Introduction and Scope
I prepared these notes as a thought experiment – a way to synthesize my ideas regarding how Artificial Intelligence will reshape human life. It is crucial to recognize that the precise nature and degree of AI’s impact are not yet clear. And today, our observations are overwhelmed by what feels like a professional tennis match between two opposing camps:
- The Skeptics: Those who argue that AI is merely a bubble, suggesting it will trigger a massive financial crisis.
- The AI Enthusiasts (Fanboys): Those who claim AI represents an evolution greater than the Industrial Revolution, promising to displace all white-collar labor and usher in never-before-seen economic growth.
The Dynamics of the Hype Cycle
We are witnessing the typical build-up: Skeptics’ warnings gain traction, focusing on inflated company valuations and AI companies struggling to prove profitability from their ventures. However, these critical warnings are invariably followed by a massive surge of “miracle” news – claims that mathematics, physics, biology, genetics and robotics have been “solved”. We hear claims of novel inventions or solutions to decades-old mathematical problems (like the Millennium Prize Problems).
The pattern is predictable: A wave of spectacular success stories emerges, only for subtle doubts or inconsistencies regarding those breakthroughs to surface. Yet, these moments of doubt never generate as much media buzz or market noise as the initial headlines do: The whispers surrounding OpenAI executives making veiled threats related to solving the Navier-Stokes equations, or the previous claims that AI could improve upon methods detailed in a mathematical paper through a single prompt, sometimes claiming entirely novel inventions – but these spectacular achievements seem confined exclusively to the employees of companies like OpenAI, while ordinary users never replicate such performance. The infamous “Trust Me Bro Benchmarks”, which are hailed with every new model release, have become an absurdity in themselves.
The Echo Chamber Effect: Manipulation on Both Sides
Skepticism is not immune to irrationality. I find certain YouTube channels constantly publishing formulaic content, labeling all AI-generated software as “AI Slop,” repeating the same core message to generate a narrative. They seem to have realized that the loudest voice can bend reality, and they attempt to extend their relevance by manipulating the narrative.
The Search for Truth in Chaos: Deconstructing the Noise
Given this environment of propaganda, conflict, and sheer noise, how is it possible to find the truth? Several things can help clarify the picture:
1. The Dot-Com Bubble Analogy
AI hype is often compared to the Dot-Com bubble, leading some to predict a similar catastrophic collapse. However, when the Dot-Com bubble burst, the result was not a total corporate collapse but rather a cleansing of overvalued assets. For example, those who thought a simple online marketplace for used school books would be worth $50 million – without analyzing profit margins or realistic transaction volumes – saw its value revert to the sober reality of $0-$100k. Investors’ Fear of Missing Out (FOMA) has a hefty price tag!
2. The Enduring Value of Labor – Good News for Us Plebs
While investors lost money, the IT workers and developers benefited immensely, enjoying an authentic “Golden Age” over the last 25 years. Starting salaries of IT jobs caught up with, or even surpassed those of Harvard Law School graduates. This era continued almost uninterrupted until the rise of AI.
The Parallel: Similarly, professionals actively engaged in the implementation of AI within corporate structures are poised to gain significant, perhaps irrational, economic advantages in the coming years. Small teams of entrepreneurs leveraging AI could complete enterprise-grade software rapidly, revitalizing small and medium enterprises (SMEs).
3. The Lagging Effect: Incremental Integration
A slower but highly significant change will occur even without a massive leap in AI’s fundamental capacity. As AI is slowly integrated into existing practices – education, medicine, law – the result will be demonstrable improvements in quality alongside dramatic cost reductions.
Illustrative Example (Law): An attorney preparing a for a case can use AI to analyze every legal transcript on record and generate an optimal game plan with all possible counter-arguments baked in. This allows the lawyer to anticipate and pre-empt counter-arguments for the opposition’s every point, which he could not have anticipated and be prepared for beforehand, if it wasn’t for AI. Such preparation can exhaust the counter-party’s strength from all possible angles. There will be a massive gap between the lawyer who masters this technique and those who cannot adapt.
4. Global Capital Dynamics: The “Next Big Thing” (NBT) Chase
For peripheral economies like Turkey (where I live), comprehending Western capital surpluses is challenging. Trillions of dollars cannot simply sit in time deposit accounts, or be poured into the construction sector. This massive surplus must flow somewhere, often leading to speculative bets on the NBT.
The frenzy once centered on VR (a bubble that thankfully deflated before becoming catastrophic). Facebook, after acquiring a VR company Oculus for USD 2 billion and changing its own name to Meta for the supposed raise of VR, is now moving more cautiously through the AI era, ready to jump off if necessary.
While Meta acts more cautiously, the hard fact still remains: The combination of Western capital accumulation surplus and Silicon Valley’s “fake it till you make it” ethos guarantees future bubbles. We may see an AI bubble burst at some point, which may first impact the financial system and then spill over to the rest of the economy, but the AI’s effects on both the economy and human life will still be valid.
The Core Technical Divide: Where AI Excels (and Struggles)
The Strength of Software Engineering
Software remains an area where AI has a unique advantage due to its near-perfect feedback loop. AI generates/proposes code. The compiler/interpreter executes it, providing instant confirmation of whether the code runs or fails (deterministic validation). It can also verify whether or not the end result that the program generates is what the user asked for. This closed loop means that much of the software development cycle can be executed without constant human hand-holding. This is where probabilistic computing (AI) is coupled with automated deterministic verification of the outcomes, and it minimizes the requirement for human intervention.
The Resilience of Human Judgment: Cybersecurity and IT Governance
However, not every field has such a clear feedback loop. Cybersecurity is comparatively more resilient for humans to lead.
- Automated Tools: Non-AI tools have long been used for automated scanning, but these tools are notorious for generating false positives (FPs). Even when results are not FPs, they might lack contextual significance.
- The Pentester’s Edge: A skilled penetration tester remains vital because they can efficiently filter and interpret results – a skill built on years of practice. When AI integrates into this process, the key limitation remains: many test results still demand human reasoning. AI predictions/interpolations/extrapolations are prone to error.
- The Client Perspective: Clients never want false positives. If they could tolerate them, they would never ask a vendor, “Are you doing just automated scanning, or do you also conduct manual testing?” This popular question proves that IT Governance and contextual review remain paramount.
Case Study: AI’s Limitation in Risk Assessment (The Pentest Anomaly)
While using AI tools like Claude Opus via CursorAI for a mobile application pentest, I successfully leveraged ADB (Android Debug Bridge) to connect to an Android Studio emulator with root permissions. Using Frida, the AI attached to the runtime memory of the mobile app, gaining access to encrypted data stored in Android Secure Storage and managing to decrypt it after two hours of searching Google for errors and trying alternative commands – all without my direct intervention.
The Critical Flaw: When I asked the AI to document its findings, I found a clear crack in its logic: Risk Assessment.
- My detailed methodology provided a scale: Low/Medium (“optional”) vs. High/Critical/Urgent (“mandatory”).
- The AI assigned the vulnerability as Critical.
- The Reality Check: The method used to gain access (root privileges via an emulator) is virtually impossible for an actual end-user attacker. This dramatically caps the likelihood of exploitation.
- A true assessment required downgrading the risk level from Critical to Medium, based on low likelihood.
When prompted to re-evaluate, and after I explained the real-world constraint (that a targeted user must either be using a rooted phone or must run the mobile app on an emulator with a specific image), the AI correctly downgraded it to Medium Risk.
The Question: Why did it initially fail?
Conclusion: The AI excels at following well-documented frameworks. Its “near-perfect feedback loop” allows it to follow the breadcrumbs and even arrive at somewhat “novel” results (that mobile app was not exploited by anyone before; therefore, the exploitation steps were not on the Internet, so AI did that for the first time ever). However, its output is divorced from genuine reasoning. It lacks the capacity for contextual judgment, which requires a higher level of abstraction capacity, some secret sauce that today’s AI clearly does not have. When faced with an unknown variable (like user behavioral constraints), it defaults to pattern matching – finding pre-written severity levels online for similar pentest findings, rather than calculating a true severity rating based on the specific operational context.
Final Synthesis: Redefining Human Value
This entire process leads to one undeniable conclusion:
Using AI can generate novel results, insights, and artifacts (the what), but it cannot assess their category or importance – AI lacks the metacognitive ability to determine why they matter or how likely they are in a real-world context.
The most valuable skills of the future will not be technical knowledge itself, but Meta-Cognition:
- Connecting Disciplines (T-Shaped Thinking): The ability to see how an AI solution for biology impacts legal liability frameworks.
- Reality Filtering: The critical skill of distinguishing between technically possible outcomes and practically probable ones.
My advice to a student nearing graduation would be to try to become a good, curious generalist, and on top of that picking an area to do a deep-dive, for the sake of learning how to do a deep-dive. The combination of both skills will provide the basis that the new white collar will need going forward in their career journey.