AI and Ergonomics: The Future Is Knowing How to Put It All Together

Artificial intelligence is going to change ergonomics. In many ways, it already has.

Today, you can give an AI system a photograph of a workstation and ask it to identify potential ergonomic concerns. You can describe an industrial task and ask what risk factors might be present or which assessment methods might be appropriate. AI can help analyze injury data, interpret assessment results, organize worker feedback, draft reports, and generate potential recommendations. As computer vision and other technologies continue to improve, many of these capabilities will become faster and more sophisticated.

At Matheson, we see tremendous opportunity in that. We believe AI will become an important part of the future of ergonomic assessment, and we expect Matheson to be actively involved in figuring out how to use these technologies well.

But the more we work with AI, the more apparent something becomes: AI can only analyze the picture we give it, and work is much bigger than a picture.

That distinction may ultimately be one of the most important skills for the next generation of ergonomic professionals.

A Picture Can Tell Us a Lot, but It Cannot Tell Us the Whole Story

Imagine taking a photograph of someone working and asking AI to perform an ergonomic analysis.

It may do an impressive job. It can identify reaching, recognize awkward joint positions, evaluate workstation configuration, estimate postures, and identify potential areas of concern. With better computer vision, some of those measurements will likely become remarkably accurate.

The problem is not necessarily what AI sees. The problem is everything that exists outside the image.

A photograph may show someone holding a tool, for example, but how much force are they producing? How much resistance does the tool create? Is the worker lightly gripping it or exerting considerable force? Does the force change depending on the material being processed? Does the tool behave differently after several hours of use?

The same questions arise with repetition and duration. The image may show an awkward shoulder position, but does that position occur twice per hour or twice per minute? Is it maintained for three seconds or three minutes? Is the worker performing that task for twenty minutes or for most of an eight-hour shift? What happens between repetitions, and how much opportunity does the worker have to recover?

Those differences matter because ergonomic exposure is not simply posture. Force, repetition, frequency, posture, duration, and recovery interact over time. Other factors can modify that exposure as well.

So we give AI more information.

We provide the photograph and explain the force. We describe how frequently the task occurs and how long it lasts. We explain the rest of the work cycle. We provide worker feedback, symptoms, assessment findings, and perhaps injury history. We describe the equipment, environment, and production process.

The analysis gets better as the context gets better.

That is exactly why the knowledge of the person using AI remains so important. You have to understand ergonomics well enough to know what information AI is missing.

The Job You See Today May Not Be the Job That Exists Tomorrow

There is another problem with treating a snapshot of work as the work itself: jobs are not static.

An evaluator may observe a job on a relatively quiet Tuesday morning and see something very different from what occurs during the company's busiest season. Routine overtime can change exposure duration and available recovery. Staffing shortages can alter workload. Equipment problems can change how a task is performed. Production deadlines can increase pace. A worker who has been on vacation for two weeks may respond differently to the physical demands of the job than someone who has been performing it continuously.

Even the individual worker changes.

Someone experiencing discomfort may modify how they perform a task without consciously thinking about it. They may shift a load to the opposite side, change their grip, avoid a particular movement, work at a different pace, reposition themselves, or develop another strategy that makes the task more tolerable.

AI analyzing that worker may accurately describe the movement it sees. What it may not understand is why the person is moving that way.

That is an important distinction. The movement we observe could be contributing to the problem, or it could be the worker's response to a completely different problem.

Understanding that requires context.

Some of the Most Important Parts of Work Cannot Be Seen

Then there are the parts of work that do not appear in a photograph at all.

Consider two employees performing essentially the same physical job at two different companies. The equipment may be similar. The workstations may be similar. The physical demands may even look nearly identical.

Their experience of the work can still be very different.

At one company, employees may feel comfortable telling a supervisor when something is becoming difficult. The safety team may have a strong relationship with the workforce and respond quickly to early concerns. Line leaders may encourage employees to ask for assistance, adjust work when needed, and participate in identifying solutions.

At another company, employees may feel pressure to keep the line moving regardless of discomfort. They may believe that reporting a concern will be viewed negatively. Production expectations may routinely take priority over recovery or task modification. Workers may have little control over pace or how the work is performed.

Those differences matter.

Psychosocial pressures, workload, perceived support, communication, worker autonomy, supervisor relationships, safety culture, and organizational expectations can influence how work is performed and how workers respond to physical demands. They can also affect whether an ergonomic recommendation has any realistic chance of being adopted.

AI may eventually become very good at helping us analyze these factors too, but they first have to become part of the information being considered. A camera pointed at a workstation will not automatically tell us what the relationship is like between the worker and the line leader.

Someone has to know to ask.

The Better AI Gets, the Better Our Questions Need to Become

This creates an interesting paradox. It would be easy to assume that as AI becomes better at ergonomic assessment, professionals will need less ergonomic knowledge.

We think the opposite may happen.

Consider two people using the same AI system to evaluate the same workstation. One uploads a photograph and asks, “Is this workstation ergonomic?” The other recognizes that the photograph is only the beginning. They want to know about force, repetition, frequency, duration, recovery, symptoms, task variation, workload, production demands, worker behavior, organizational factors, injury history, and what happens during the rest of the shift.

The technology is identical. The quality of the assessment may not be.

The difference is that one person knows what questions still need to be answered.

This becomes even more important when the information does not neatly agree. A worker may report symptoms that do not align with a structured assessment score. An assessment method may identify a concern that does not appear in the worker's report. Injury data may point in one direction while direct observation seems to point somewhere else.

At Matheson, we place considerable importance on looking for convergence across different sources of information. When worker report, appropriate assessment methods, exposure analysis, and other available evidence point toward the same concern, our confidence in that finding becomes stronger. When they do not agree, that divergence is information too. It should make us ask why.

AI could become extraordinarily useful in helping evaluators process all of this information and recognize patterns. But the evaluator still needs to understand what each source can tell us, what it cannot tell us, and when the evidence suggests that we need to investigate further.

Assessment Is Going to Get Easier

This is where we think AI becomes particularly exciting.

A considerable amount of ergonomic work has historically involved gathering measurements, reviewing video, calculating scores, documenting observations, organizing data, comparing information, and preparing reports. AI has the potential to make many of those activities dramatically more efficient.

Computer vision may increasingly quantify posture and movement automatically. AI may help calculate structured assessment methods, identify patterns across injury data, summarize interviews, compare findings across workers or departments, and help evaluators organize enormous amounts of information that would be difficult to process manually.

We should want that.

There is little value in spending twenty minutes manually performing a calculation if technology can perform it accurately in seconds. The purpose of an ergonomic professional should not be to protect the manual labor involved in ergonomic assessment.

The more interesting question is what happens after the assessment becomes easier.

Because identifying a problem has never been the ultimate purpose of ergonomics.

The purpose is to improve the work.

The Recommendation Is Where the Work Becomes Valuable

Suppose AI eventually becomes exceptionally good at identifying exposure. It recognizes a concern, selects an appropriate assessment method, performs the calculation, and produces a reliable result.

We still have to decide what to do about it.

The technically ideal solution may be to raise a work surface, but perhaps the surface is integrated into a production line. Moving materials closer might reduce reach for one employee while creating another problem downstream. Job rotation might appear to reduce exposure until we realize that employees are being rotated between tasks with similar physical demands. Automation may solve the problem beautifully but require an investment that makes no sense for a product line scheduled to disappear next year.

The recommendation has to exist inside the reality of the company.

That means understanding production, equipment, workflow, staffing, budget, physical space, organizational priorities, worker behavior, and who actually has the authority to make a change. It also means understanding who will receive the assessment. A safety professional, operations leader, supervisor, human resources representative, and individual worker may all need different information from the same assessment.

The findings should not change depending on the audience, but how we translate those findings into action often should.

This is where the future of ergonomic expertise becomes particularly interesting. The value may increasingly move away from simply being able to identify and calculate exposure and toward being able to interpret it, prioritize it, communicate it, and do something meaningful about it.

Implementation May Become the Real Differentiator

Even a very good recommendation accomplishes little if it remains in a report.

Someone has to help move the organization from “we identified a problem” to “we changed the work.” That may require working with employees, supervisors, safety professionals, operations leaders, engineering, purchasing, or senior leadership. It may require modifying the original recommendation after learning more about the process. It may require testing a solution, gathering feedback, and determining whether the change actually reduced the exposure without creating a new problem.

That work is difficult because organizations are complicated.

And that is precisely why we believe it will become more valuable.

As AI lowers the effort required to perform portions of an ergonomic assessment, the ability to generate an assessment may become increasingly common. Producing a professional-looking report may become easy. Calculating a risk score may become nearly automatic.

The meaningful differentiator will be whether the person using those tools understands the work well enough to turn all of that information into change.

The Future of Ergonomics Is Putting It All Together

None of this is an argument against artificial intelligence. At Matheson, we are excited about what AI can bring to ergonomics, and we believe there are opportunities to do things with this technology that would have been difficult to imagine only a few years ago.

But we also think AI changes what it means to be good at ergonomics.

A photograph is one piece of information. A posture measurement is another. A structured assessment score is another. Force, repetition, duration, recovery, worker symptoms, injury history, production demands, psychosocial pressures, organizational culture, and the realities of the company all add pieces to the picture.

AI can help us collect, calculate, organize, and interpret more of those pieces than ever before. That is enormously valuable.

The professional still needs to know which pieces are missing, how they fit together, what they mean, and what should happen next.

That is why we believe the future is not human expertise versus artificial intelligence. It is human expertise amplified by artificial intelligence. As the assessment portion becomes easier, greater value will come from understanding the complete situation, developing the right recommendation, communicating it to the right people, and helping the organization actually implement change.

The future of ergonomics is not simply getting better at measuring work. It is getting better at putting the whole picture together and using it to make work better.