When People Ask Me About AI…

When People Ask Me About AI

People talk to me about AI a lot these days.

Sometimes they’re curious. Sometimes they’re excited. Sometimes they’re skeptical.

And sometimes I can tell from the question that they’ve already decided what they think about it before the conversation even starts.

I get it.

AI is everywhere right now.

Every company has an AI strategyor lack of one. Every software platform suddenly has an AI feature. People are generating pictures, writing emails, building agents, automating workflows, making music, analyzing data, conducting research, and doing things that would have sounded like science fiction not that long ago.

And somewhere along the way, we’ve reduced a very complicated conversation to one question:

“Did you use AI to do that?”

I never quite know how to answer.

Because usually the answer is:

Yes.

But the more important answer is:

Yes. And I did a whole lot of thinking too.

I didn’t arrive at AI because it was trendy.

My relationship with AI didn’t start with ChatGPT, nor is it limited to it. I regularly work with several different models, products, and platforms.

And my experience with AI hasn’t been limited to prompting a polished generative AI tool and seeing what comes back.

During my data science work, I spent time much further down the stack—building environments, working in Docker, experimenting with models and tools, managing dependencies, troubleshooting failures, and getting closer to what is actually happening underneath the interfaces most people interact with today.

That experience changed how I think about AI.

When you have to build the environment and understand how the pieces connect, a model stops feeling magical.

You start seeing the machinery underneath the answer:

The data.

The architecture.

The assumptions.

The parameters.

The experimentation.

The failures.

The iterations.

The evaluation.

And all the human decisions that happen before anything produces an output someone else eventually calls “AI.”

That is one reason I sometimes struggle with how casually we talk about AI now.

The interface has become so easy that the technology underneath it can feel almost magical.

It isn’t.

There are models. Data structures. Computational environments. Constraints. Design choices. And people making decisions at every layer.

My interest in this started long before generative AI.

For more than a decade, my professional work has lived at the intersection of healthcare, data, analytics, digital products, enterprise systems, workflow, and decision-making.

I’ve worked in environments where information has to move between systems, data has to mean the same thing to different people, business rules matter, exceptions matter, and seemingly small decisions can create very real downstream consequences.

Over time, I became increasingly interested in a problem that existed long before generative AI entered everyone’s vocabulary:

How do we take all the information an organization has and turn it into better, more consistent, explainable decisions?

That question eventually became much bigger than analytics for me.

My graduate work in data science gave me another lens through which to explore it.

Data science reinforced something I now think about constantly:

Producing an answer is not the same thing as producing a trustworthy decision.

A model can be technically impressive and still be unusable.

A prediction can be statistically valid and operationally irrelevant.

An algorithm can perform exactly as designed and still create problems because its assumptions are wrong.

And an AI system can produce a beautifully written, incredibly confident answer that is completely inappropriate for the situation in which you’re trying to use it.

That distinction has shaped nearly everything I’ve built since.

And since we’re here: please don’t mansplain AI to me.

I say that with humor.

But only a little.

You absolutely do not have to agree with me about AI.

There are legitimate questions about how we use it, where we use it, what we automate, what we trust, and what boundaries we place around it.

I spend a significant amount of my own time asking those questions.

But please don’t assume that because you see me using a finished AI product, that represents the extent of my understanding of the technology.

It doesn’t.

I have graduate training in data science. I’ve worked closer to the underlying technology. I’ve spent years professionally at the intersection of data, analytics, healthcare technology, enterprise systems, automation, and decision-making.

And now my academic work is taking me deeper into questions involving evidence, governance, decision support, human-technology interaction, and how we responsibly translate information into action.

So when I use AI to help edit a blog post, research a concept, challenge an argument, create music, or organize my thoughts, I’m not confusing the polished interface in front of me with the entirety of artificial intelligence.

I know there’s machinery underneath the magic.

I’ve worked with some of it.

That doesn’t mean I know everything about AI.

Nobody does.

It does mean I probably don’t need the introductory lecture.

AI is only one part of the decision system.

One of the things I’ve become increasingly interested in is separating pieces that organizations too often collapse together.

There is the data.

There is the business knowledge that tells us what the data means.

There are the rules and policies that establish what is allowed.

There is the evidence required to support a decision.

There is the decision logic that determines what should happen.

There is the technology that executes or recommends an action.

And there is the outcome, which should teach us whether the decision was actually good.

AI can participate in that system.

But AI should not automatically own all of it.

That distinction becomes increasingly important as organizations race to build AI agents.

Because once a system begins acting instead of merely answering, the stakes change.

We are being asked to move faster.

This is another part of the AI conversation I think gets overlooked.

In business, AI is no longer simply an optional experiment.

There is a very real mandate to use it.

Work faster.

Learn faster.

Research faster.

Build faster.

Analyze faster.

Automate work that doesn’t require human judgment.

Find opportunities we might otherwise miss.

Reduce the amount of time we spend on repetitive work and increase the amount of time we spend on work that genuinely requires thinking.

I fundamentally agree with that direction.

I don’t think the answer is to avoid AI because it makes us uncomfortable.

I think the responsibility is to learn how to use it well.

Become faster without becoming careless.

Learn more quickly without confusing speed with understanding.

Automate where automation makes sense without pretending every decision belongs to a machine.

And recognize that if AI accelerates the pace of work, governance cannot remain something we bolt on at the end.

Governance has to accelerate too.

Human-in-the-loop isn’t a weakness.

One phrase that appears constantly in AI conversations is human in the loop.

Sometimes it is discussed as though the human is temporary—as though the system simply isn’t mature enough yet, so we’ll leave a person involved until the technology gets better.

I don’t think that is always the right framing.

There are decisions where a human should remain involved because human judgment is part of the design.

Not because the AI failed.

Because the decision requires judgment.

Context.

Nuance.

Accountability.

Ethics.

An understanding of circumstances that cannot always be reduced to a rule.

In some situations, AI should recommend.

In others, it should summarize.

In others, it should identify an exception.

And in some clearly bounded situations, it may be appropriate for AI to act autonomously.

The sophistication isn’t in removing humans from everything.

The sophistication is knowing where humans belong.

And then there is traceability.

This may be one of the least exciting words in the AI conversation.

It is also one of the most important.

If an AI system recommends something—or if an agent acts on someone’s behalf—I want to know more than what it did.

I want to know why.

What information was available at the time?

Which rule was authoritative?

Which version of that rule was used?

What evidence influenced the recommendation?

Was there an exception?

Did a human override the recommendation?

If so, why?

What happened afterward?

Could we reconstruct the decision six months later?

That is traceability.

And once AI begins participating in consequential decisions, traceability becomes foundational.

Because:

“The AI said so” is not governance.

AI makes research faster. It does not eliminate the research.

I use AI professionally, academically, creatively, and sometimes personally.

One of the things I’ve intentionally learned is where it creates leverage—and where I still have work to do.

Research is a good example.

AI can help me find a theory, framework, methodology, author, research area, or keyword much faster than I could have discovered it on my own.

And then I go read.

That distinction matters.

I have done literature reviews the traditional way, academically and professionally.

Finding the literature is work.

Determining what is relevant is work.

Reading it is work.

Evaluating the methodology is work.

Comparing findings is work.

Identifying gaps is work.

Synthesizing all of it into something meaningful is work.

AI can accelerate discovery and help organize bodies of information.

It cannot relieve me of responsibility for the work.

I still have to recognize when a search is too broad.

When concepts are being conflated.

When a source doesn’t actually support the claim being made.

I still have to return to the original literature.

Evaluate the evidence.

Understand the methodology.

Decide what belongs in the final synthesis.

And ultimately, if my name is attached to the work, I am responsible for what it says.

AI may reduce the time it takes me to get to the reading.

It does not eliminate the reading.

The prompt matters more than people think.

There’s a strange assumption that using AI means pressing a button and receiving something useful.

That is not my experience.

I still have to understand enough about the problem to frame it.

I have to know which constraints matter.

I have to provide the right context.

I have to recognize when I’ve asked a bad question.

Sometimes an answer is poor because the AI failed.

Sometimes it is poor because the prompt was poor.

Sometimes it is technically correct but completely useless because I asked the wrong question.

And sometimes the most valuable thing that happens is that AI gives me exactly what I asked for—and I realize what I asked for was wrong.

That is useful too.

The output isn’t the intelligence.

This may be one of the biggest misconceptions I encounter.

We look at an output and attribute the intelligence entirely to the machine.

But increasingly, I think the more interesting question is:

What did the human bring to the interaction?

AI can generate an answer.

That doesn’t mean it asked the right question.

It can identify patterns.

That doesn’t mean it understands which patterns matter.

It can generate twenty ideas in seconds.

That doesn’t mean any of them are good.

It can sound extraordinarily authoritative while being spectacularly wrong.

And it can absolutely reinforce a bad assumption if the person using it doesn’t know enough to recognize that the assumption is bad.

That’s why I don’t believe expertise becomes less important in an AI world.

I think it becomes more important.

When generating something becomes incredibly easy, judgment becomes incredibly valuable.

Knowing what to accept matters.

Knowing what to reject matters.

Knowing which question to ask next matters.

Knowing enough to say:

“Wait. That’s wrong.”

matters.

Recognizing:

“You’re conflating two different things.”

matters.

And realizing that the machine gave you exactly what you asked for—but you asked for the wrong thing—may matter most of all.

I don’t want AI to think for me.

I want it to think with me.

There’s a difference.

I don’t hand AI a problem because I don’t want to think about it.

Usually, I hand it a problem because I want to think about it more.

From more angles.

Challenge me.

Show me what I’m missing.

Find the weakness in my argument.

Help me connect ideas I haven’t connected yet.

Break something enormous into manageable pieces.

Point me toward research I didn’t know existed.

Take something I’ve been staring at for three hours and ask the question my tired brain isn’t asking anymore.

Sometimes I want acceleration.

Sometimes I want friction.

Sometimes I need a blank page to stop being blank.

Sometimes I need an editor.

Sometimes I need a research assistant.

Sometimes I need a coach.

And sometimes I just need something to push against.

That last one may be my favorite use of AI.

Because there have been plenty of times when AI gave me an answer and my immediate reaction was:

Absolutely not.

And suddenly I knew exactly what I thought.

And yes, I’ve used it to make music.

That one tends to generate some interesting conversations.

People hear a song and ask whether AI created it.

Well…

AI was involved.

But AI didn’t live the life behind the lyrics.

It didn’t experience what I was trying to process.

It didn’t decide what I believed about it.

It didn’t know when a line sounded like something I would never say.

It didn’t know when something felt too cheesy, too polished, too generic, too churchy, too corporate, too safe—or simply not me.

I did.

Anyone who has watched me work with AI for more than five minutes knows I spend an unreasonable amount of time saying:

No.

Try again.

That’s not what I mean.

You’re missing the point.

That’s technically correct, but it isn’t true to the situation.

Go deeper.

Take that out.

You’re conflating two different things.

Start over.

Those interactions are probably a much better representation of how I use AI than the finished product ever could be.

But I’m not blindly optimistic about AI either.

The more capable these systems become, the less comfortable I am with the idea that simply because we can automate something, we necessarily should.

Especially when AI moves beyond drafting an email, searching for research, organizing notes, or brainstorming an idea and begins recommending—or taking—actions that affect people, customers, businesses, money, healthcare, access, opportunity, or risk.

At that point, the question cannot simply be:

Can the AI do it?

We also have to ask:

Should it?

Based on what information?

Using whose rules?

Who determined those rules were correct?

What evidence is required?

What happens when the information conflicts?

What happens when an exception occurs?

Can a human intervene?

And six months later, can we reconstruct why that decision was made?

Those aren’t merely technology questions.

They are human questions.

Governance questions.

Business questions.

Ethical questions.

Accountability questions.

Ironically, the more sophisticated AI becomes, the more important those questions become.

AI has made me think more about what is uniquely human.

Maybe that is the part I didn’t expect.

For all the conversations about artificial intelligence, using it has made me think an awful lot about human intelligence.

Curiosity.

Discernment.

Context.

Experience.

Taste.

Wisdom.

Empathy.

Conviction.

Creativity.

The ability to recognize that something can be logically consistent and still be wrong.

The ability to understand why something matters.

The ability to know when the evidence is insufficient.

The ability to recognize nuance that doesn’t fit neatly into a rule.

The ability to determine not only what can be done, but what should be done.

AI can help me write a sentence.

It cannot decide what I want my life to say.

It can help me explore an idea.

It cannot determine what I believe.

It can help me find something to read.

It cannot decide whether the evidence changes my mind.

It can help me create.

But it doesn’t have something it needs to create because it lived through it.

I do.

We do.

And I don’t think we should surrender that distinction simply because the technology is impressive.

So yes. I use AI.

A lot.

Probably more than most people realize.

I’m fascinated by it.

I experiment with it.

I push it.

I question it.

I research with it.

I build with it.

I create with it.

I use it to edit.

I use it to challenge assumptions.

I use it to find things I need to go read.

And occasionally I argue with it like it is a coworker who has confidently misunderstood the assignment for the fourth time.

But I don’t see AI as a substitute for my intelligence.

I see it as leverage for it.

And the quality of that leverage still depends heavily on what I bring to the conversation:

My experience.

My education.

My domain knowledge.

My questions.

My judgment.

My creativity.

My values.

My understanding of the context.

My willingness to verify what I’m being told.

And my willingness to say:

No. That’s not right. Let’s try again.

We are being asked to work differently.

To learn faster.

Research faster.

Experiment faster.

Use AI where it creates leverage.

I think we should.

But moving faster doesn’t eliminate our responsibility to understand what we’re building, writing, citing, automating, or deciding.

It increases it.

As AI participates in more of the decisions organizations make, I don’t believe the differentiator will simply be who has the most agents or who automates the most tasks.

It will be who can build systems that are:

Fast and trustworthy.

Automated and accountable.

Intelligent and explainable.

AI-enabled, while still knowing exactly where human judgment belongs.

And when a decision matters, we should always be able to answer:

What happened?

Why did it happen?

What information was used?

What evidence supported it?

Who had authority?

Where was the human?

And can we prove it?

That, to me, is a much more interesting conversation about AI.

I’m not interested in handing my thinking over to a machine.

I’m interested in discovering what becomes possible when human expertise, governed knowledge, good data, strong judgment, rigorous research, and increasingly powerful technology work together.

There’s a very big difference.