The Most Interesting AI Problems Aren’t Always On a Screen
Much of the conversation about AI involves things on a screen.
Writing emails, summarizing meetings, creating content, responding to customer service requests.
These are all valuable applications, but I’ve been more interested recently in a different set of problems: one in which a machine is changing behaviors, needing maintenance, or a worker has to respond to some activity with limited information.
These are interesting problems because the information often isn’t easily accessible, and can often be found as a series of disconnected pieces of information.
This story was written with the assistance of an AI writing program and reviewed by the author.
Real World Problems Rarely Present Themselves On a Screen As Clean Information
Imagine a piece of equipment that needs maintenance:
“Machine 14 is making a strange noise.”
The truth is, an organization may well have other information beyond what the maintenance request may say: other maintenance records, sensor data and logs, work orders, the hours and location of the machine, inspections, photographs, and more.
The interesting question isn’t whether or not the AI can consume a text query, but whether or not the system overall will be able to combine relevant pieces of information.
This moves the question from something like
Request → AI → Answer
to something like
Request → Context → Analysis → Recommendation → Human Decision → Outcome.
It’s a critical, and extremely valuable, difference.
Context Makes These Problems Interesting
A single sensor reading rarely tells the whole story.
If a machine’s vibrations are increasing, is the system starting to fail? That may be the case.
However, the system could also be seeing increased loading or a change in speed, or that the sensor isn’t accurate, or that maintenance was recently performed.
Get Nayantara P S’s stories in your inbox
Join Medium for free to get updates from this writer.
AI is often most interesting then when we aren’t looking at individual outliers, but instead evaluating multiple data points.
That’s why IoT and AI can often go together: IoT can offer context to AI, enabling richer analysis, while AI often has the power to unlock insights from IoT.
AI Doesn’t Always Need to Make Decisions
Much discussion of AI focuses on how much autonomy a system can have.
That’s a discussion that’s far less interesting in many physical systems.
If an AI observes some equipment failure mode and alerts a maintenance worker, that seems like a win.
Could the AI have scheduled the maintenance? Sure, but that doesn’t necessarily answer the better question of who should be responsible for making that decision.
The responsibility always rests with people, and while automation is desirable, workers should make decisions based on their own experience and judgment.
The Most Important Part of the System May Not Be the Cool Demo
An interesting demo for AI might involve examining a picture and identifying objects in the image.
However, what happens after the result is known is often far more important.
Is there an alert generated? Did it notify the right person? Will the information make its way into a work order or some other system? Is there the option for a worker to review and update information? Can this information be used for training or other downstream insights?
Without that ability, an AI system may well become nothing more than another dashboard for workers to ignore.
The most valuable AI features are often the ones that support an operational workflow.
Focus on the Annoying Problem
One common approach to these problems is to try to forget about AI for a moment: what operational or technical problem really wastes the most time for workers?
It might be the maintenance request example above, or it might be inventory records disagreeing with what’s actually on the floor, or the same information being in multiple systems requiring workers to find and update all of them when something changes, or some other process.
Once one or more problems are identified, the next task is to decide whether AI, IoT, automation, or something else would help.
At Aperture Venture Studio, we focus on building AI and IoT ventures focused on the physical world, linking data, context, and operational workflows together.
The future of AI may very well be focused on enabling workers to make decisions about the physical world.
A chatbot embedded in every system may be interesting, but the more valuable opportunities are often those that help workers interact with information and make decisions about the physical world.
