A grey-haired Industrial Engineer’s perspective on Garbage In, Garbage Out, fake news, and not letting shiny new toys replace common sense
AI is impressive. No argument there.
It writes reports, summarizes documents, generates images, drafts emails, answers technical questions, and never asks for PTO or even coffee breaks. That’s the “new tool” part.
But here’s the uncomfortable part most vendors won’t put into the slide deck:
If you bolt AI onto old behaviors, weak processes, poor data hygiene, and sloppy decision-making, you don’t get transformation. You get SSDD — Same Stuff, Different Day — at a higher cost.
And now it happens faster, with less time available to respond and recover.
Garbage In = Garbage Out (Now in High-Definition)
Manufacturing figured this out decades ago. Automating a broken process doesn’t fix it — it just produces defects at machine speed. AI is no different.
If you feed AI bad data, vague prompts, outdated assumptions, biased historical records, unverified sources, unclear objectives, it will return something that looks polished, confident, and authoritative — even when it’s fundamentally wrong.
That’s what makes AI different from traditional technology. Old systems failed loudly. AI fails quietly and convincingly. A wise man once told me “AI is 100% confident and 50% accurate.” Which brings us to the real problem…
Welcome to the Fake News Era (Now with AI-Assist!)
We already live in a world where misinformation spreads faster than verification and facts.
AI accelerates this facet of humanity, and it can be frightening to watch in real time. Misinformation can be generated instantly, written professionally, customized to the audience, and repeated in high volume.
This means the most important professional skill of the last decade just became even more critical: Validate everything that matters — regardless of the source.
Doesn’t matter if it comes from:
- a news outlet
- a consultant
- a dashboard
- a coworker
- social media
- or an AI system
If decisions, safety, money, or reputation are involved, blind trust is not a strategy.
The Real Risk Isn’t AI Being Wrong — It’s People Believing It Too Easily
One of the biggest dangers with AI is something engineers have seen before: Automation Bias. When a system looks smart, people assume it is smart.
We saw this with:
- autopilot systems
- alarm systems
- scheduling software
AI just makes the problem worse because it writes and speaks in full sentences. Humans start deferring judgment: “The computer said so.” That’s not decision-making; that’s outsourcing accountability and responsibility.
And the machine will happily accept the job — without liability.
New Tools + Old Behaviors = SSDD (Premium Edition)
In observing over 100 factories across dozens of industries and sectors, there are some common “leading indicators” I’ve seen related to poor performance in a troubled plant.
The Troubled Plant historically:
- rushes implementations
- avoids root cause analysis (or does it in a vacuum)
- ignores data quality & consistency
- de-prioritizes training (or skips it altogether)
- treats validation as an optional step
- chases shiny tools instead of fixing fundamentals
Now you want to use AI to solve your aches and ills. What could possibly go wrong? (hint: a lot)
You are setting yourself up for SSDD — Same Dysfunction, Different Dashboard.
Now “enhanced” with:
- subscription costs
- infrastructure expenses
- reputational exposure
- regulatory risk
- faster mistakes
Technology, especially AI in its current buzzword mega-hyped status, won’t fix your culture or change your organizational DNA, it rather exposes them to the light. There’s no “bad time” to work on fundamentals like Kaizen, RCA, and waste elimination / process improvement. Focus there first and then move to the advanced tactics – it’s what your organization can handle RIGHT NOW, for LOW COST.
Feeding AI Properly: Treat Inputs Like Engineering Specifications
Most “bad AI output” problems start with bad inputs.
Garbage Input Example:
“Give me the best KPIs for warehouse operations.”
If that’s your approach, then just “Google it” and be done. Typing it into an AI interface doesn’t make it “better” or more “modern.” The input is weak.
What warehouse? What industry? What maturity level? What organizational goals? What operational definitions? AI doesn’t read your mind (yet), so you must own the quality of the inputs. Ask an engineering question if you want an engineering answer.
Engineered Input Example:
“Generate KPIs for a mid-sized food manufacturing distribution center focused on throughput, inventory accuracy, and labor efficiency. Use standard industry definitions where possible and flag assumptions.”
Now you’ve:
- provided context
- defined scope
- established priorities
- asked for transparency
That’s how engineers talk to systems.
Validation: The Boring Step That Prevents Expensive Mistakes
AI output should never be the final answer. It should be the first draft. Before using AI-generated information:
Ask:
- What decision depends on this?
- What assumptions exist?
- What must be verified externally?
- What happens if this is wrong? (VERY IMPORTANT)
High-risk decisions deserve high scrutiny. Sometimes decision points result in a poor decision and thus, a CLM (Career-Limiting Move).
Pro Tip: Try to avoid CLMs by properly validating information.
******************************************
Why You Start Small (And Don’t Fly into a Storm on Your First Flight Lesson)
Learning AI is like learning to fly. You don’t start with bad weather, complex instrumentation, and needing to follow live emergency procedures. One of my colleagues is a pilot in his spare time, and he verified this.
You start simple: low risk use cases, training runs, with intense supervision, and gradual skill building.
Organizations that rush straight to “full automation” without building internal understanding usually become case studies.
Best in Class organizations:
- experiment safely, pilot new toys
- build expertise, develop internal SMEs
- create standards and templates (and follow them)
- scale up as the organization learns and embeds the new ways
The Upside: When AI Is Used the Right Way
We know the benefits we are pursuing with AI. When paired with good habits, AI becomes truly powerful in delivering faster documentation, better knowledge access, and reduced manual workload. It can also improve analysis support and reduce inefficiencies resulting in more iteration cycles (more with same).
But only when paired with validation discipline, accountability, and common sense. AI works best when it amplifies rather than replaces human thinking.
Final Thought from the High-Mileage IE
AI is not dangerous because it’s powerful. It’s dangerous because it can create unneeded churn and unintended consequences – specifically when people fail to think, fail to verify, and fail to challenge the implied “authority” of the tool.
The organizations that win this next hill won’t be the ones who “use AI the most.” They’ll be the ones who use it with discipline. They will learn the method by using it to improve efficiency in incremental and sustainable steps.
Garbage in still produces garbage out – It just arrives faster now, in Cortana’s soothing yet authoritative voice.
—Steve Wamsley, St. Onge Company
