The Field Is More Than a Place
What Does “The Field” Really Mean?
Designing AI Around How People Actually Work
Recently, I have been thinking a lot about what we mean when we say “the field.”
The question came from an uncomfortable experience.
In one activity I had been involved in alongside my main job, I was told, in effect, that I did not value the field enough.
That hit me harder than I expected.
The project had gradually become complicated. Responsibilities were not always clearly defined. New tasks appeared along the way, directions changed, and I often covered work beyond what I had originally expected because of the relationship I had with the people involved.
There were also discussions about visiting the actual site, but even the purpose and timing of those visits were not always clear.
I have a full-time job. My life has also changed over time. My time, energy, and money are limited.
Of course, there were probably times when I could have visited more often.
But it was never because I did not want to understand the site.
I had been trying to support the work in whatever way was possible at the time: remotely when appropriate, by thinking about cost, by testing technical alternatives, and by trying to reduce the burden on the people who would actually use the system.
Still, I experienced a situation where not visiting often enough seemed to become a judgment about my attitude and about the work I had done so far.
What I had wanted to discuss was the scope of my role and how to proceed.
The conversation eventually moved beyond the work itself.
It was quite a shock.
I am now gradually stepping away from that activity.
But afterward, one question stayed with me.
What had I actually meant by “the field”?
The Field Is More Than a Physical Place
Going to a farm matters.
Going to a factory matters.
Meeting users face-to-face matters.
There are things that can only be understood by seeing the actual environment or speaking directly with the people who work there.
But I have started to feel that the “field” I have been trying to understand is larger than a location.
For example, when building systems with generative AI, getting the technology to work is often only the beginning.
A system can technically work and still never become part of someone’s actual job.
There may be too many steps.
People may not trust the output, so they check everything manually anyway.
The cost may be too high.
The users may already be overloaded.
Or the organization may simply not allow the workflow we imagined.
The AI works.
But it does not fit into the work.
That is when I have to ask:
What did they actually need?
So I try something, listen, and revise it.
Instead of bringing in the technically ideal solution unchanged, I try to understand how the person already works, what matters to them, and how much change they can realistically absorb.
I am beginning to realize that this is part of what I have meant by working with the field.
Every Person and Organization Is Surrounded by Forces
I sometimes imagine people and organizations surrounded by arrows pointing in many different directions.
They want to achieve something.
But money, time, and energy are limited.
They want to try new technology.
But they may already be overwhelmed by their existing work.
They want greater efficiency.
But they may also worry about what happens when the system is wrong.
There is time, cost, technology, equipment, relationships, organizational structure, health, values, long-term goals, and external factors such as weather that nobody can control.
And these things do not operate independently.
Money affects time.
Staff changes affect operations.
A change in weather can suddenly change which problem matters most.
A person exists inside all of these relationships.
That is why the phrase “understanding the whole activity around the other person” has started to feel more natural to me than simply “going to the field.”
The physical place matters.
But so do all the conditions surrounding what the person is trying to do.
Technology Should Fit Into the Activity That Already Exists
For me, introducing technology is not the goal itself.
Get Tomohiro Mori’s stories in your inbox
Join Medium for free to get updates from this writer.
People already have jobs, routines, responsibilities, relationships, and ways of doing things before new technology arrives.
What I increasingly want to do is not simply bring technology into that environment, but make it fit naturally into the activity that already exists.
If I introduce a new system, I want it to preserve what already works while making something a little easier, or enabling something that was previously difficult.
That means looking beyond the technology itself.
Who will use it?
How much can they afford?
How much extra work will it create?
How much can they trust the result?
Who will maintain it when something goes wrong?
Sometimes a slightly less capable model is better if it is much faster and cheaper.
Sometimes full automation is worse than letting a human make the final decision.
Sometimes the technically elegant solution is simply the wrong solution for the person who actually has to live with it.
The answer cannot be determined by technology alone.
It has to emerge through interaction with the people who will use the system.
AI Does Not Remove Ambiguity
This has also changed the way I think about AI itself.
In computer vision and generative-AI projects, it is tempting to think that the main challenge is building the model.
But often the harder question comes first:
What exactly are we asking the AI to decide?
In agriculture, animal-related research, and many other real-world domains, apparently simple categories can become surprisingly ambiguous once people actually begin labeling data.
Sooner or later, someone asks:
“Which side of the boundary does this example belong to?”
And sometimes the honest answer is:
“I’m not sure.”
That ambiguity does not disappear just because we use deep learning.
If people cannot explain what they want the AI to recognize, the same problem eventually comes back during evaluation.
The model produces an answer.
Then someone has to decide whether that answer is correct.
AI can reduce annotation cost.
It can automate parts of a workflow.
It can accelerate development dramatically.
But it does not remove the need to decide what matters.
In some cases, it makes that ambiguity more visible.
Leave Materials for Judgment, Not Just Answers
Respecting someone’s activity may also mean not forcing our own “correct answer” onto their situation.
I have started thinking about this through a much smaller project as well: experiments with planters, sensors, and AI on my balcony.
I do not want to leave behind instructions that say:
“Do exactly this.”
The next person will have a different location.
A different budget.
Different weather.
Different time constraints.
Even my future self will be operating under different conditions.
So what I would rather leave is something like:
Under these conditions, I saw this problem, made this decision, tried this approach, and observed this result.
That is not a command.
It is material that someone else can use to make their own decision.
I think the same applies to technology, research, and AI.
Caring about the field does not necessarily mean visiting the same place over and over again.
It means trying to understand:
What is the other person trying to do?
What is making that difficult?
What constraints are they operating under?
What do they care about?
And then trying to find a form that both sides can sustain without unnecessary strain.
I still do not think the experience that triggered this reflection was a good one.
But it did help me put something into words that I had been trying to understand for a long time.
For now, I am beginning to think of “the field” not simply as a place, but as the other person’s activity and the whole set of conditions surrounding it.
