AI: Trading Understanding for Speed

AI is in a very bizarre place right now.

According to a recent survey by Verasight, trust in AI is actively falling. Concerns over safety, accuracy, societal impacts, and the environment are just some of the major issues eroding what little trust users had.

However, despite the bevy of concerns, the use of AI is growing. Nearly two-thirds of Americans use AI regularly, and half do so at least once per week. It’s an interesting dichotomy: AI usage continues to grow, but trust and acceptance are shrinking even as its reach expands. 

Nowhere is this more evident than with computer programming. A study of 1,100 IT professionals by Sonar found that 96% of respondents don’t fully trust AI code. However, 72% said they use coding assistants every day, but fewer than half of users verify AI-generated code before committing it. 

People know that AI-generated work shouldn’t be trusted, but are doing so anyway. 

In programming, this can represent a serious issue. According to YouTuber Imran Gardezi, the blind use of AI-generated code doesn’t just create technical debt, but cognitive debt. When something goes wrong with AI-generated code, humans don’t understand it well enough to fix it. This creates downtime, raises costs and creates other issues.

But that cognitive debt isn’t just for companies or computer programmers. As a now-famous MIT study found, using AI causes the brain to engage far less than traditional writing does. While there are legitimate criticisms of this study, the principle that writing supports learning has been a tenet of education for at least forty years.  

As one 1993 thesis put it:

“The authoring process engages students in a multitude of learning experiences. Students use writing as a tool to grow beyond themselves. The act of writing helps students synthesize and analyze what they have learned about life.”

That learning and understanding may be the biggest sacrifice with AI. Creation isn’t just about ticking a box or completing a requirement; it is part of the learning process.

However, we lose that when we skip to the final step of a project. 

Writing Isn’t Just Words

To be clear, I’m going to focus heavily on writing in this article because it is the craft I am most familiar with. However, the same principles can be applied to any form of creativity, including, but not limited to, art, photography, music, filmmaking, and computer programming. 

When I sit down to write an article, typically, less than one-third of the time I spend is writing or editing. I am a very fast typist who normally gets over 100 words per minute, so if I just wanted to fill the screen with random words, I could likely “write” a 1,000-word article in approximately ten minutes.

But that wouldn’t add value to me as the author, nor to anyone who reads it. It would be beyond meaningless. 

The vast majority of my time is spent reading and researching. This includes reading articles, pulling down legal documents, comparing the story to any previous coverage I’ve done, and finding all the information I can about it.

While I hope this work adds value to the reader, its greatest value is to me. Over the past 20+ years, I have written nearly 6,500 articles for this site. When I started this site, I had some understanding of these issues. But by writing about these topics regularly, researching things that interest me and publishing my thoughts, I’ve learned more than I ever thought possible.

While there is still plenty for me to learn, there always is, I’ve managed to make a career for myself as an expert in this space. This would not have been possible if I had generated those articles using AI.

Even if the AI articles were every bit as accurate, well-researched and original as writing, I would have lost out. If generative AI had been available in 2005, when I launched this site, and I had used it to write these articles, I would not have a fraction of the understanding I have now.

That’s something I wouldn’t trade for the world.

AI’s “Bargain”

Every time you use AI to generate a final product, you’re making a trade. You’re saving a great deal of time (and possibly money), but you’re losing your understanding of the finished product and the knowledge you would have gained creating it.

In some cases, this may be a fine trade. Not everyone needs to be an expert in everything. You may also completely understand something, and generating the final product doesn’t harm you.

But one thing that’s always surprised me is how much I don’t know. Even mundane tasks such as writing emails and creating reports can be very educational. That, in turn, is one of the problems with this trade: you have no way of knowing what you are actually sacrificing. 

There are also ways that you can mitigate that lack of understanding. Carefully reviewing the finished product, writing explanations for what it means, refining it, etc. 

But all those approaches take time. There simply is no shortcut to understanding. To understand something, you need to take the time to engage with it. One of the best ways to do that is to create something. 

To understand something, you must engage with it. Creativity forces engagement. This is one of the reasons that, despite all the issues with the academic essay as an assessment tool, it has remained the standard. It doesn’t just assess, it teaches.

Even if AI produces acceptable work (not that anyone would be able to tell without proper understanding), it robs the “creator” of the chance to learn and grow as part of the process. 

That’s an opportunity that, in many cases, won’t come around a second time.

But then, somehow, things get worse.

The Dunning-Kruger Machine

By itself, the loss of understanding might not be particularly bad. If people are aware of the tradeoff, limited the use of AI in certain spaces and encouraged steps to claw back some of that understanding, this drawback could be blunted.

However, a pre-print study submitted in September and highlighted by PsyPost in January, shows that sycophantic AI chatbots actually inflate users’ confidence in their own abilities.

That led Frank Landymore at Futurism to call AI chatbots “Dunning-Kruger Machines”, referencing the effect that causes people to overestimate their capabilities. 

In short, using an AI to create something not only robs you of the understanding that would come with doing the work yourself, but it gives you the feeling that they gained an understanding that you did not.

This means that when you attempt to engage with the topic, with or without AI, not only will you be less capable than if you hadn’t used AI initially, but you will likely overestimate your knowledge.

In 20 years in this field, I am still constantly blown away by how much I have to learn. I know that there are gaps in my knowledge and understanding. Sometimes that knowledge makes me reluctant to talk about topics I really should. But that’s what part of learning is, being uncomfortable with what you don’t know.

AI, on the other hand, sells the illusion of real knowledge and understanding while robbing users of the chance to obtain it. It’s a double-whammy that makes it incredibly difficult for someone to reverse course. 

Once again, maybe that trade is an acceptable one for you. But you should at least be aware that you’re making it. 

Bottom Line

Though many still claim that AI is just like using a calculator for learning, I’ve already addressed this. Not only is AI prone to mistakes that require understanding to catch, but there are many classes where calculators are still banned. 

The reason is that students need to understand the basics of mathematics to use the calculator properly. If you don’t understand the fundamentals, the calculator can’t and won’t help you.

AI will produce a finished(ish) product regardless of the user’s understanding. Worse still, that understanding may never follow. That just makes you more and more reliant on the AI, rather than more self-sufficient. Furthermore, calculators don’t attempt to flatter their users into thinking they have skills they don’t. 

Once again, there may be situations where that is fine and a trade people are willing to make. But it’s a trade that everyone should be aware of. It’s a decision that everyone should make consciously.

And that is the ultimate point of this. While I could list all the problems with AI and all the reasons not to use it, most of those reasons are well-known and haven’t changed the arc.

What I want is for people to think about what they are giving up when using AI, especially to create a “finished” product. While it’s tempting to skip to the end of a project, doing so doesn’t just potentially rob your audience; it robs you. 

It’s something to think about the next time you submit a prompt.