February 16, 2026
AI and Social Impact
Some thoughts examining the socio-economic impacts of AI, highlighting both its disruptive potential and the urgent need for human oversight, equitable governance, and nuanced understanding to ensure AI benefits society as a whole.

Disclaimer
Before you dive in, a few notes on how this piece came together:
These thoughts have been brewing for some time. However, the viral essay by Matt Shumer - "Something Big Is Happening" provided the necessary trigger to finally put structure to these thoughts.
This post began as handwritten note on my Supernote. I later used AI to convert it to digital text and as a collaborative editor to help refine the language and restructure the flow while preserving my original perspective and tone.
The included list of references is indicative and was collated with the assistance of AI. While I have not personally audited every individual link, these events and studies align with my own independent reading of widely reported industry news and academic research.
Introduction
The AI wave is going on strong, and it's not stopping any time soon. I have seen first-hand what it can do, albeit in a narrowly scoped world and from a one-person perspective. It is a very capable tool in the right hands, but one that outputs garbage if you don't know what you want. For some specific scenarios, it produces garbage even when you are explicit about what you need. I was frustrated when trying to create specific, technically accurate images, and resorted to creating them manually. AI still hallucinates when there isn’t enough information and gets locked in a loop when the context becomes too large. Some of these have workarounds if you've spent enough time with the tools. And that brings me to my main thought: AI tools are advanced and becoming more capable with each new generation, but eventually, they are tools, bending to the will of the wielder.
So, I want to take all the hype with a pinch of salt. Now that I've had some time to think through the "Something Big is Happening" blog over the weekend and spoken to a few friends, I agree with the broad strokes, but find some nuance on second order effects missing.
What I Agree With
Almost every industry will be disrupted by AI, knowledge work more immediately, and physical jobs when robotics is finally advanced and cheap enough to perform general-purpose actions.
Lower-level jobs with high repetition and low-value outputs will vanish. One AI-enabled person will produce the output equivalent of ten grunt workers.
We should tighten our belts and save/invest. Living below our means is solid financial advice.
We should find passion in something and use AI to accelerate execution. This leaves more time to build deep expertise, which will be the only moat for having a career in the future.
What I don't Agree With
Jevon's Paradox is real and has held firm for several previous disruptions, so it should apply to AI-based disruption too, with caveats. There will be jobs lost, but new types of work will emerge that we can't yet imagine. The work my son will do hasn't been invented yet. I remain hopeful for the future. The pace of change is unprecedented, and our economic order may not have enough time to adjust. And if the new jobs created also play to the strengths of AI, human work may be in danger.
Moravec’s Paradox applies here, too. AI has the exceptional ability to analyze vast amounts of data and text. Yet it can’t navigate office politics, build client trust, evaluate the applicability of an analysis to a specific situation, and pivot. These remain deeply human, built on years and decades of intuition. Even the most advanced models still hallucinate, and you need human oversight. This is especially true in areas where the cost of being wrong is unforgiving - healthcare, banking & finance, insurance, law enforcement & defense, and many others.
The comparison to COVID is an exaggeration. I don’t disagree with the scale of change, but the speed. This change will unfold over months and years, not upend the world in days and weeks.
The Missing Nuance
Conflict of Interest: The doom and gloom tone feels alarmist. The loudest voices in the industry come from sources with a deep conflict of interest if AI does not meet the hype. Companies pouring trillions into hardware, data centers, and frontier models, and individuals benefiting from it, cannot be trusted to provide reliable information if there is a scaling problem with AI. The call to use the latest and greatest model has me conflicted. While paid models are substantially better than the free ones, only a privileged few can afford to spare a couple of hundred dollars on AI subscriptions and API tokens. On the other hand, it seems like a direct sales pitch to increase adoption. Outside of tech and tech-adjacent areas, most people think of AI as a toy for creating images, videos, or smart-sounding generic content. The world could benefit from better adoption, but the message must come from someone without a conflict of interest.
Vanishiing Career Ladder: I argued above that deep domain knowledge and the soft skills to navigate human situations form the competitive moat for the future. These skills need a gut feel and a pulse on the industry that only comes with experience. That in turn needs time in the trenches, learning the ropes, building the intuition to take the human judgment call. The post completely sidesteps the question of what happens to succession plans and finding the next set of leaders if fresher and early career hiring vanishes. Who will take the mantle when the current leaders retire over the next couple of decades?
Income Inequality: If 1/10th of the workforce can generate the same output with AI, what will the rest 90% of the working age population do? I don't see things like UBI and universal healthcare becoming a reality anytime soon. The financial beneficiaries of the AI revolution are a handful of people, so this will accelerate the accumulation of wealth at the top, leaving the majority destitute. Everyone wants to build robotaxis and automate warehouse operations and factories. Who is going to buy these services and products if people must choose between food and rent? The current runaway late-stage capitalist model needs to give way to something more equitable with a safety net. No wonder the rich are building bunkers.
Model Collapse: The AI promise may be overhyped. Yes, things will continue to improve, but model collapse and regression to the mean are real. You need human beings for true original creations and to maintain the tail, the outliers, the data that prevents model collapse. Without that, AI feeding on AI-generated content will result in future outputs being generic, meaningless slop. That makes true human ingenuity, with all its flaws, a premium commodity. There already is an emerging trend of people subtly sensing AI-generated content and quickly disengaging. And outside of coding/data analysis/text generation/some creative areas, AI still sucks at real-world tasks. It needs a lot of growing up to do.
Parting Thoughts
The biggest threat of AI is not immediate job losses or industry disruption. We are a resilient species. We will survive this change and adapt to whatever becomes the new normal. The biggest threat is letting AI do the thinking for us. Our critical reasoning skills are what set us apart from other animals, and now, thinking machines. Schools and parents already struggling to deal with screen addiction will have a much broader challenge - how to ensure the next generation is in charge of their mental faculties? We are already seeing effects of outsourcing our thinking - students submitting LLM-generated essays, police picking up the wrong people blindly trusting AI IDs, doctors operating on the wrong parts based on AI diagnosis. AI is a capable tool, but it needs a master to command it and push back when it errs. The best AI outcomes are when you iterate with it, reason, plan, and then, and only then, execute.
AI is only as good as the data it is trained on, and it reflects the same biases we have. Social adoption should pause till governance is in place to ensure equitable and just outcomes, free of biases. In the corporate world, too, the promise of 10x productivity will be a distant dream unless companies clean up their data exposed to AI models. Till then, it is Garbage in - Garbage out.
I use AI almost every day in my work, and I am in awe of what these systems can do. But there are real challenges that need to be addressed - social equality, copyright protections, fact-checking, regulatory protection, bias removal, etc. - before we can usher in a true age of AI where it is helping everyone, not just the ones living on the edge.
PS: If you've read so far, thank you! I must explain why the hero image. That is an image of the pre-dawn sky which to me reflects the dual nature of our reality where AI is a double edged sword. It also is an example of the human judgement I talked about in this post. I could have asked AI to generate an image with a description, but its not as fun as standing at 6am in your balcony, shivering, trying to capture the unusual dual tone sky :)
References & Evidence
I. The Technical “Inbreeding”: Model Collapse
Shumailov et al. (2024), Nature — The seminal empirical study defining model collapse, where LLMs trained on synthetic data lose tail‑end diversity and drift toward incoherence.
Harvard Journal of Law & Technology (2025) — “Model Collapse and the Right to Uncontaminated Human‑Generated Data.” Argues that AI requires access to real human data to avoid degenerative drift.
IBM Research (2024–25) — “What Is Model Collapse?” Clear explanation of diversity collapse and regression toward the statistical mean when models recursively train on their own output.
ACM Communications Blog (2024) — Practical walkthrough of model collapse emerging in real workflows where AI is forced to ingest AI-generated content.
https://cacm.acm.org/blogcacm/when-ai-tools-train-on-ai-output-model-collapse-in-daily-workflows/
DataCamp Technical Summary (2024) — Accessible breakdown of how collapse emerges over training cycles.
II. The “Vanishing Ladder” & Real‑World Failure
MIT / NANDA Report (2025–26): “The GenAI Divide” — Found that 95% of enterprise GenAI pilots fail to deliver measurable business value.
(Covered in Futurism summary) https://futurism.com/ai-agents-failing-companies
Carnegie Mellon + Salesforce Research (2025) — AI agents fail ~70% of practical multi‑step office tasks, especially anything requiring UI interaction or coordination.
https://www.theregister.com/2025/06/29/ai_agents_fail_a_lot/
Yahoo Finance (2025) — Independent confirmation that AI agents cannot finish real-life work, failing on nearly all tasks beyond narrow domains.
https://www.yahoo.com/news/articles/ai-still-fails-completing-real-173358385.html
RheoData (2025) — Consolidated statistics showing extremely low reliability of generative AI in real-world workflows.
“Paper Ceiling” Data (2025) — Widespread reporting shows double-digit declines (≈16%) in entry-level tech hiring, validating concerns about the collapsing talent pipeline.
(Representative coverage available on WSJ, Insider, CompTIA.)
III. Human Cost & Algorithmic Harm
Innocence Project (2025) — Documentation of multiple wrongful arrests from facial-recognition systems in the US, including the Trevis Williams case.
https://innocenceproject.org/news/when-artificial-intelligence-gets-it-wrong/
Law Quadrangle (University of Michigan) — Deep dive into flawed facial-recognition technology leading to a historic legal settlement.
CriticalTake (2025) — Compilation of police misuse of facial recognition, showing consistent racial bias.
AI Incident Database — Documented incidents of wrongful arrests due to faulty AI matches.
The Independent (2025–26) — AI-assisted surgical systems misidentifying body parts, causing strokes, arterial damage, and wrong‑site operations.
MedBound Times (2025) — AI navigation systems linked to dozens of botched surgical outcomes.
https://www.medboundtimes.com/medicine/ai-operating-rooms-safety-concerns-botched-surgeries
Yahoo News (2025) — Summary of widespread AI-driven surgical navigation errors.
https://www.yahoo.com/news/articles/ai-operating-room-botched-surgeries-193131330.html
Futurism (2025) — Google’s medical AI “invented” anatomical structures that do not exist in humans.
https://futurism.com/neoscope/google-healthcare-ai-makes-up-body-part
IV. Academic Integrity & Social Friction
Associated Press (2024) — AI is making plagiarism detection nearly impossible as students blend ChatGPT and Grammarly into essays.
https://apnews.com/article/ai-cheating-school-chatgpt-4f89a552e9093ce2180471b4d4736675
The Guardian (2025) — AI cheating cases tripled in a single academic year across UK universities.
NBC News (2025) — Students now use AI to evade AI plagiarism detectors, escalating the arms race.
https://www.nbcnews.com/tech/internet/college-students-ai-cheating-detectors-humanizers-rcna253878
US News (2025) — Schools are redefining what counts as cheating because of AI ubiquity.
BBC / Douglas Rushkoff (2024–25) — Investigation into billionaire “bunker mentality,” citing fears of social unrest and inequality exacerbated by AI-led disruption.
https://www.bbc.com/future/article/20240115-why-the-super-rich-are-prepping-for-apocalypse