The Mirror of Slop: Why AI’s Biggest Problem is That It Learned from Us
“The biggest issue with AI is that it learned from us!”
I spent yesterday trying to train a Gemini model to stop thinking like a 1990s database developer. I needed it to stop obsessing over pre-defined lookups and start thinking about contextual data. It struggled. It kept trying to add more fields, more variables, more rigid rules.
Today, I spent the day QA-ing the outputs, trying to stop the model from adding irrelevant or inaccurate context. "Be careful what you wish for," right? But as I sat there wrestling with the model, I realized something: my expectations were unrealistic. It’s easy to think, “It’s a computer; it should be consistent and smart.” But as it kept making mistakes (learning a rule, unlearning it, skipping a step, or missing a detail) I had a moment of clarity.
Whoa. Stop. This is exactly what it’s like working with humans.
Back when I was writing in Pascal, Fortran, and VBS, life was binary. If the program failed a QA test, it was 99% my fault. I wrote the code wrong. Period. But when I moved into leading large development efforts, a new layer appeared, interpretation. Now, a failure could be because the programmer coded it wrong, OR because they misinterpreted the requirements I wrote.
Every layer you move away from hands-on "switch-flipping" adds a layer of cloudiness. AI is simply the ultimate layer of cloudiness. It isn't a calculator; it’s a mirror. If it’s producing "slop," it’s because it was trained on the decades of slop we’ve produced in our boardrooms, spreadsheets, and warehouses.
The 2-Bug Fiasco: When the "Official Record" Lies
We expect AI to be accurate because it has "data." But data is often just a documented lie. Take my favorite example of human error, the 2-Bug “Fiasco.” I was blocking a major system implementation for a client. The Project Manager (PM) and Product Owner (PO) were furious. They escalated all the way up to the President of the company that I was being "unreasonable" because there were "only two bugs."
The President stopped me in the hall, fortunately he had curiosity and asked me about the situation. I explained the impact of those "two" bugs: “Clicking 'Load File' and clicking ‘Initiate Campaign’ do nothing. We literally can't run a single one of our 360 test cases.” I replied yes when he asked if the PM was aware of that (he was.) So, the president marched out of his office and fired the PM and PO on the spot.
The point? The official log said, "2 bugs." The reality was "0% functionality." If an AI were to read the emails and chats from that project, it would conclude I was the problem. It would learn that "2 bugs" is a minor issue, even if those bugs are catastrophic. AI can only read what it finds, and humans are notoriously bad at documenting the truth.
The Efficiency Trap: When "Lean" is Actually "Lazy"
We also see this in how we define "success." I once worked with a shipping specialist who wanted to "be more efficient." Instead of doing labels one at a time, he batched them in groups of seven. Management loved it. It looked like the mass production techniques used in the big plants.
The reality? He was mixing up labels on 20% of the orders. He wasn't increasing throughput; he was just creating more "down-time" for himself between batches. Until I forced the General Manager to go out and actually QA the boxes, the "official" narrative was that this shipper was a hero of our "dynamic problem-solving culture." He wasn't lean; he was lazy. But on paper, he looked like a genius.
Context is the Only Cure for Slop
Whether you are dealing with a LLM or a leadership team, the "slop" comes from the same place, asking for X without understanding the context of Y and Z. We have a habit of asking for outputs without engaging in the dialogue, the challenge, or the "Why." We want the AI (or the employee) to just "do it." But without a shared system of interpretation (without Decision Architecture) everyone just improvises.
So, give the AI a break. It isn’t failing because it’s a computer. It’s failing because it’s doing exactly what we taught it to do. It’s prioritizing the loudest signal over the most accurate one. It’s choosing "relief" (giving an answer) over "progress" (finding the truth). And, it’s following a playbook instead of reading the room. AI can’t run on its own any more than the humans it learned from can. If you want better results (from your tech or your team) you have to be the architect of the context.
At Growth Spectrum, we don’t just fix processes. We fix the Decision Systems that dictate how those processes are interpreted. Because if you don't clear the cloudiness at the top, you'll just end up with a faster version of the same old slop.
Continued Reading
The Age of Outsourced Discernment
The philosophical sibling to this post.
The Talent Isn’t Missing. The Discernment Is.
How a lack of context and discernment helps ATS and AI accelerate human slop in hiring.
The Invisible Moat
How context and "unseen" value are the only things AI can't replicate.
See If We Can Help
Decisions Systems Framework
Our approach and methodology to integrated cross-functional diagnosis and alignment.
Decision Architecture
Our services stack blending leadership, marketing, delivery, and decision systems together.
Case Studies | Growth Spectrum
See examples where we deliver 70-90% overhead reduction and 2x-3x scalable growth.
Risk-free Clarity Conversation
Reach out to see if we’re a good fit for a low-entry-cost, quick diagnosis, and plan.
