When and How Not to Use AI

I've been touting the mantra "AI can solve everything" in my social circles for more than a minute, but today, I'm about to switch tunes and indulge in a bit of self-contradiction. That is, once we've got a grip on the capabilities and limitations of AI, understanding what it can and cannot do, it's crucial not just to dive headfirst into using AI for everything. Instead, we should judiciously decide where AI should step in to truly achieve the outcomes we desire.

Let's get down to brass tacks. Recently, it dawned on me that the ultimate use of AI might just be in peddling courses. And to sell a course, you need an introduction that really packs a punch. Consistent with my usual stance, I figured it'd be worth seeing how well AI could handle crafting such an intro. But, like many other decisions, choosing whether or not to use AI isn't a black-and-white issue. Often, our choices lie somewhere between fully embracing AI and steering clear of it altogether. The real skill lies in pinpointing exactly where on that spectrum to land.

For this particular pickle, I weighed up three different approaches. The first was the path of least resistance: if it's possible to use AI, then why not? I gathered all the potentially relevant info and relayed it to GPT, explaining the target audience for the AI course and the specifics of what my intro needed to highlight, such as showcasing my credibility and drawing in sign-ups. Here's what AI churned out for me:

Dr. Ya Ge, a visionary in the field of Computer Vision and Machine Learning with a Ph.D. from Columbia University, stands at the forefront of innovative applications of GPT. With 10+ U.S. patents, 20+ papers on top conferences/journals, and impactful algorithms used by billions globally, Ya's expertise bridges the gap between theoretical research and real-world applications. His commitment to exploring GPT's capabilities since its inception underpins his belief that humanity is on the cusp of a new industrial revolution, where proficiency in GPT can provide a decisive productivity advantage. Ya's diverse achievements, from being an Olympic torchbearer to hosting a photography exhibition with Leica, underscore his multifaceted approach to innovation and makes him an unparalleled mentor for those seeking to leverage GPT's transformative power in their professional journey.

The second tier involved me, the human, picking out the most crucial bits of information, drafting an outline, and then articulating it in the most straightforward language possible for AI to tidy up. In this scenario, AI's main job was akin to clerical work. Here's the AI-produced self-introduction that followed.

Dr. Ya Ge is deeply committed to creating functional products by developing specific mechanisms, fostering continuous innovation, and utilizing effective tools. His algorithm has been integrated into Photoshop, Bing, and Pinterest, benefiting millions of users. His achievements have also attracted global media attention. Ya has contributed to advancing human science by publishing over 20 papers at top-tier conferences and securing more than 10 U.S. patents. Since the emergence of GPT, Ya has been diligently exploring its potential, gradually forming a strong belief that we are at the cusp of a technological revolution akin to that of steam engines. He posits that individuals proficient in AI could have a significant advantage over those who are not. Leveraging his proven career coaching skills, Ya has discovered a practical and effective method to enhance productivity.

The final tier was going full manual—manually selecting materials, organizing the narrative, and refining the language myself. Below is the version I personally crafted, directed, and wrote.

Dr. Ya Ge has devoted his passion into building working solutions to complicated engineering problems since he got his PhD degree in Columbia University. With emphasis on tailored mechanism, continuous innovation, and specialized tools, Ya consistently delivered algorithms and systems to Photoshop, Bing, and Pinterest, which have benefited millions of people and got widely reported by global media. Ya also contributed to pushing the boundary of human science by publishing 20+ papers on top conferences/journals and inventing 10+ US patents. Since the initial appearance of Large Language Models (LLMs, often referred as AI), Ya has been exploring its potential and gradually built a belief that, we are at an intersection of technology revolution similar to steam engines, and people who know how to properly use AI may build a decisive advantage over those who don’t.

Throughout this process, I distinctly felt a significant disparity between the first version and the subsequent two. Lacking my thoughtful consideration, the AI-selected materials, while seemingly related to credibility and adorned with elaborate language for self-introduction, lacked the depth of the latter two versions.

In discussions with my course partner Kedaibiao, I realized a noticeable difference between the second and third versions, even though their core content and structure were identical. The AI's phrasing felt particularly diplomatic, creating a sense of detachment. If human writing is akin to delivering a passionate speech on stage, AI writing is more like a diplomat coldly reading a script. This detachment results in a lack of engagement and persuasiveness. For instance, the human-written version starts with "has devoted his passion into building working solutions." Words like devote, passion, and working, though few, are opinionated. In contrast, AI’s "deeply committed to creating functional products" feels bland.

Kedaibiao also noted AI's use of language can feel subtle or even odd, like its preference for “utilize,” which not only sounds overly formal but also hints at a lack of familiarity with technology. It's like saying, "Today, I utilized chopsticks to accomplish the remarkable feat of eating," subtly suggesting unfamiliarity with chopsticks as if using them is something extraordinary. In comparison, human language is more direct and natural, opting for “use” instead. Thus, although the content volume may be similar, human writing naturally exudes tension, vitality, and engagement, whereas AI writing, though ornate, lacks depth and soul. The class rep likened this contrast to Michelin-starred meals versus pre-made dishes, a comparison I find spot-on.

My learnings boil down to three points:

  1. Deciding whether to employ AI isn't a simple black-or-white issue; it's about figuring out where to position ourselves on the spectrum or what combination of tiers to adopt.
  2. The decision to use AI isn't just a technical consideration of whether AI is capable of completing a task but also involves non-technical aspects. For example, in serious scenarios like selling AI courses, presenting customers who've spent hundreds of dollars with a pre-made dish is fundamentally disrespectful. Hence, choosing not to use AI isn't solely due to its capability limits but is more about showing respect to our users.
  3. This experiment has deepened my understanding of the differences between AI and human writing. Previously, I merely thought AI-written articles looked somewhat odd, with a distinct AI flavor. But after in-depth experiments, reflection, and discussion, I've gained a deeper appreciation of the subtle differences in word choice and the ability to engage.