Artificial intelligence — you can’t go a day without hearing it mentioned in the news. On LinkedIn, some forensic engineers forcefully state that AI should not be used at all. So what is the issue with AI in forensic engineering? Should it be embraced, or ignored entirely?
After more than 30 years of failure analysis and work as a forensic engineering expert witness, my answer is: neither. AI is a tool. Like every engineering tool, it is only as reliable as the data behind it and the engineer who checks its work.
Key Takeaways
- AI is only as good as the data it is trained on. Bad data means bad results.
- Every engineering tool — AI, CAD, multi-physics software, or a spreadsheet — requires an engineer who understands its capabilities and limitations.
- The scientific method still governs forensic investigations. The data must support the conclusion.
- The engineer whose name is on the report is responsible for every conclusion in it, no matter what tools were used.
The Good and Bad Sides of AI
AI, as with most things, has its good and bad sides. There is great potential for it to be useful in many areas: streamlining customer service (what we love to hate), aiding in simple medical diagnoses, summarizing massive amounts of information, supporting engineering design, and the list goes on.
The downside is that AI is no better than the data it is trained with:
- Bad data — bad results.
- Limited data — the potential for erroneous results.
- Comprehensive, good data — the potential for good outcomes.
Every New Technology Brings Both Praise and Fear
Should we embrace AI? Should we ban it? Powers much greater than I will make that decision, of course. But every new technology brings the good and the bad. Remember when the internet was just getting started? Some praised it, and others questioned its value and its potential negative effects on society. That debate (social media, etc.) continues to this day.
What about the introduction of the telephone? A quick search on its history shows the same pattern — praise for its potential, and fear of the negative things that might result from its use. AI is following the same path.
AI Is Just Another Engineering Tool — Know Its Limits
In forensic engineering, tools must be used with care. That tool might be AI, an advanced CAD/CAM package, a complex multi-physics engineering analysis code, or something as simple as a spreadsheet. Whatever the tool, the engineer using it needs to understand its capabilities and limitations.
If you are performing complex engineering analyses on new products or designs, you had better understand what the solution “looks like” before you start running the analysis. Otherwise, how will you know whether the answer is reasonable, or even correct? The engineer has the responsibility to verify the results until they are confident the answers are correct and properly represent the physics of the problem.
A Lesson From a Seismic Stress Analysis
Many years ago, a design engineer was performing a structural analysis of a table undergoing seismic loading. The results showed what they thought were unrealistically high stress levels in the table, and they could not understand why.
I asked how they had set up the problem and found they had fixed the legs rigidly to the floor — allowing absolutely no movement. After a brief discussion, they realized that was not a realistic boundary condition for a leg sitting on a floor. That assumption caused the high stresses in the legs and table.
The computer code worked perfectly for the conditions it was given. The same will be true of any new engineering tool, including AI: the right inputs and information are needed to ensure a reasonable answer is obtained.
How AI Tools for Forensic Engineering Must Be Validated
The introduction of AI tools for forensic engineering will be no different. Tools will be developed, and before they can be trusted in an investigation:
- They must be tested on relevant data sets and realistic failure scenarios.
- Their results must be checked against cases where the answer is already known.
- The user must verify that every outcome is realistic and supported by the physical evidence.
The Scientific Method Still Comes First
In forensic engineering — and in failure analysis, for that matter — the use of the scientific method is critical. Without going into a full-blown discussion, the concept is simple: data is collected, data is analyzed, and hypotheses are proposed and tested. The data must support the conclusion reached.
If it doesn’t, more data is needed if it exists. If all available data has been used and still does not support a hypothesis, the cause must be classified as undetermined. This is the same discipline used in fire origin and cause investigations and every root cause analysis we perform.
In litigation, this matters even more. An expert’s opinions must rest on sufficient facts and reliable methods. A conclusion produced by an AI tool that the engineer cannot explain, reproduce, or support with evidence will not hold up to scrutiny.
What Ford’s AI Experience Teaches Engineers
A recent article in the trade journal Design World (August 2026, p. 14) described Ford rehiring 350 veteran engineers — called “gray beards” by internal staff. Why? The company had implemented AI-based quality tools, and they were not working as intended.
According to Ford, the process failed because many experienced engineers left before their knowledge made it into the system. Ford VP Charles Poon was quoted as saying AI is “only as good as the information you use to train it.” Just like any engineering tool, AI needs engineering knowledge and good data to be useful.
Where AI May Help Forensic Investigations
AI tools for forensic engineering, just like those in any other branch of engineering, will be developed to streamline the work, support the investigation, and help reach conclusions that are well supported. This will take time, effort by many people, and vast amounts of data.
It is my understanding that some organizations are already streamlining the report-writing process with AI. The engineer with their name on the report must, of course, read the report, edit it, and make sure every conclusion is properly supported.
Documentation Matters More Than Ever
For an AI tool to reach the right conclusions, the underlying record must be complete. That means the engineer’s field notes, photographs, measurements, and failed components (the evidence) must be properly documented and preserved. Poor documentation leads to poor AI output — and to conclusions that cannot be defended.
The Bottom Line: Use AI in Forensic Engineering With Care
For now, AI needs to be watched closely and properly trained if it is to be used successfully. As in every field of engineering, these tools will work their way into forensic engineering over time. It is up to developers and users to ensure good data is used to train the tools and that the proper conclusions are being reached.
Need a Forensic Engineer Who Verifies Every Conclusion?
Dr. Randy Clarksean, Ph.D., P.E., CFEI, brings more than 30 years of experience investigating mechanical failures, fires, and explosions for attorneys, insurers, and manufacturers. Every opinion is grounded in the evidence and the scientific method.
Call (218) 371-1967 or request a case review today.
Frequently Asked Questions About AI in Forensic Engineering
Yes, as a supporting tool. AI can help organize and summarize large amounts of information, but it must be tested on relevant data, and a qualified engineer must verify every result against the physical evidence.
No. An expert witness must explain and defend their methods and conclusions. AI output that cannot be traced to evidence and sound engineering principles will not stand up in litigation, and the engineer who signs the report remains responsible for its contents.
The biggest risks are poor or limited training data, incorrect inputs (like an unrealistic boundary condition), and conclusions that are not supported by the evidence. Any of these can lead to the wrong root cause.
Follow the scientific method: collect the data, analyze it, and test each hypothesis against the evidence. If the data does not support a conclusion, gather more data — or classify the cause as undetermined.
