir podcast. When recording podcasts, AIIM uses AI-enabled transcription in Zoom. We then use that transcription as part of a prompt with Claude Pro, Anthropic’s AI assistant. AIIM staff (aka humans) then edit the output from Claude for accuracy, completeness, and tone. In this way, we use AI to increase the accessibility of our podcast and extend the value of great content.