Of Minds, Machines, and the Curious Business of Thinking
On learning, authorship, intellectual curiosity, and the strange ways we decide whose thoughts are worth considering.
I’ve been thinking a lot lately.
Which, for anyone who knows me, probably ranks somewhere alongside the revelation that water is wet. I have a rather unfortunate habit of finding something interesting in almost everything, and an equally unfortunate inability to leave a particularly interesting thought alone once it has taken up residence in my head. One question becomes three, those three become a dozen, and before long I’ve wandered sufficiently far from the original subject that explaining how I arrived at my present conclusion requires its own lengthy explanation.
This is not a new development. I’ve been doing it for as long as I can remember.
Lately, though, there seem to be an unusually large number of threads competing for my attention. I’m pursuing graduate degrees in computer science and business, navigating the usual complexities of a career in software engineering, pursuing projects of my own, and participating in conversations about education, philosophy, technology, organizational behavior, and the peculiar ways human beings interact with one another.
Somewhere in the midst of all this, I’ve found myself increasingly preoccupied with a question that seems to connect a surprising number of otherwise unrelated observations:
Have we become so concerned with evaluating how people acquire, express, and demonstrate knowledge that we’ve begun to lose sight of what it means to actually contribute something worthwhile?
(I should probably acknowledge that this particular essay may be disproportionately colored by my ongoing coursework in Georgia Tech’s CS6750: Human-Computer Interaction. I’ve noticed that my tendency to examine the world through whatever intellectual lens happens to be most recently positioned beneath my magnifying glass is, apparently, not a particularly temporary condition. At present, that lens happens to involve cognition, mental models, affordances, and the sometimes spectacular disconnect between what people intend, what they communicate, and what everyone else actually understands. I suspect my next course will provide an entirely different vocabulary with which to overanalyze otherwise perfectly ordinary human behavior. Such is the occupational hazard of being perpetually curious.)
The connections between these subjects have become interesting enough that I wanted to explore them together. I suppose I could separate everything into several neatly organized essays, each with a narrowly defined thesis and a tidy conclusion, but that isn’t particularly representative of how I think, and I’m not convinced the ideas would benefit from being separated merely because they happen to cross disciplinary boundaries.
Besides, where’s the fun in that?
When Did Having Something to Say Become Less Important Than How We Say It?
I’ve become increasingly frustrated by a tendency I encounter in academic discussions, professional conversations, and particularly online discourse: the inclination to evaluate the legitimacy of a contribution by examining its provenance, presentation, or conformity to expectations before engaging with its substance.
Someone writes a thoughtful essay, and rather than examining the argument, another person declares that it sounds like something generated by artificial intelligence.
Someone presents an unfamiliar perspective, and the response is to question their credentials rather than address their reasoning.
Someone communicates an otherwise reasonable observation using imperfect grammar, unconventional phrasing, or language that doesn’t conform to the stylistic expectations of a particular community, and suddenly the conversation becomes an assessment of their ability to communicate rather than an examination of what they communicated.
Now, presentation matters. Attribution matters. Intellectual honesty matters enormously, and I don’t mean to imply otherwise.
But I find it troubling how readily we treat those considerations as substitutes for examining the substance of an argument.
A beautifully written assertion can be complete nonsense. A poorly articulated observation can reveal something profound. An individual without formal credentials in a discipline can occasionally identify a problem that people immersed in that discipline have overlooked, just as someone with an impressive collection of academic qualifications can be spectacularly mistaken.
None of this means expertise is irrelevant, or that we should treat every assertion as equally credible.
It means that credibility, truth, understanding, and the presentation of an idea are related considerations, not interchangeable ones.
There’s also a particular kind of intellectual performance that I find exhausting: the individual whose principal contribution to a conversation seems to be demonstrating that the contributions of others are unworthy of consideration.
Criticism is indispensable. If I’ve misunderstood something, overlooked a relevant fact, or arrived at a conclusion that doesn’t follow from my premises, I want someone to point that out. Preferably with enough substance that I can understand the objection and either revise my position or explain why I believe it remains defensible.
But announcing that someone’s work is worthless because of who produced it, which tools they used, or how they chose to express it doesn’t strike me as much of an intellectual contribution.
Sometimes identifying a flaw is itself valuable, even without proposing a solution. We shouldn’t require every critic to produce a replacement argument before they’re permitted to identify a problem.
But criticism ought to involve some actual examination of the thing being criticized.
Otherwise, what precisely has been contributed?
I occasionally wonder whether some of our public conversations have become less about discovering what is true, useful, or interesting, and more about demonstrating who is entitled to be taken seriously.
And I find that a depressing use of our collective intellectual capacity.
The Curious Case of Artificial Intelligence
Which brings me to a subject that has become almost impossible to avoid: artificial intelligence.
Or, more specifically, the increasingly enthusiastic application of the term AI slop to practically anything someone suspects may have been produced with the assistance of a large language model.
Before anyone misunderstands me, AI slop absolutely exists.
There’s an enormous quantity of thoughtless, derivative, misleading, and occasionally completely fabricated material being produced and distributed with virtually no meaningful human involvement. Some of it exists to manipulate search rankings, some to manufacture the appearance of expertise, and some apparently for no greater purpose than demonstrating that generating content has become considerably easier than producing something worth reading.
I’m not particularly fond of it either.
But I take issue with the increasingly common suggestion that the involvement of artificial intelligence, by itself, establishes that the resulting work is intellectually deficient.
The presence of AI in the process does not establish the absence of human thought.
And to understand why I find that distinction important, I think it’s worth considering how authorship and editorial assistance have worked for a very long time.
The Author, the Editor, and the Curious Question of Who Actually Wrote It
Consider the role of an editor.
For centuries, writers have relied upon editors, proofreaders, and literary collaborators to help transform their thoughts into published works. Newspaper editors review articles, challenge assertions, restructure arguments, and occasionally rewrite substantial portions of a journalist’s prose. Book editors work with authors to refine narrative structure, eliminate redundancies, clarify meaning, and improve the overall effectiveness of a manuscript.
A good editor does considerably more than correct punctuation.
They might tell an author that an argument is unsupported, that a passage is unnecessarily complicated, or that an entire section, however elegantly written, doesn’t accomplish what the author intended.
Sometimes they suggest a different approach entirely.
And yet, we generally don’t conclude that a novelist has somehow surrendered authorship of their story because someone helped shape the manuscript, or that a journalist is no longer responsible for their reporting because an editor made substantial changes to the article.
Indeed, thoughtful editorial collaboration is often regarded as evidence of professionalism.
Why?
Because we’ve long recognized that developing ideas, investigating a subject, composing prose, and refining a finished work are related but distinguishable intellectual activities.
An editor can improve how an argument is presented without originating the argument. They can identify weaknesses in someone’s reasoning without becoming the source of the underlying experiences or observations. They can make a manuscript substantially better without necessarily becoming its author.
Of course, there are limits to this distinction.
An editor who fundamentally transforms an argument, introduces major original ideas, or substantially develops the intellectual content may deserve acknowledgment as something more than an editor. Ghostwriting raises additional questions. And anyone familiar with publishing knows that the boundaries between editing, collaboration, and authorship can become complicated.
But that’s precisely my point.
The nature and extent of a contribution matter more than the mere fact that assistance occurred.
If I engage a human editor to help organize my thoughts, identify weaknesses in my reasoning, and improve the effectiveness of my prose, that is generally regarded as an acceptable part of the writing process.
If I engage a large language model to perform substantially similar functions, some people would dismiss the resulting work as intellectually worthless without bothering to consider how much thought I contributed.
Now, human editors and language models are not interchangeable. A human editor brings professional judgment, lived experience, accountability, and an understanding of meaning that we shouldn’t casually attribute to a machine. Language models also introduce their own problems, including fabricated information, misleading confidence, and the possibility of producing persuasive material whose apparent intellectual substance exceeds that of the person presenting it.
Those differences are important.
But if the objection is simply that receiving assistance with composition or revision invalidates authorship, then we have a substantial historical inconsistency to explain.
Nobody seriously believes that every published novelist, journalist, or academic personally originated every editorial improvement appearing in their finished work.
Why, then, should we suddenly demand such a standard merely because a new kind of editorial assistant happens to run on computers?
But Isn’t Writing Itself a Form of Thinking?
There’s a stronger objection to my position than simply declaring that using tools constitutes cheating, and I think it deserves serious consideration.
Writing isn’t always the transcription of thoughts we’ve already completed.
Sometimes writing is how we think.
Anyone who has attempted to articulate a complicated argument knows that something can seem perfectly reasonable while it remains comfortably abstract in our head, only to fall apart spectacularly when we attempt to put it into words.
The process of finding those words can expose assumptions we didn’t realize we were making. It can force us to confront contradictions, discover relationships between ideas, or recognize that an argument we thought we understood was less developed than we imagined.
That friction is valuable.
And there’s a legitimate concern that delegating too much of the writing process to a language model could allow someone to bypass precisely the intellectual work that would otherwise have revealed those weaknesses.
I don’t dismiss that concern.
In fact, I think it’s one of the most important questions surrounding AI-assisted writing and education.
But I also don’t believe productive intellectual friction is confined to the solitary act of composing sentences.
Consider a conversation in which someone challenges your reasoning.
You propose an interpretation. They disagree. You attempt to explain yourself more clearly, discover that they misunderstood your original point, and in doing so recognize that your original explanation was incomplete.
Perhaps their objection reveals a flaw in your reasoning. Perhaps your response reveals a flaw in theirs.
Either way, the process can improve your understanding.
I’ve experienced this repeatedly while working with language models.
I might begin by explaining an idea in my usual fashion, which is to say with considerable enthusiasm, several qualifications, and enough parenthetical observations to occasionally lose track of the sentence I originally intended to finish.
The model responds with an interpretation or proposed structure.
Sometimes it’s useful.
Sometimes it misses the point completely.
And sometimes it presents a remarkably confident explanation of something I know to be wrong.
At that point, I have to decide whether the problem lies in my original explanation, the model’s interpretation, or the reasoning underlying the idea itself.
That’s intellectual work.
I’m not suggesting that interacting with a language model necessarily produces the same kind of reflection as writing independently, or that the two processes are always equivalent.
They aren’t.
But the distinction worth preserving is between productive intellectual effort and effort that adds little to the quality of our thinking, not simply between writing with assistance and writing without it.
Sometimes the difficulty of finding the right words is where the thinking happens.
Sometimes that thinking happens when someone challenges the words you’ve already found.
And sometimes it happens when you realize that the beautifully structured argument offered by your editorial assistant doesn’t actually say what you believe.
Of course, there’s a boundary here, even if it isn’t always easy to identify.
Someone who develops an argument independently and asks for editorial assistance is doing something different from someone who begins with a vague intuition and develops the argument through extended collaboration. Both differ again from someone who requests a finished essay, finds it convincing, and publishes it without meaningful examination.
I don’t believe the second process is necessarily less intellectually legitimate than the first. Much of what we learn, and indeed much of what we eventually come to believe, emerges through conversations in which other people contribute ideas we hadn’t previously considered.
But understanding an argument, agreeing with it, taking responsibility for it, and having originated it are not the same things. We ought to be capable of recognizing those distinctions without assuming that collaboration invalidates authorship or that merely approving someone else’s work establishes originality.
And perhaps that’s where intellectual honesty becomes especially important: not in proving that nobody assisted us, but in being truthful about what we contributed, what we learned from others, and what we’re actually claiming as our own.
The important question is whether we’re still doing the work of understanding, and whether we’re representing that work honestly.
What My Own Process Actually Looks Like
I use large language models regularly, and I don’t find that particularly embarrassing.
I also don’t use them because I’ve somehow lost the ability to read, write, research, or think.
Anyone who has received one of my longer messages can attest that I’m perfectly capable of producing an intimidating quantity of prose without technological assistance.
My problem has rarely been a shortage of words.
Quite the contrary.
I have a tendency to explore ideas while I’m expressing them, which means an observation that might reasonably occupy a paragraph can become an extended examination of its implications, exceptions, historical context, and relationship to three other subjects that seemed relevant at the time.
Sometimes that’s useful.
Sometimes it means I’ve written several pages before realizing that the question I originally intended to answer is still waiting patiently somewhere near the beginning.
Working with an AI assistant can help me organize those ideas, recognize where I’ve been unclear, identify assumptions that deserve examination, and structure the resulting argument into something another person might reasonably enjoy reading.
But I don’t simply accept the first result.
I question it, correct it, reject suggestions, revise interpretations, and occasionally argue at considerable length about why a particular conclusion doesn’t follow from what I originally said.
And that brings us to an amusing example involving the very essay you’re reading.
During its development, I worked with more than one language model. One helped me assemble and organize the ideas, while another critiqued the resulting draft.
The criticism was useful. It identified repetitive rhetorical patterns, areas where the argument needed stronger opposition, and places where concrete examples would accomplish more than abstract observations.
It also suggested that because I mentioned my recent HCI coursework, I ought to apply HCI concepts more explicitly throughout the essay.
Which sounded reasonable until I considered what the essay was actually trying to accomplish.
My coursework influenced the conversations from which the essay developed, but I wasn’t attempting to produce a formal HCI analysis of intellectual discourse. The course was part of the intellectual environment, not necessarily the subject.
Interestingly, the first model initially agreed with the second model’s criticism.
I disagreed with both of them.
We discussed it, clarified the distinction, and revised the direction accordingly.
Now, I recognize that none of this proves the resulting essay is profound, original, or even particularly good. Those are separate questions.
But I do find it difficult to reconcile that process with the assertion that using AI necessarily involves surrendering one’s capacity for independent thought.
If anything, having two systems offering competing interpretations creates additional opportunities to exercise judgment.
Neither model gets to determine what I believe.
Neither gets to decide what I intended to communicate.
And neither is entitled to my agreement merely because its response happens to sound convincing.
(Although watching two language models effectively critique one another while I adjudicate the disagreement has been entertaining. Somewhere in this process is a joke about peer review, but I suspect the academic community has enough to contend with already.)
Provenance, Attention, and the Problem of Too Much Content
There is another argument against AI-generated writing that deserves more consideration than it often receives in these discussions.
We have limited attention.
There is already vastly more information available than any individual could possibly consume, and generative AI has dramatically reduced the effort required to produce additional material.
When readers encounter an unfamiliar article, essay, or commentary, they must make judgments about whether it deserves their time.
Provenance can be a useful signal.
If a publication has a history of careful editorial review, if an author has demonstrated expertise, or if a piece includes verifiable sources and transparent methods, those characteristics can help a reader decide whether further attention is warranted.
Conversely, discovering that material has been automatically generated in large quantities, without meaningful oversight or verification, might reasonably lower one’s expectations.
That’s not necessarily intellectual gatekeeping. Sometimes it’s a practical response to information overload.
But here’s the difficulty.
A useful heuristic is not the same thing as a reliable conclusion.
If someone begins treating particular stylistic characteristics as proof of AI involvement, and AI involvement as proof of intellectual emptiness, they’ve compressed several separate judgments into one.
An essay might exhibit characteristics commonly associated with AI-generated prose because it was substantially generated by a language model.
Or because it was written by a human and subsequently edited with AI assistance.
Or because the author naturally favors formal transitions, parallel constructions, and carefully structured arguments.
Or because a human editor encouraged exactly those stylistic choices.
The visible characteristics of the finished work don’t necessarily reveal the intellectual process that produced it.
And even when the involvement of AI is known, the extent and nature of that involvement remain relevant.
I don’t expect readers to investigate the composition history of every article they encounter.
That’s neither realistic nor desirable.
But there’s a difference between deciding that something isn’t worth your limited attention and declaring that it has no value because of assumptions you’ve made about its creation.
The former is a personal judgment about how to allocate your time.
The latter is a claim about the intellectual merit of someone else’s work.
And I think the distinction matters.
Education, or the Performance of Being Educated?
My recent experiences in higher education have made these questions particularly interesting because education provides another environment in which we attempt to infer something meaningful from observable results.
I genuinely enjoy learning.
Not merely accumulating credentials, satisfying academic requirements, or achieving particular grades, although I certainly recognize the practical significance of those things.
What I enjoy is encountering ideas that force me to reconsider how I understand the world.
Human-computer interaction has been especially interesting in that regard, partly because it provides theoretical language for problems I’ve encountered throughout my career as a software engineer.
Why do people struggle with interfaces that seem perfectly logical to their designers? Why do relatively minor design decisions cause disproportionate frustration? Why do we so often blame people for failing to use systems that were supposedly designed to serve them?
And perhaps most importantly, how do we distinguish between what a system was intended to accomplish and what people actually experience when attempting to use it?
Those questions occasionally follow me out of the classroom.
(One might reasonably argue that they have escaped entirely and are now roaming unsupervised through the rest of my intellectual life.)
But my broader concern with education predates this course.
We have developed an elaborate collection of mechanisms for measuring and certifying learning: examinations, assignments, rubrics, participation requirements, standardized assessments, learning management systems, and increasingly sophisticated methods of monitoring whether students completed their work in the prescribed manner.
Many of these mechanisms are necessary.
I wouldn’t want a physician who had never demonstrated medical competence, an engineer who couldn’t apply the principles of their discipline, or a pharmacist who lacked the knowledge necessary to dispense medication safely.
Assessment matters.
But assessments are tools intended to measure learning, and like any measurement system, they’re only useful to the extent that they accurately reflect what we care about.
Consider the difference between understanding a concept and recalling its formal terminology.
During my HCI coursework, I’ve encountered situations where I understood the general principle being discussed, could recognize its application, and could reason about the problem it addressed, but struggled to retrieve the precise term or definition expected in an assessment.
I’ve also encountered the reverse: terminology that was familiar enough to recognize without necessarily possessing the depth of understanding required to apply it effectively.
Those are different deficiencies, and they deserve different responses.
Sometimes terminology is essential. Disciplines develop specialized vocabularies because precision matters, and an inability to distinguish related concepts may reveal an important gap in understanding.
But I wonder whether we always distinguish between assessing conceptual mastery and assessing the ability to retrieve a particular formulation under prescribed conditions.
The same question applies to assignments.
A difficult assignment can be valuable because it forces a student to confront unfamiliar material, develop a new skill, or reason through a complicated problem.
But an assignment can also be difficult because its instructions are ambiguous, its administrative requirements are unnecessarily cumbersome, or the tools involved are poorly designed.
I don’t believe those two forms of difficulty are educationally equivalent.
And I certainly don’t believe that enduring unnecessary friction should automatically be regarded as evidence of greater learning.
I also recognize that identifying imperfections in the way we assess learning is considerably easier than developing practical alternatives. A professor responsible for hundreds of students cannot reasonably conduct an extended examination of every individual’s understanding, and educational institutions must contend with legitimate concerns about consistency, fairness, academic integrity, and the cost of assessment.
Those constraints matter, and I don’t imagine that questioning the limitations of an examination somehow resolves the difficulties involved in replacing it.
But acknowledging the practical necessity of imperfect measures doesn’t require us to mistake those measures for the things they’re intended to represent. If anything, it makes understanding their limitations more important, particularly as technology changes what students can produce without necessarily demonstrating what they’ve learned.
Which brings the conversation back to artificial intelligence.
If a student uses a language model to generate answers without understanding the material, that’s a legitimate educational concern.
But if the student uses the same tool to explore alternative explanations, challenge their assumptions, practice applying concepts, or discover deficiencies in their understanding, I see a substantially different activity.
And we ought to be capable of evaluating those differences, even if developing practical, scalable methods for doing so is itself a difficult problem.
The question shouldn’t simply be whether technology made a task easier.
It should be whether the student actually learned what the task was intended to teach.
Which is, perhaps, a more difficult question to answer.
But I’d argue it’s also the more important one.
Expertise Is Valuable. Intellectual Gatekeeping Is Not.
All of this naturally leads into another subject I’ve been contemplating: expertise.
I respect expertise deeply.
I’ve spent much of my life developing technical knowledge in disciplines that reward careful reasoning and frequently punish unfounded confidence. Software engineering, for example, has an unpleasant habit of revealing the difference between a persuasive explanation and a correct implementation.
You can explain with absolute confidence why your system ought to work.
The compiler, runtime, network, or production environment may have a different opinion.
And ultimately, the system’s behavior matters considerably more than how convincingly you described your intentions.
That’s one reason I value evidence, reproducible results, and people who have invested the time necessary to develop genuine expertise.
But I don’t believe expertise grants anyone exclusive ownership of a conversation.
Credentials can provide evidence of preparation. Experience can provide context and practical judgment. Formal study can develop understanding that would be difficult to acquire independently.
None of those things makes someone infallible.
Nor does their absence automatically render someone else’s observations worthless.
I’m something of a generalist by temperament, which means I have interests extending well beyond the boundaries of my professional responsibilities.
Software engineering, computer science, business, organizational behavior, psychology, education, aerospace, philosophy—I’m interested in all of them, among quite a few other subjects.
I recognize, of course, that breadth comes with limitations. No one can develop equal depth in everything, and there is a substantial difference between being interested in a discipline and possessing the expertise required to practice it responsibly.
But I have never understood why curiosity outside one’s established specialty should be treated as something faintly embarrassing.
Some of the most interesting insights I’ve encountered have emerged from recognizing connections between subjects that are ordinarily studied separately.
A concept from psychology may illuminate a software design problem. An observation about education may raise useful questions about professional training. A principle from organizational behavior may explain why a technically sound engineering initiative failed despite everyone’s apparent agreement that it was a good idea.
Those connections don’t eliminate the need for expertise.
They create opportunities to engage with it.
And if I misunderstand something, I’d much rather someone explain why than simply remind me that I haven’t earned the right to discuss it.
I’ve also found that the more I learn, the more frequently I discover that something I thought was an original observation has already been examined by someone else, sometimes centuries earlier.
I love discovering that.
It means someone else has explored the same question, probably with a level of depth or historical context I hadn’t considered, and now I have an opportunity to learn from their work.
I don’t consider that a defeat.
I consider it the beginning of another conversation.
Perhaps one of the stranger habits we’ve developed is treating knowledge as though it were principally something to possess, display, and defend, rather than something to pursue.
As though expertise were a permanent condition one achieves rather than a continuing relationship with a subject.
After decades of working with computers and software, I routinely encounter technologies, concepts, and perspectives that remind me just how incomplete my understanding remains.
I don’t find that humiliating.
It means there’s more to learn.
And I hope that remains true for the rest of my life.
The Trouble With Being Right
Of course, there’s an uncomfortable complication to all of this.
I like being right.
I suspect most people do.
It’s satisfying to discover that an idea you’ve spent time developing corresponds reasonably well with the available evidence, particularly when you’ve invested considerable effort in understanding the subject.
But wanting to be right can become a serious obstacle to learning.
When every disagreement becomes a contest, every correction a personal affront, and every concession an admission of defeat, we stop examining ideas and start defending identities.
And that’s a particularly unfortunate outcome in environments ostensibly dedicated to expanding knowledge.
I’ve also had conversations where the disagreement itself turned out to be more interesting than determining which participant was correct. A recent discussion about affordances in human-computer interaction provides a useful example.
The question concerned whether adding a handle to a door introduces an affordance or merely provides a signifier indicating how the door should be used.
At first glance, this might seem like a fairly narrow terminological distinction, particularly to someone who doesn’t spend much time contemplating the design of doors.
(Which, incidentally, is probably most people. Donald Norman and the HCI community have successfully ensured that I may never look at a door quite the same way again.)
But the conversation revealed why the distinction isn’t quite as straightforward as it initially appears.
One interpretation treats the handle primarily as a signifier, something communicating how an existing action should be performed. Another observes that adding a handle can materially change the actions available to someone interacting with the door, introducing an opportunity to grip and pull that wasn’t previously available in the same way.
Both observations can be relevant, depending on the relationship between the person, the object, and the action being considered.
And the distinction becomes clearer when we examine how affordances have been understood in different theoretical frameworks, including James Gibson’s ecological account and Donald Norman’s treatment of affordances and signifiers in design.
What began as a question about whether a handle changes what a door affords became a discussion about the difference between an action being physically possible and the design communicating that possibility to someone attempting to use it.
What’s particularly interesting is that Norman himself provides an example of the kind of intellectual development I’m trying to describe.
When he introduced affordances to the design community through The Psychology of Everyday Things in the late 1980s, the concept helped designers think about the relationships between people, objects, and possible actions. But the terminology also became a source of confusion, particularly as designers increasingly applied it to visual cues and software interfaces.
Rather than merely insisting that everyone else had misunderstood him, Norman revisited how the distinction was communicated. In his 2008 essay Signifiers, Not Affordances, and subsequently in the 2013 revised edition of The Design of Everyday Things, he gave signifiers a more prominent role in explaining how people discover what actions are available to them.
I find that telling. Here is someone whose work helped shape an entire field, recognizing that the way an idea had been presented was contributing to misunderstanding and refining his explanation accordingly.
That doesn’t invalidate his earlier work, nor does it mean everyone who interpreted the terminology differently was being unreasonable. It illustrates that even established experts can benefit from reconsidering how their ideas are understood, communicated, and applied.
And, returning to the original conversation, it wasn’t necessary for someone to have been entirely wrong for the discussion to be valuable. The useful part was recognizing what each interpretation captured, where the assumptions differed, and why the distinction mattered.
That’s the kind of conversation I enjoy.
Not a competition to see who can recite the correct terminology most confidently, but an opportunity to examine a concept closely enough that everyone involved comes away with a more complete understanding.
And it’s the kind of intellectual engagement I wish we encouraged more consistently.
I don’t need everyone to agree with me.
In fact, I’d probably find universal agreement rather unsettling. It would suggest either that I’d somehow achieved an implausible level of intellectual perfection or, more likely, that nobody was bothering to challenge what I said.
What I want is engagement.
If an argument is wrong, explain why.
If there’s a better interpretation, show me.
If I’ve overlooked something important, I’d genuinely like to know.
That doesn’t mean I’ll necessarily agree with every criticism.
And, for what it’s worth, I don’t exempt myself from the tendency to become defensive or stubborn. I can be argumentative, enthusiastic, and occasionally so occupied with developing my own explanation that I risk overlooking the point someone else is attempting to make.
But I try to recognize that being corrected doesn’t diminish what I know.
It improves it.
The alternative is to become so invested in maintaining the appearance of knowledge that we gradually lose the willingness to acquire more.
And that strikes me as a terrible bargain.
When the Tool Becomes the Measure of the Worker
There’s another dimension to the discussion of artificial intelligence that troubles me, and it has less to do with writing than with the incentives surrounding technological adoption.
I’ve noticed an increasingly enthusiastic narrative in some executive and technology circles suggesting that substantial consumption of AI services is itself evidence of professional productivity or organizational sophistication.
The implication, sometimes quite explicit, is that employees who consume more AI tokens are somehow demonstrating greater value, as though the amount spent interacting with a tool were a meaningful substitute for evaluating the results produced.
I find this peculiar.
If I build a system that satisfies its requirements, performs reliably, remains maintainable, and accomplishes the intended business objective, why should the amount of money I spent on AI assistance be treated as a measure of the quality of my work?
Conversely, if I consume an enormous quantity of tokens producing something that is incorrect, unnecessarily complicated, or unsuitable for its intended purpose, what exactly have I demonstrated?
Certainly not competence.
I may have demonstrated considerable enthusiasm for consuming tokens.
But that’s a different accomplishment.
There’s a familiar principle here, commonly referred to as Goodhart’s law, which originated in economics but has since found application in disciplines ranging from organizational management to education.
The anthropologist Marilyn Strathern, writing about university assessment in her 1997 paper ‘Improving Ratings’: Audit in the British University System, expressed it in its now-familiar form: “When a measure becomes a target, it ceases to be a good measure.”
The observation traces back to the economist Charles Goodhart’s work on monetary policy, although Strathern’s formulation, building on a discussion by Keith Hoskin, generalized the principle well beyond its original economic context.
I particularly enjoy the intellectual lineage of that observation. An economist identifies a problem involving monetary indicators, the idea finds new relevance in discussions of accountability and higher education, and here I am applying it to software engineering and corporate enthusiasm for artificial intelligence.
Apparently, I’m not the only person inclined to let ideas wander across disciplinary boundaries.
I’ve encountered variations of this problem throughout my professional life.
Lines of code are not a reliable measure of software engineering productivity. Hours spent at a desk don’t necessarily correspond to meaningful output. The number of meetings attended tells us remarkably little about the quality of someone’s contribution, and the number of completed tickets may tell us more about how work was divided than about how much useful work was accomplished.
Treating AI-token consumption as evidence of professional value strikes me as another manifestation of the same problem, albeit one with a particularly enthusiastic marketing department.
The measure becomes the objective, while the outcome it was supposed to represent receives less attention.
There’s also a more serious concern.
If organizations increasingly require employees to use proprietary AI services, and then begin treating access to those services as part of the compensation equation, they risk creating an uncomfortable dependency between professional capability and externally controlled infrastructure.
Models change.
Pricing changes.
Terms of service change.
Capabilities that exist today may be modified, restricted, or discontinued tomorrow.
And questions surrounding ownership, intellectual property, confidentiality, and the long-term implications of AI-assisted creation remain significant.
I don’t particularly relish the prospect of employees being encouraged, or effectively required, to develop workflows so dependent on proprietary services that their ability to perform their own jobs becomes contingent upon continued access to those services.
Technology should expand our capabilities.
It shouldn’t quietly become a substitute for maintaining them.
I’m an enthusiastic user of AI tools, but I have no intention of surrendering my understanding of software engineering, my ability to reason through problems, or my responsibility for the systems I design simply because an increasingly capable model can perform portions of that work more efficiently.
I want the benefits of the technology.
I also want to understand what I’m doing.
And I don’t believe those objectives are in conflict.
Which brings us back to the distinction between the process and the outcome.
Whether we’re evaluating an employee, a student, an author, or an engineer, the tools involved may be relevant to understanding how something was accomplished.
But the fact that someone used a tool, or consumed a particular quantity of it, cannot substitute for evaluating what they actually contributed.
The Universe Is Entirely Too Interesting for This Nonsense
Perhaps the strangest part of all this is how much energy we devote to policing the boundaries of intellectual participation when there’s so much left to understand.
We live in a universe of almost incomprehensible complexity.
We’ve mapped distant galaxies, investigated the fundamental structure of matter, developed technologies that would have seemed indistinguishable from magic to earlier generations, and accumulated more accessible information than any individual could reasonably hope to absorb in a lifetime.
And we still don’t fully understand consciousness.
We continue to debate the nature of intelligence, the complexities of learning, the foundations of human behavior, and the ways our institutions shape the people who participate in them.
Even technologies we use every day contain layers of complexity that most of us rarely stop to consider.
There is so much to learn, investigate, and build.
And somehow, amid all that possibility, we frequently find ourselves arguing over whether someone used the appropriate tool, possessed the expected credential, adopted the preferred tone, or endured a sufficient amount of inconvenience before expressing an idea.
I find that deeply disappointing.
Not because I believe every idea deserves acceptance. Quite the contrary. Some ideas are demonstrably wrong, some are poorly supported, and some are dangerous precisely because they’re presented with more confidence than the evidence warrants.
Those ideas deserve scrutiny.
But scrutiny ought to involve examining the reasoning, the evidence, the assumptions, and the implications.
If someone’s argument is wrong, identifying the error is valuable.
If it’s incomplete, explaining what’s missing can move the discussion forward.
And if it turns out to be useful, perhaps we could acknowledge that fact without first requiring the author to demonstrate that they arrived there through a process we personally approve of.
I sometimes think about how remarkable it is that a person can devote an entire lifetime to learning and still encounter new questions almost every day.
I have ambitions that are probably unreasonable by most conventional standards.
I’d like, for example, to travel to the Moon, land there, and return safely to Earth in a spacecraft of my own design and construction.
Preferably alive.
(I feel that last qualification deserves explicit mention. One-way lunar travel is a different engineering problem, and not one I’m particularly interested in solving through personal experimentation.)
I recognize that this is an ambitious undertaking, to put it mildly.
But the ambition itself reflects something I’ve believed since childhood: the boundaries of what we might accomplish are worth investigating, even when the practical obstacles appear overwhelming.
I don’t expect to reach the Moon by pretending I already possess everything necessary to get there.
I’d have to learn an enormous amount from people who know things I don’t.
I’d have to study unfamiliar disciplines, revisit assumptions, test ideas, make mistakes, and undoubtedly discover that some of my initial designs were spectacularly impractical.
I’d also need tools.
A great many tools.
Computer-aided design systems, simulation software, mathematical models, manufacturing equipment, measurement instruments, and probably technologies that don’t yet exist in forms suitable for what I’d want to accomplish.
Some of those tools might incorporate artificial intelligence.
Others might not.
I can’t imagine dismissing the legitimacy of the endeavor because I used software to perform calculations, simulations to evaluate engineering alternatives, or an AI assistant to help investigate concepts outside my existing expertise.
The responsibility for understanding the design, validating its assumptions, evaluating its safety, and making the relevant decisions would remain mine.
And I suspect the Moon would be indifferent to whether I calculated a particular trajectory with pencil and paper, a conventional computer, or some future system whose capabilities I can only imagine.
The spacecraft will either perform as intended or it won’t.
The physics involved has no interest in my credentials, the tools I used, how eloquently I explained my design, or whether someone thought my documentation sounded suspiciously like it had been edited by a language model.
Physics, after all, has a rather unforgiving way of evaluating the merits of an argument.
And I think there’s something worth remembering in that.
Not every human endeavor has the same kind of objective standard, of course. Literature, philosophy, art, and public discourse involve interpretations and judgments that can’t be reduced to whether a machine successfully reaches its destination.
But the underlying principle remains useful.
At some point, we have to evaluate what someone actually accomplished.
Not merely what we assume about how they accomplished it.
A Final Thought (Which, Knowing Me, Is Rarely the Final Thought)
I don’t imagine any of these observations will resolve the debates surrounding artificial intelligence, education, authorship, expertise, or the quality of public discourse.
And I certainly don’t imagine I’ve arrived at some uniquely original philosophical revelation.
What I hope, more than anything, is that we can preserve a little intellectual generosity.
The willingness to distinguish disagreement from hostility, assistance from substitution, expertise from authority, and genuine curiosity from the performance of knowing.
The willingness to consider that someone might have arrived at a worthwhile observation by a route different from the one we would have chosen.
And perhaps the humility to recognize that even our most carefully developed understanding of the world remains incomplete.
I decided when I was seven years old that I wanted to become a software engineer. I had some fairly specific ambitions about my education, the work I wanted to do, and the kind of things I hoped one day to build.
Decades later, I’ve accomplished a number of those things, encountered challenges I never anticipated, changed my mind about subjects I once thought I understood, and discovered interests I couldn’t possibly have imagined when I was seven.
And I’m still learning.
I’m still finding things I don’t know, discovering ideas that challenge my assumptions, and occasionally getting carried away explaining why something I encountered in one discipline has interesting implications for something entirely unrelated.
I consider that a wonderful state of affairs.
Because the objective, at least as I’ve always understood it, was never to reach a point where there was nothing left to learn.
It was to spend a lifetime discovering just how much there is.
And if I happen to use a book, a professor, a colleague, a search engine, an editor, or an artificial intelligence to help me along the way, I suspect the universe will survive the indignity.
After all, the point was never to prove that I could think without assistance.
The point was to keep thinking, to continue developing my understanding, and perhaps, if I’m fortunate, contribute something worthwhile along the way.
(And yes, for anyone wondering, this essay was developed with the assistance of artificial intelligence. More than one, actually. The process involved conversations, drafts, criticism, disagreements, revisions, and occasional corrections of the models themselves. Some of the language was suggested by those systems, some was revised in response to my objections, and the underlying ideas emerged from experiences and conversations I’ve been having for quite some time. You’re welcome to challenge any of the conclusions, and I’d genuinely enjoy discussing why you disagree. But if your principal objection is simply that I used tools to help organize and communicate thoughts I could have written out independently, albeit with more effort, then I suspect we’ve arrived rather neatly back at the point of the essay.)
TheIrishGeek — Engineer by profession. Geek by nature. Student of just about everything.
Keep learning. Keep questioning.