AI Found Your Root Cause. Should You Trust It?
Artificial intelligence just reviewed the problem.
It analyzed the defect history, corrective actions, process data, inspection results, maintenance records, PFMEA, control plan, work instructions and supplier information.
Then it gave you an answer:
Here is the root cause.
Now what?
Do you trust it?
Maybe You Should
I think we're going to have to get comfortable with something that may make some quality professionals uncomfortable:
AI can find root cause.
AI can process enormous amounts of information, recognize relationships, compare records, identify inconsistencies and connect pieces of evidence that a person might overlook.
That capability is increasingly relevant to quality management. The ISO 9001 Auditing Practices Group Guidance on Auditing a Quality Management System that uses Artificial Intelligence identifies applications of AI in areas including design and development, supplier evaluation, predictive maintenance, quality control, production automation, process optimization, supply chain, training, and compliance and risk management.
So dismissing an AI-generated root cause simply because a machine found it doesn't make much sense.
But neither does automatically accepting it.
Then Again, Why Do We Trust People?
Suppose a quality engineer with 25 years of experience looks at the same problem and says:
"I've seen this before. I know what's causing it."
Would you accept that as the root cause?
I wouldn't.
Experience matters. Expertise matters. Process knowledge matters.
But none of them makes someone automatically right.
The engineer could recognize something everyone else missed.
The engineer could also be anchored to a previous experience that looks similar but isn't.
That brings us to the more interesting question.
Maybe we shouldn't be asking:
Can I trust AI?
Maybe we should be asking:
Why do I trust any root-cause conclusion?
The Answer Isn't the Root Cause
This is where quality investigations often get into trouble.
We feel pressure to find the answer.
So we perform a 5 Why, create a fishbone diagram, hold a meeting, review some process data and eventually arrive at something everyone agrees sounds reasonable.
Maybe it's:
Operator failed to follow instructions.
Inspection failed to detect the defect.
Preventive maintenance was inadequate.
Supplier material was inconsistent.
Training was insufficient.
The procedure wasn't followed.
Root cause found.
Corrective action assigned.
Problem solved.
Except sometimes it isn't.
The defect comes back.
And when it does, we usually start another corrective action.
But perhaps the first investigation never developed enough understanding to justify the conclusion in the first place.
AI doesn't create that problem.
We've been perfectly capable of doing that ourselves.
AI Changes the Investigation
Here's where I think AI becomes genuinely interesting for quality professionals.
Consider how much information exists around a significant manufacturing problem.
There may be years of:
Nonconformances
Scrap and rework
Process parameters
Inspection results
Customer complaints
Supplier corrective actions
Audit findings
Equipment maintenance
Engineering changes
Training records
PFMEAs
Control plans
Work instructions
Previous corrective actions
How much of it does the typical investigation actually examine?
AI gives us the ability to work through far more of that evidence and ask questions across it.
Maybe the defect isn't simply associated with Machine 4.
Perhaps AI discovers that every occurrence also followed a tooling change within a particular window, involved material from two supplier lots, and happened after a specific maintenance activity.
That could be enormously valuable.
AI hasn't made the investigation weaker.
It may have taken us somewhere we wouldn't have gotten on our own.
Artificial Intelligence (AI) for Manufacturing from NIST is specifically examining fit-for-purpose AI methods and human-AI teaming in manufacturing, including manufacturing use cases such as production scheduling, manufacturing-system integration, and preventive maintenance.
The potential isn't simply automating what quality professionals already do.
It's expanding what we're capable of examining.
But AI Said So Isn't Evidence
Here's where I draw the line.
If an AI system tells me:
"The root cause is..."
my next question is:
"Show me why."
What evidence supports the conclusion?
What evidence contradicts it?
What assumptions were made?
Are there other explanations that fit the available evidence?
Does the proposed cause explain when the failure occurs?
Does it explain when the failure doesn't occur?
Can we reproduce, remove, isolate or otherwise verify the relationship?
Those aren't special rules for AI.
Those are the same questions we should have been asking all along.
If an engineer identifies the root cause, prove it.
If a supplier identifies the root cause, prove it.
If a corrective-action team identifies the root cause, prove it.
And if AI identifies the root cause?
Prove it.
Trust the Evidence
That's where I think the conversation around AI and quality needs to go.
We shouldn't blindly trust artificial intelligence.
But we shouldn't blindly distrust it either.
And we certainly shouldn't pretend that human judgment has somehow been immune from assumptions, bias, incomplete information or incorrect conclusions.
AI gives quality professionals another extremely powerful analytical capability.
Use it.
Let it search.
Let it compare.
Let it challenge.
Let it find relationships.
Let it find the root cause.
Then determine whether the evidence demonstrates that the conclusion is actually true.
Because ultimately, it shouldn't matter whether the answer came from a quality engineer with 30 years of experience, a cross-functional team, a statistical model or artificial intelligence.
The standard should be the same.
Can you demonstrate that you understand why it happened?
That's the part that matters.
That's True Root Cause.
TRC Quality Consulting LLC
Look deeper. Understand why. Improve what matters.
