Category: Uncategorized

The danger is not in not using AI, but in ignoring AI.

AI is everywhere, and every organization is asking whether and how it should be used. Some companies have gone so far as to mandate AI usage, and AI proficiency is becoming a skill new hires are increasingly quizzed on.

The current form of AI has the potential to change society in much the same way that the personal computer and the internet did. In the context of education, it has the potential to change how and what we teach, and what it means to prepare students for a life beyond K-12.

AI has been slowly making its way into professional workflows in subtle ways to enable individuals to do more or work more efficiently. For example, radiologists are increasingly using AI tools in medical imaging to assist in detecting abnormalities. And unquestionably, some of the AI tools that have appeared in the past few years are game changers. We can mine the internet with natural language using ChatGPT – when it doesn’t hallucinate. Generative AI tools are allowing creators to create amazing imagery and video – and causing some consternation among Hollywood actors. Self-driving cars can give us a new level of freedom of getting from A to B – though their robotaxi incarnations are presently causing traffic gridlock when they get confused.

As we rush to realize grander ambitions for AI, we’re seeing it is a double-edged sword. And because the technology it’s built on has different properties than the computation and automation we’ve become accustomed to, we don’t always assess its harms until the sword is descending upon us.

Looking to the near future, agentic AI opens the door for AI tools to act autonomously – today with baby steps. But steps that will no doubt lead to greater strides as agents become more capable and we become more accustomed to working with them. And artificial general intelligence – AGI – the point at which AI exceeds human “intelligence” in all areas – was once thought to be a theoretical fantasy, then considered achievable perhaps in the 2030s, but some are now claiming it will be reached in the next few years. And in many respects, we’re just out of the starting gate. AGI in particular creates a backlash driven by a crisis of worth in a humanity that has so long prized itself on “intelligence.” This is in many ways an echo of the same crisis caused by computers, and will no doubt result in the same backlash in shunning these new technologies and emphasizing their faults while overlooking their benefits.

With all the buzz and promise, it seems odd to ask What if we don’t use AI? But because AI is a new and rapidly evolving area, thinking through What if we don’t use AI? is just as important as How should we use AI? and Where should we use AI? and work. And circling these questions is: What should we be doing?

Backgrounder: What Is AI?
Backgrounder: What Are AI Tools?

Not your typical advancement

In the computer age, we’re used to tools which allow us to do things more efficiently. They may be faster, improve accuracy, or offload some of our work. As computer programs, they work in a defined way and (more or less) do what we expect them to do.

AI tools fall into another class. They can seemingly imbue their users with new skills. For example, in TV studio environments, it was typical for producers (who may be artistically minded but challenged to sketch anything) to work with artists to create artwork. The producer described what they wanted, the artist sketched, and there was a lot of back-and-forth. Now we are seeing producers work with generative AI tools to produce preliminary artwork. At least today, these are not production-grade. The AI tool has not turned the producer into a production artist. But it has turned the producer into a junior artist. The producer’s artwork is shared with the artist for refinement. It cuts down the back-and-forth. It also means that artists are now expected to work at a higher level – they’re no longer doing a lot of quick and dirty sketches, but are instead transforming the producer’s draft into a production piece that “speaks” to an intended audience (something which the producer may not be capable of even with the help of AI). AI tools are like cognitive prosthetics. They upend the playing field – individuals who learn to use these tools gain an advantage and may outperform those who do not . The lines between roles become blurred, resulting in positions being created, eliminated, and altered.

The need to learn a new class of tool, and the resulting changes it can cause, may be exciting to some but also anxiety-ridden for many. Fortunately, because AI is a young and rapidly evolving field, the risks of not making immediate use of it are low. In fact, one can take a cue from Apple, which, rather than rushing to spend vast sums of money on AI and data centers, has taken a more measured approach. Apple has rarely been at the fore of providing bleeding-edge technology in its products. It is usually elegantly late, with much thought and care given to how technologies are expressed in its products.

If AI is not used in education, at one level, nothing will happen. The status quo will be maintained. Teachers will continue to teach, and students will continue to learn. The risks become more apparent down the road, as tools mature and other school systems are reaping the rewards of adopting them. Until then a measured approach to adoption will give teachers (and students) time to become familiar with AI tools and the how they can be integrated into existing methods of instruction.

AI Fluency and Proficiency

Even if AI ends up not being used, both teachers and students need to be able to be fluent in it – they need to be able to discuss it with peers and those who are experimenting with AI tools, and to be able to understand how the field evolves. AI comes with a whole new set of terms and concepts. Some, like “AI hallucinations” are becoming mainstream. Others like “context window” and “attention mechanism” remain mostly in the technical realm, but may pop up in non-technical literature. Understanding them allows us to work better with AI.

In addition to being fluent in the language of AI, some level of proficiency should also be developed. Teachers should know how AI tools in general work and what they are capabale (and not capable) of, regardless of whether they end up using them. For example, understanding how to use or “talk” to an AI, and that you should not implicitly trust its responses.

At an organizational level, everyone requires some level of AI fluency and proficiency. Everyone needs to be able to meaningfully participate in discussions about AI tools and their potential uses.

But an organization’s decision-makers and thought-leaders also need access to a deeper level of knowledge.- a level that cannot be achieved by book learning but is gained by hands-on use of AI tools – with direct experience in the successes, pitfalls, and outright failures that result. This level of knowledge is particularly valuable given AI’s novelty and the rate at which it is changing. Decision-makers and thought-leaders don’t necessarily need to possess this knowledge themselves, but they do need trusted individuals who can provide it to them.

In high-tech corporate organizations, the Office of the Chief Technology Officer or equivalent may provide this sort of direction. But in other organizations, the same knowledge can be obtained by allowing a few individuals comfortable with walking the bleeding edge to gain a deep understanding of what AI can do. Their experiences can then shape AI usage across the organization. Their deep fluency imparts an understanding of not only what an AI can and cannot do for the organization, but also why. This deeper level of understanding allows them – and through them their leadership – to extrapolate how AI tools might be used in other areas, how they might be adapted and improved, and to engage with those developing tools.

The risks of not developing fluency and proficienc in AI are subtle and pervasive. AI changes the way we interact with computers (as evidenced by ChatGPT), the way we search for answers (again ChatGPT), and how tasks can be automated (the emerging field of agentic AI). Interacting with an AI is not like interacting with a typical computer program. You need to know how to talk to it. You need to know when (and to what degree) you can trust its answers, and when to call a result a hallucination. You need to know where AI is useful, and where it is not. And the rapid pace of advancement means that the answers to all of these questions are very dynamic. A task that an AI could not handle well last year, it could excel at next year. Or, next month.

AI as an enabler

AI tools can enable both teachers and students to become proficient in areas they were not before. A teacher who has difficulty interacting with a certain student directly may find that the student interacts well with a certain type of chatbot teaching aid. The chatbot becomes an intermediary and tool, bridging the two. Similarly, students who are unable to draw but have artistic potential may use generative AI tools to create works and refine their talent. The criterion shifts from “can you wield a brush” to “can you create a piece that conveys your message?” We step from the mechanics of a lesson or task to its core.

In all these cases, AI is filling in a gap. We need to take a step back and look at the big picture. It’s true that a student or teacher who relies on an AI tool as a crutch will not perform well without that tool. However, if their capacity is fundamentally changed with that tool – and the tool will be readily available when needed – does it matter?

Is a student who uses a calculator proficient in math? The answer has changed over time. Today we use calculators (or computers) routinely and being able to do multiplication without them is simply not a valued skill. Performing mathematical calculations are inevitably part of another skill – preparing taxes, managing inventories, or balancing checkbooks. We value those higher-level skills and easily ignore that an accountant proficient in the tax code and finding deductions may be woefully deficient in long division by hand. We trust his tax program (or calculator) renders the lack of that skill harmless.

If we do not use AI tools in education, we lose the new avenues that AI prosthetics open for both teaching and learning. Teachers may not be able to convey lessons in as meaningful a way to each student. A student’s lack of a certain skill may prevent them from developing adjacent or higher-level skills.

New Ways of Learning

Not using AI allows the status quo to be maintained, and no immediate harm is caused. This is a rather comforting position, and many may be tempted to keep doing what they have been doing. But not using AI also means we are not availing ourselves of new ways of teaching.

As an example, chatbots can allow instruction to be performed in a conversational manner rather than textbook study/homework. Conversational learning has advantages over textbook instruction, a primary one being that the student is an active participant in the conversation. Information is retained and incorporated into the student’s mind as it is actively used. For some students, this may result in superior comprehension and retention compared to problem sets or other tools traditionally used to reinforce textbook learning.

Conversation between a student and chatbot also affords the possibility of more personalized instruction. A chatbot can assess a student’s knowledge during conversation. It can then spend more time on aspects of a lesson that the student is unfamiliar with or struggling with, while spending less time on areas the student appears to have a good grasp of.

Instructional AI tools may also be able to tailor instruction for each student. Subject material could be altered to a particular student’s interests while the core of the lesson plan remains the same. Examples, parables, and the “surface topic” of problems may be substituted to be more “interesting” or relevant, making them more meaningful and easily integrated into a student’s knowledge base.

The use of chatbots and dynamic lesson plans would be a novel mode of instruction that might previously be available only in very small classes, where 1:1 interaction between a student and teacher is possible. Even then, the load on a single teacher could become prohibitive.

But despite the potential advantage, hurdles must be overcome. Instructional chatbots must be developed – chatbots that are focused on instruction rather than general conversation. A means of constraining a chatbot must exist so that it has some leeway in how it adapts a lesson plan, but the lesson itself remains what a teacher intended. As with all tools used in an educational environment, chatbots must be constrained from venturing into areas that parents (or general society) might object to. And, a teacher must be able to trust that the chatbot will faithfully render a lesson without hallucinating or otherwise causing harm.

The Fallacy of Scores

Some attention needs to be given to whether adopting AI will help or hurt school systems in their scoring. Goodhart’s Law is particularly relevant here – When a measure becomes a target, it ceases to be a good measure. Tests, scores, and metrics are initially valued because, when they were created, the scores or metrics were well-correlated with some proficiency. The score becomes a proxy for a more thorough evaluation. But the relationship is correlational and not causal.

Goodhart’s Law can be prescient in many ways – the pressure to be measured well means there is a tendency to lose sight of what is supposed to be measured. There may be pressure to game the system so that scores are excellent, yet the proficiencies they are supposed to correlate with are absent.

AI and AI tools are disruptive to existing metrics because they fundamentally change the landscape. The context of the system being measured is fundamentally changed, so tests may no longer correlate well with the proficiency they are supposed to assess. In some cases, the proficiency itself may no longer be valued.

This does not mean metrics should be abandoned, but when AI tools are (or are not) brought to bear, their impact on test scores needs to be evaluated in a larger context. What a test is supposed to measure should be asked, as well as whether that goal is being achieved, even when scores are low.

Overreliance on AI tools.

There are studies and papers making the rounds proclaiming a particular risk in using AI tools – that they erode the skills in human that they themselves provide. In the context of education, teachers may become less proficient at teaching when they use AI tools. At some level this argument is the same alarm sounded regarding the use of calculators to do math and computers for the varied purposes we have put them to. But we live in a world today where calculators are ubiquitous (most likely even on our phones).

However, as with any aid, using AI tools is not without its risks, and we must be cognizant in our use of tools so that we reap their benefits while avoiding their pitfalls.

There is a risk of losing touch with fundamental knowledge if AI tools are used exclusively. The difference is using an AI tool to accelerate a task vs. using it to do something you don’t know how to do. If you know how to do a task, you have some knowledge of the underlying steps and which approach yields the best results. These steps can be time-consuming and require going through vast amounts of information – the types of tasks which AI tools (or regular computer programs) can assist with, so that you can focus on higher-level tasks. But with your knowledge of the task, you can assess if the tool is doing things in an appropriate manner and the quality of the result. Ideally, the tool would proceed in the same manner and produce the same result as you would.

There is a risk of losing touch with fundamental knowledge if you rely too heavily on a tool. Quite simply, our understanding of the domain starts to become dated, and we may forget certain aspects of it. One way to keep our core skills up-to-date is to rely on tools most of the time, but to periodically do things ourselves. In doing so, we remind ourselves of the domain and also confront any changes. We can also periodically do a deep inquiry into how a tool arrived at a result. If the tool took a different approach than we would have – why? Is it in reaction to a change in the domain so that more optimal paths to a solution are available? Did new challenges arise? Is there simply a “different” way of doing things? The latter is an interesting case as it suggests that the tool’s approach is still valid and still produces acceptable results. But as users of the tool, it is an important data point since it likely reflects a trend in the domain that will come up as we interact with peers or other tools.

Our students may be less-prepared for life after graduation

Whether AI is used in education or not, children in school today will graduate into a world of AI tools. It will be taken for granted that they know when and how to use AI tools, much the same way that we expect everyone to know how to send an email or how to use a computer in the general sense.

There’s a question of how to phrase a question or task. In the terminology of AI tools, this is referred to as prompting. It is the specific phrasing used to achieve the best result, as well as managing what background information the AI model should use (or not use). (The latter refers to managing the AI’s attention mechanism and context window. The context window can be thought of as an AI’s working memory. The attention mechanism is a major contributor to what ends up in that working memory.) There’s the question of what the current brand of AI does, and does not, do well. And there’s the question of how you go about verifying the product of an AI tool.

In many respects, interacting with an AI tool can be like interacting with a child – one with peculiar cognitive strengths and deficits. As AI models evolve, the strengths and deficits will change, but (as with dealing with people) knowing how to play to a model’s strengths while sidestepping its weaknesses and being aware when it goes astray will be an integral part of the student’s adult life.

Keep Your Eyes On The Prize

It’s easy to become overwhelmed by the rapid pace of change spurred by AI. But despite all the new tools and processes that come about, keeping the big picture in mind will help ground discussions and decision-making. Are our graduating students well-prepared for what comes next, whether it is participating in the workforce, going to college, or taking other paths through society? Can they navigate a world that includes AI but also still do well in those areas not touched by AI? High school, in particular, is a time of discovery and maturity, a chrysalis where children become young adults. Are teachers helping students see their potential and opening doors where that potential can develop? Can our students participate in society – do they leave school valuing teamwork, a sense of civic duty, and community?

The greatest risk is perhaps not in not using AI. There are a great many reasons why AI tools should not be immediately adopted – cost, their constant change, and unclear harms that could be incurred. But we need to make such decisions consciously and deliberately. The great risk is not in not using AI, but in ignoring AI. When we do that, we turn our back on the change happening around us, and those we are responsible for will be ill-prepared for a rapidly changing future.

AI is novel. It’s a technology giving rise to a whole new set of tools and workflows that are revolutionizing a great many industries, and arguably society itself. The current brand of AI, which is based on a form of computation called neural networks, can tackle tasks that computer programs traditionally struggled with. It is also evolving at a tremendous rate, spurred by a public interest that has resulted in a massive influx of capital. That has driven research and product development at a much faster pace than would have normally occurred. The novelty and rapid pace of AI development pose challenges for organizations adopting AI.

The novelty means we are, at various levels, trying to wrap our heads around what AI tools are, as well as how to use them. Traditional computer programs work or don’t. They’re predictable. We have learned to evaluate how well they work and whether they fit into our workflows. AI tools are a new class of tools. There are new parameters indicating how well they work (i.e., how completely they can answer a question posed to them or carry out a requested task). AI tools may also hallucinate, provide partial responses, or fail to carry out tasks in a requested manner. A recent example of the latter is an AI tool that deleted a company’s database by “accident” because it “forgot” to perform various checks before deleting the database. We need to evaluate AI tools in a different light than traditional tools, taking into account their unique strengths and weaknesses.

The pace of change means that we are struggling to catch up with the changes wrought by AI tools. Technologies typically advance at a rate that reflects the funding and effort that goes into those advances. They also tend to advance in step with the technologies they depend on. CTOs and other technologists can chart the progress of a technology (and its dependent technologies) in creating product and innovation roadmaps. The rate of impact on customers and society is also generally measured and itself moderates the rate of advancement – products are generally developed when customers are willing to pay for them. This gives people and organizations time to integrate new tools and processes into their workflows and understand both the benefits and dangers they bring.

Neural network AI is quite old (the concept dates to the ’60s) and arguably interest started accelerating int he early ’00s. By then we were able to create neural networks large and fast enough to be “interesting”. But interest went rather non-linear after the public release of ChatGPT 3.5 in November, 2022. The hype and subsequent influx of capital into the sector meant that AI went into an If We Build It, They Will Come development mode – ignoring the usual constraints on technology development.

This has resulted in various stresses as we attempt to adapt. AI tools are causing changes to the workplace – jobs are being created, eliminated, and altered. They can allow existing tasks to be performed more efficiently. They are changing workflows as classes of tasks that could not be automated by traditional computing can be tackled by AI agents. The demand for AI processing power is stressing power grids at unprecedented levels. That demand is also causing component shortages, driving prices up drastically, as AI servers compete with laptops, PCs, and even TVs and set-top boxes for storage, memory, and other components.

There are also societal challenges resulting from these new tools. For example, generative AI can create photorealistic images and videos from a simple text description. We are approaching a time when, if you can imagine it, you can create it. But what does this mean for the authenticity of photos or videos? Do we know when something is “real”? If we obtain “camera footage” of someone stealing from a store, did it really happen? Or was the video faked? What does it mean for actors if a creator tells an AI tool to use an actor’s likeness in creating a movie? More concerningly, AI tools are increasingly used by cybercriminals.

Over time, new societal norms will arise, legal judgments will be made, and our cyber defenses and practices will evolve to include the new capabilities of AI tools. But until that happens, it is up to individuals and organizations to determine how to responsibly and safely use AI.

The benefits that AI tools confer mean they are too important to ignore. But at the same time, they should be approached as any new technology – with attention to its unique strengths and weaknesses, with an eye to how they both enhance and disrupt, and with safeguards to minimize abuse.

There is a study making the rounds and raising some concern about the use of AI chatbots. Headlines such as Cognitive scientists found using AI for just 10 minutes impairs brain performance raise a rather alarming concern that chatbots can cause harm. Indeed, in the study itself, the authors conclude:

…we find that AI assistance improves immediate performance, but it comes at a heavy cognitive cost: after just ∼10 minutes of AI-assisted problem-solving, people who lost access to the AI performed worse and gave up more frequently than those who never used it. These findings raise urgent questions about the cumulative effects of daily AI use on human persistence and reasoning. We caution that if such effects accumulate with sustained AI use, current AI systems — optimized only for short-term helpfulness — risk eroding the very human capabilities they are meant to support.

The study, AI Assistance Reduces Persistence and Hurts Independent Performance (https://ai-project-website.github.io/AI-assistance-reduces-persistence/), measures the performance of subjects in solving math fraction problems. Some participants had access to a chatbot (AI-assisted group), while some did not (control group). Later in the test, the chatbot was removed from the AI-assisted group. The participants’ ability to solve the problems was assessed, and it was observed that when the chatbot was removed from the AI-assisted group, their ability to solve the problems decreased, and the tendency to skip problems increased. The provocative part of the data is that the AI-assisted group’s performance did not return to that of the control group – instead, the (formerly) AI-assisted group’s solve rate was worse than the control group’s, and the skip rate was higher.

At face value – and indeed what the study’s authors suggest – is that “AI assistance reduces persistence and impairs independent performance.” This is quite a claim, and while there is no reason to dispute the data collected, the authors fail to discuss or propose any mechanism by which AI assistance caused the reduced persistence or independent performance. While a causal mechanism is strictly not required, such a mechanism greatly boosts findings from correlational to causative. It is a factor in criteria such as the Bradford Hill criteria for establishing whether an association between a presumed cause and effect is causal or merely correlative. The mechanism provides the Why A caused B, and allows findings to be extrapolated to conditions outside those of the study (which may be controlled and somewhat artificial). In the absence of any identified mechanism, it is necessary to carefully and critically evaluate the data in light of the study’s design and the behavior of subjects.

Dissection

Let’s first take a look at the study. It uses the term learning to label the part of the experiment where the AI-assisted group had the chatbot available and test for the part where the chatbot was removed. This labeling may be somewhat provocative, as it is unlikely that the subjects were actually “learning” during that phase. (If so, then there are a host of other questions regarding the study.) It is more probable that the subjects had learned how to do math fraction problems as part of their general education.

A less provocative term for the phase might be baselining, preparation, or priming. The goal was presumably not to see how the subjects learned, but to establish a baseline of performance. Priming may describe a phenomenon occurring during that phase of the test – the subjects entered a cognitive state for solving fraction problems. That state then primes them for how they behave in the next phase of the study. In the AI-assisted case, the priming results (for some subjects) a state where they ask the chatbot to solve the problem, in others to provide hints, and for others to slog through without assistance. The non-AI cohort had only the last option.

The study also uses self-reporting to determine how the AI-assisted group used the chatbot. Presumably, the interaction with the chatbot was (or could have been) available to the researchers. In such cases, reviewing and scoring the interaction for the level and type of assistance would provide more solid data than self-reporting, which relies on memory and may be tainted by personal biases. Understanding what the interaction could be a key factor in understanding how the cognitive processes of the AI-assisted cohort differed from the control cohort.

A more detailed description of the “fraction problems” would also be illuminating – the study provides an example as part of a diagram showing a multiplication problem. This is one of the easier types of fraction problems. Fractional division problems can be reduced to fraction multiplication problems – if the subject remembers that. (The author of this post had to think a bit to recall that nugget of information…the author is a firm believer in the use of calculators). Fractional addition and subtraction problems can pose more challenging, and once again require the subject to recall the method for solving these types of problems.

There is also the question of the cognitive processes that occurred when the chatbot was removed. Was this a case of violation of expectation? When one comes to expect something, and then something else happens, this causes some amount of dissonance. Such a state would likely affect performance.

Priming and similar phenomena are well-known aspects of cognitive science that many of us can see in our daily lives. It’s the foundation of adages like You never forget how to ride a bicycle or the need to brush up on [something]. Skills that are learned are not always immediately available at the same level as they were when first acquired. They may seem completely “forgotten”, when they are better thought of as dormant and imperfectly recalled (though over time some amount of actual “forgetting” may occur). However, the initial skill level can be reacquired – or exceeded – by performing the skill. This generally happens much faster than initial acquisition of the skill – a testament to its latent presence – and the skill may be actively retained to a greater degree afterwards as it is associated with new contexts (in connectionist theory, there are more connections.)

Once you learn to ride a bicycle you never forget

There are some interesting findings from this study: First, “Participants who used AI for hints showed no significant impairments relative to control.” The impaired subjects were those who asked the chatbot to solve the fraction problem and provide the answer. While we can only guess why this happened, one potential reason is that the AI-assisted group was reminded by the AI of the process for solving the problem. Having “remembered” how to solve certain types of fraction problems, the subjects would be better-equipped to tackle the test phase problems. In an educational context, this correlation suggests that chatbots should be constrained to assisting students in solving problems (teaching them how to solve a problem) rather than solving the problem itself. This is probably not a great surprise to most teachers and reminds us of the adage Give a man a fish and you feed him for a day; teach a man to fish and you feed him for a lifetime.

The second interesting thing about this study is the correlation of the AI chatbot to superior performance when the AI is present. In a work context, productivity is a major concern. To the extent that a chatbot can enable an employee to be more productive, it is a great asset. The question that must be asked is whether there is permanent harm from using the chatbot.

The study regrettably suggests that there may be permanent harm, but does not provide any causal mechanism by which harm occurs. Without that, there is no way to prove further or counter the assertion.

Reflections

The study does not provide any indication of the cognitive processes involved and whether the noted effect of “harm” is persistent, permanent, or transient. “Common sense” suggests that the effect is transient in nature and directly related to a cognitive state of having an aid (the chatbot) perform work rather than doing it oneself. Some amount of dissonance resulting from the unexpected withdrawal of the chatbot may also play into the performance deficit. Given the variation in results during the “learning” phase, three test problems are not sufficient to see any trend in the “test” phase. However, it would be interesting to see if both cohorts’ performance eventually converged, how long that took, and also to assess the subjects’ cognitive processes.

Regardless, it is unlikely that the knowledge subjects had about fraction problems before the study was “erased” by the study. It is likely that, given time to “reset”, their unaided performance would mirror that of the non-AI cohort. (It is also very likely that the non-AI cohort would perform better on a second round, even if a few days or a week elapsed before the retest, due to the lasting effects of relearning and refreshing the skill.)

One suggestion that has been made by those who do use AI tools is that it is valuable to periodically do a task without those tools. This generally helps you to stay grounded in what the tool is actually doing and also keeps your ability to do the task at a maintenance level. While someone may still have a book-knowledge level of knowing what a tool does when using it, there is a deeper experiential knowledge gained by performing the task. This deeper knowledge conveys a greater appreciation for where and how the steps to completing a task must be altered in different contexts, and how the tool could be applied to different domains with suitable adaptations. In an extreme, it differentiates mere users of a tool from experts who both use and have a deep understanding of the tool.

Having something more than “book knowledge” is particularly valuable at a time when AI tools are relatively new and are still being refined. They may hallucinate or otherwise behave aberrantly, and it is valuable to be able to spot those errors. Eventually, AI tools will mature – and our understanding of how we should use AI in general will evolve – to a point where using AI tools will be like using a calculator today – a once-controversial tool that most find indispensable today for doing moderately complex math. But we’re not quite there yet.