What We Never Get To

There is a part of every AI training session I deliver that I refer to as "The Good Bit." This is the time when we stop talking about policy and actually do something. It is the part for which I prepare, practice and in many cases stay up until midnight in my efforts to perfect it. Importantly, it is the part we never get to experience.

The Policy Ceremony, or: How Fifty Minutes Become Eleven

The session is scheduled for 2 hours. In the planning phase, this represents an embarrassment of riches, an excess of time and a complete spectrum of possibilities. During the preceding week, I imagine how I will use those 2 hours. I plan it with the precision of an architect: There will be a 15-minute period of conceptual development, 30 minutes of interactive exploration, an opportunity for participants to actually try the activity and experience with almost exhilarating satisfaction that it actually works. Finally, there will be an opportunity for which I have planned specifically, for someone in the back row to exclaim, "Oh, wait a minute!" in a manner that indicates that the penny has finally dropped.

But what actually happens is that policy arrives first and remains seated in the centre of the room without moving.

Policy is not a person, but by the end of the session has developed a personality that is highly conservative, extremely repetitive, intensely concerned with issues of data sovereignty and legally incapable of producing summaries. The resulting acceptable use policy is presented in 12 minutes, and discussion of data classification requires 20 minutes because someone (and here I can only express professional enthusiasm) always asks whether his or her e-mail signature represents sensitive information. This leads to an ongoing discussion of the extent to which AI applications will compromise the principles of academic integrity and thereby modify our definitions of plagiarism in the information age. Finally, we receive the same question at the conclusion of each of our 14 consecutive sessions: "Will our use of AI result in loss of our jobs?"

The answer to this question is never a brief one. In principle, it should be. The more time I spend thinking about AI training, the more forcefully I am reminded of my mortality and the transient relevance of what I do. Rather than being a question about AI at all, it ends up consuming every remaining minute of my time with that question, because it is ultimately a question about human expertise in real time and a manifestation of deep anxiety about education in a rapidly renegotiating relationship with expertise.

I have tried providing brief answers, detailed answers and even the "let's set that aside for now and return to it later" strategy that all of us know really means "we will never return to it." Nothing works. The question of jobs represents the tidal force of every session of AI training; it pulls all other considerations toward it and releases its grip only when the room is beginning to empty.

By the time I've worked through policy, resolved jobs, clarified that no, participants may not request that an AI compose a complete annual performance review and sign it (a uniquely memorable event about which I will say no more). And after experiencing the 12-minute low-stakes, highly generative warm-up activity that I designed, I have only 11 minutes remaining.

Eleven minutes. With my lesson plan running silently in the background like an optimistic daemon, I quietly close the tab.

What Lives at Minute 47, and Why It Matters That Nobody Sees It

Here, then, is what constitutes the best part of my session. Because someone must know that it exists.

It is the time when I stop describing what AI is and begin using it to accomplish some novel and highly ridiculous task that nonetheless illustrates an important principle. It is the moment when I provide a large audience with an extremely bizarre request and observe the response with them. I do not describe the experience, but simply allow the audience to respond. Finally, it is the experience of answering a question such as, "Can it do that?" with a reply of, "Let's find out." In doing so, I experience the excitement of a shared experiment that may or may not succeed.

This represents, frankly, the purpose of my entire session. My use of policy is primarily instructional. The discussion of jobs is important but certainly not terminal. My experience with the warm-up activity is largely infrastructural. Ultimately, it is my experience of learning to use these tools with a sense of genuine curiosity, rather than merely fulfilling an obligation of occasional use, that constitutes the most important goal of my session. It is also the experience of learning to exploit the potential for unexpected results that is inherent to the use of AI. Finally, I learn to treat a request for service not as a search for information, but as the beginning of an unstructured conversation with no pre-specified destination.

In the end, I learn that the box for requests of service is primarily a stimulus to the consequences of my imagination. That the quality of my requests determines the quality of my experiences with these tools. To use AI as responsibly as possible, students must understand a fundamental principle: "Boring in, boring out." Conversely, to motivate students to use AI in their future work, they must appreciate that "interesting in, interesting out" provides an equally important rationale for responsible use.

I fail to achieve even the most basic objectives. With only 11 minutes of class time, I have sufficient time to display three slides and invite questions. There are always questions about policy, however, and this makes it impossible to achieve my other objectives.

In Which I Am Forced to Confess My Only Real Excel Achievement

I have been asked to include Excel in my training delivery. Basic Excel, practical Excel and Excel for those who use it for work in the manner that ordinary people would: to track budgets, organise schedules and produce clean, functional worksheets that convey competence without requiring any internal effort.

The problem is that Excel bores me. I am not proud of this fact. It is a powerful tool with appropriate applications for almost every professional activity. Many highly competent individuals regard Excel as a true skill for which they have considerable pride. But I am completely unable to generate any sense of involvement with the product, and entirely blame myself for this condition.

My last experience of excitement about a spreadsheet was with the development of a generator for Persian sentences. Let me explain. The word order of Persian is exceptionally regular and deterministic. The rules of grammar are so well defined that the system, in principle, should be amenable to complete and systematic representation. The verb always appears at the end of the sentence, and the object precedes it. This results in a subject-object-verb structure of extraordinary elegance. Consequently, I constructed a spreadsheet of remarkable complexity with validated lists of subject pronouns, of verb forms selected from a dictionary of Persian verbs to which I had previously had access, and of cell colours indicating different tenses. The final formula for assembly of the complete sentence produced grammatically correct output in the appropriate direction of right-to-left display.

The results were absolutely magnificent. I used the system only once to demonstrate it to a colleague who expressed polite interest and then abruptly diverted the conversation. Since that time, the spreadsheet has remained in a file on my desktop as a permanent reminder of the experience of providing a person with a background in linguistics with a copy of Excel and complete freedom of a weekend afternoon. The point is that when I attempt to generate enthusiasm for teaching someone to apply a SUM formula to column B, some part of my brain signals this as an enormous step back from my Persian verb dictionary. Which is, of course, a completely irrational reaction. Standard training in Excel does not require knowledge of a Farsi vocabulary, nor is it necessary to know that my personal experience with spreadsheets was at its peak in a fugue state of linguistic enthusiasm at 11 on a Sunday night. That is a problem for me to solve.

My solution was, appropriately and professionally, and with the full exploitation of AI as an authentic component of training, to request that Claude assist me in developing a Train-the-Trainer course for Excel.

And here things become complicated.

Saturday Night and the Ozark Pivot: A Story About Prompt Engineering

I started with a reasonable brief: I needed training materials that were interesting, made pivot tables seem consequential and VLOOKUP seem like a valuable skill. In other words, I was not at all bored with the concept of Excel, but profoundly bored by the absence of real stakes. Conventional Excel training represents a disaster of epic proportions: No one cares whether a SUM function is correct, because the fictional budget used in the exercise is obviously fictional and well understood by all involved.

What would happen if the budget were not fictional? What if the resulting level of concern were substantially greater than concern for the appearance of accuracy in quarterly variance figures?

My thinking was entirely in the shoes of Marty Byrde.

What followed was about three hours of the most intense spreadsheet-based teaching experience of my professional career. I produced a complete manual, with learning objectives that resembled threat assessments. My case study involved a lakeside resort with inexplicably strong cash reserves and a payroll that, upon close inspection, did not reflect actual staffing levels.

Exercises in conditional formatting were specifically designed to identify discrepancies prior to (as the manual delicately put it) "discovery by external stakeholders." My tutorial on the use of VLOOKUP focused on the problem of reconciling two slightly different versions of the same invoice. The resulting spreadsheet work was technically flawless, with fully correct formulas and excellent instructional support. Learning objectives were clearly defined, measurable, achievable, relevant and limited to a specific time period in a manner that my usual experience with spreadsheet instruction had never achieved, because of the tremendous advantage provided by the organisational context of criminal activity.

In addition, I developed and administered a quiz based on a darkly styled, Ozark-themed self-assessment. Questions included, for example: My analysis of variance for Q3 revenues from lakeside accommodations revealed a $47,000 discrepancy. I have 90 minutes before my meeting.

Which course of action is most appropriate? All response options were functions of Excel, and the correct response was unquestionably valid. Importantly, the entire clinical experience was completely fictitious.

My efforts to achieve a highly attractive visual design resulted ultimately in production of a complete, fully polished resource for educational purposes that I will certainly never be able to implement in any professional setting, because description of my clinical teaching experience would require me to refer to a workplace in which the words "money laundering scenario" were obviously inappropriate. Nonetheless, I am recording this experience as representing substantial professional development.

A Diagnosis, of Sorts: What Imposter Syndrome Does When It Has a Prompt Box

I've been thinking about what this weekend revealed, and finally come to a conclusion that is reassuring to some and deeply disturbing to others, depending on one's relationship to the concept of professional identity.

Standard impostor syndrome is paralysing. It is the voice that says you are not qualified, not experienced enough, not the right person for the room. It has caused more loss of confidence among excellent teachers than I care to count. As a result, it leads to avoidance of opportunities to contribute to panels, to submit abstracts or to participate in meetings in which others with greater credentials describe your field with complete confidence.

But what we have discovered is that when given access to a language model and left to operate unimpeded over a weekend, impostor syndrome produces results that are entirely different. It becomes feral.

Removal of social barriers to expression eliminates the inhibitory effect of the impostor voice. There is no panel to avoid, no audience for whom claims to expertise are not justified, and simply you, your blank input screen and all of the freedom of your actual interests. These represent experiences that are vastly more interesting and creative than those implied by your lesson plans. Evaluation of your performance is unnecessary, because no one will ever see what you type. No one will judge the appropriateness of an Excel-based dictionary of Farsi verbs. And no one will question the degree of institutional appropriateness of a description of money laundering in the Ozarks as a vehicle for teaching the use of VLOOKUP.

Ultimately, you simply do the interesting things. And these things, we finally discover, actually work very well.

This is, I believe, the experience that I continue to attempt to provide to participants during those final 11 minutes, and ultimately fail to achieve, because the ceremony of policy implementation consumed the remaining time. Not the tools, not the functions, not even the formats of the prompts, although these certainly were important. Instead, I attempted to convey the experience of typing an input of unexpectedly great interest and receiving in return a response that was truly remarkable in its fidelity to that input. The thing they find boring can become interesting if treated as interesting. Expertise is not a static credential of which you either have or do not have; rather, it is a process of continuous curiosity and sufficient specificity and boldness to pursue leads wherever they lead.

As a result, I am a more effective practitioner on a Saturday night than on a Tuesday afternoon when addressing an audience. The experience of carefulness in a training session provides a useful antidote to my experience of being totally uninteresting to an audience of zero using a laptop in the setting of a Farsi spreadsheet with no perceived need for a result.

Finally, I am not certain what to do with all this information. I will record it and hope that is a beginning.

The Part We Never Get To

On Monday, I'll provide a session for which the acceptable use policy will arrive promptly and sit in the front row. The jobs question will follow as usual, with the inescapable timing of a familiar character who recognises his or her essential role in the story even though he or she is not strictly scheduled to appear.

We will reach 11 minutes. I will observe my lesson with the expression of someone watching an excellent meal passed by one's table for consumption by another.

And I will think about the pivot tables from the Ozarks, displayed on my desktop with the appearance of classified material inadvertently formatted to be extremely attractive. I will think about the conditionally formatted cells of the Persian verb dictionary. And I will consider what would happen if we had an additional hour of time, occasionally and only for the purpose of demonstrating that when we ceased to be professional about the work, truly interesting and unexpected things could happen.

That would be very pleasant. One of the people in the back row who had spent the session politely nodding to the acceptable use policy and waiting for something that felt real would undoubtedly enter an extremely unusual value in the box and experience almost complete surprise that the box had accepted this value seriously.

That was the point at which our experience of the work changed completely.

With 11 minutes of time, we will attempt this experience again at the next session.