Beyond the Fascination

Introduction

"Time and tide wait for no man". We hear this constantly, whether from the CEO of Nvidia or from artificial intelligence companies pushing the narrative that AI is the new norm - one we must either embrace or be left behind by.

• Everywhere we turn, people are offering advice on how to optimize AI in our daily work.
• However, while I feel a sense of excitement about these advancements, I also harbour some serious reservations.

The Illusion of the Infographic

The most immediate impact I see in daily life is the lightning-fast generation of AI videos and infographics.

• Yet, far from being excited by these excellent graphics, I find myself a bit old-fashioned; I am not a person who loves visuals for the sake of it.
• In fact, it often takes more time to digest an infographic because we have to decipher the sequence and figure out the core message.
• More critically, infographics risk omitting crucial details that we might easily dismiss.

We need to look beyond mere fascination and focus on actual substance.

• What truly shines in any work is never how the cover looks, but the originality of the ideas inside.
• Currently, AI-generated content is mostly abstract and generic. I once watched a speaker present a slide deck generated by AI; it was flashy and filled with nuanced vocabulary, but it completely missed how humans actually learn.
• We crave real, actionable takeaways.
• In the years to come, I hope attending an information-sharing session does not become like going to a modern cinema - where we no longer care about the actors' skills, so long as we can tolerate the storyline.

Furthermore, a much more dangerous issue is the hallucination of information.

• If you use AI to generate content, you must proofread it meticulously.
• You can provide the AI with perfectly accurate facts and flawless inputs, yet the final output can still be filled with glaring errors and fabrications.
• The tool simply cannot be trusted to preserve factual integrity on its own.

The Erosion of Critical Thinking

An even more dangerous trend I have noticed is how an overreliance on AI can erode critical thinking through a phenomenon known as "cognitive offloading".

• Whenever we face a difficult question, we immediately jump to an AI chatbot, robbing our brains of the opportunity to problem-solve.

With AI so readily available, children exposed to it in early education may misuse its convenience for homework.

• On the surface, they find the correct answers with ease, but they completely skip a vital developmental phase: the process of wrestling with a problem and persisting through cognitive frustration. This struggle is essential for cultivating critical thinking and psychological resilience in adult life.
• After all, the ultimate purpose of academic study is not merely to memorize operations like mathematical differentiation or integration, but to fundamentally sharpen our capacity to think.
• In the past, it was unheard of for a medical graduate to break down at home from work stress or want to quit their internship after a few tough shifts.
• It is a sobering observation that younger generations risk becoming more fragile because they are born into comfort and abundant resources, rather than growing through hardships where they must navigate hurdles and push through challenges.

I have also witnessed firsthand how an overreliance on AI can erode confidence in our innate abilities, leading us to preemptively delegate tasks and intellectual responsibilities to the machine.

• Consequently, problems are no longer solved through active critical thinking, but rather through a passive dependence on automated outputs.
• Crucially, this reliance creates a feedback loop of self-doubt; many of us now feel fundamentally incapable of expressing our own thoughts clearly or grammatically without the validation of an AI filter.

The Surface of Research

As a healthcare professional, I have noticed a rising trend of peer-led workshops focused on maximizing AI utilities in research.

• Investigators are increasingly deploying AI to identify literature gaps, conduct systematic reviews, and even draft manuscript discussions.
• While tools like NotebookLM excel at synthesizing vast volumes of data, an overreliance on them risks bypassing the rigorous, deep reading required to truly master a topic; we end up skimming the surface and commenting on what appears to be true rather than what is verified.
• AI is an invaluable efficiency multiplier for screening thousands of abstracts in minutes, but it must only serve as a precursor to deep, critical analysis of the selected literature. To become a true subject-matter expert, there is simply no algorithmic substitute for human intellectual rigor.

Beyond superficial reading, an even greater concern is the hallucination of consistency.

• We may feel we are building robust new frameworks through AI analysis, but a large language model's outputs are notoriously volatile and stochastic (probabilistic).
• Today, an AI might evaluate your conceptual framework favorably; tomorrow, a different prompt iteration or model update might dismiss the exact same logic as completely ungrounded in reality.
• Because these tools frequently cherry-pick patterns based on fluctuating search context windows, they lack a stable epistemological baseline.

Ultimately, while AI can ingest generations of human effort and generate a slick summary, it is fundamentally backward-looking - it merely feeds on the past.

• True academic progress does not come from rearranging existing data into synthesized averages.
• The only way to move a field forward is through genuine clinical innovation and the paradigm-shifting insights that only human thinkers can provide.

The Pragmatic Reality of Clinical AI

I have thought deeply about how artificial intelligence might truly personalize patient care. Broadly speaking, AI in healthcare falls into two categories:

• Classical AI, which excels at pattern recognition and predictive analytics, and
• Generative AI, which synthesizes unstructured data to draft clinical narratives and summarize patient histories.

In theory, we could deploy AI to monitor patient laboratory results and track trends to predict disease risks.

• However, looking closer, we must ask a fundamental question: do we actually need AI for this, or can a few lines of traditional code do the exact same job?
• For instance, if a patient has a prolonged activated partial thromboplastin time (aPTT), a simple hard-coded rule can automatically trigger reflex testing for hemophilia prior to surgery - after first ruling out confounding factors like warfarin use or acute infections like dengue fever.

If we can manage these clinical alerts with basic programming, the argument for AI often shifts toward operational automation, such as drafting patient notifications for critical lab results.

• Yet, even here, reality intrudes.
• Because of medico-legal liability, healthcare professionals must still review and approve every single AI-generated clinical communication.
• Furthermore, routine tasks like medication adherence tracking or appointment reminders do not require AI; they have run smoothly on standard automation for years.

Where, then, does the true value of AI lie?

Some suggest using AI to interpret every routine lab panel.

• This is expensive overkill.
• In the near future, when tech companies stop subsidizing AI computing costs, healthcare systems will face massive token expenditures without any guaranteed return on investment.

While AI positions us for a major technological shift, I am highly skeptical of using it to "vibe code" applications for basic medical tasks.

• We should focus on deterministic, client-side functionalities rather than relying on costly, server-side AI actions for problems that simple, traditional code already solves.

We often marvel at the capabilities of AI, but we must consistently ask: are these solutions only possible through AI? Or is there a more efficient, predictable, and cost-effective way to achieve them without paying a recurring subscription bill?

Tools, Not Truths

AI is undoubtedly capable of solving many things, and there are moments when I look up information using these tools myself.

• I encourage everyone to put them to use, but we must treat them strictly as tools - like a technical expert whom we consult on how to perform a task.
• In daily life, we need to remain skeptical of the information AI feeds us, questioning it just as we would question a friend, rather than accepting it with blind faith.
• Interestingly, if you debunk an AI's point and argue back with a valid argument, it will quickly change its stance and agree with you.

Ultimately, AI is a tool to find information, but it is never the source of truth.

• It should not be your only connection to knowledge.
• There remains a profound human need to look beyond the virtual world and truly connect with diverse opinions and real human beings.