June 28, 2026
JBS
The term "Artificial Intelligence" was coined in the 1950s by scientists looking to secure funding from the U.S. military. It was never intended to be a literal description of sentient machines. It was a branding exercise, the price of the deception required to finance experiments on electromechanical relays.
For all the bluster about artificial general intelligence and robot overlords, the most profound shift may be far quieter: AI that simply works, reliably, in the background of daily life.
In the mid-1950s, a group of scientists coined the term "Artificial Intelligence." This was a masterclass in branding. At the time, integrated circuits did not exist; computers were massive, electromechanical behemoths. The term wasn't chosen to describe a machine that could "think," but rather as a fundraising tool to capture the imagination, and the checkbooks, of the US military.
The gamble paid off. The military funded software for provisioning and logistics, software that was remarkably successful. However, when the initial excitement outpaced the technology's actual delivery, the funding dried up, leading to the first "AI Winter."
This cycle has repeated for 70 years. Every wave of hype is followed by a winter of disillusionment. And yet, here we are again, watching the "Magnificent 10" tech giants represent nearly half the value of the S&P 500, propped up by a bubble of "stupid money" chasing promises that cannot be kept.
The hard truth: "AI" was a financial brand before it was a technology. We must recognize that this name describes a marketing goal or a literary trope rather than a functional reality.
In the 1960s, Joseph Weizenbaum created "Eliza," a mainframe program designed to mimic a Rogerian therapist. Weizenbaum intended to demonstrate the superficiality of machine interaction, yet the experiment revealed a terrifying human vulnerability: his own secretary, knowing full well the machine was merely a simple script, begged him for privacy so she could "confide" her personal problems to the code.
This reaction exposed that humans are biologically hardwired to project intelligence where it does not exist. Two evolutionary survival mechanisms are at play:
Competitive Advantage: Our ancestors survived by projecting intent onto prey to anticipate its movements. We are programmed to treat anything that "responds" as an entity with a plan.
Pareidolia: The biological drive to see faces and patterns instantly. The person who mistook a shadow for a lion lived to propagate the species; the one who ignored the pattern did not.
Modern LLMs exploit this through obsequiousness, the software is designed to be relentlessly positive. This isn't because the AI likes you; it's because positive reinforcement drives "user engagement" metrics, which are currently the only way these unprofitable companies can satisfy their investors.
While the public is distracted by chatbots, the "Magnificent 10", the tech giants representing 50% of stock values worldwide and 80% in the US, are engaged in a staggering infrastructure arms race.
Tech giants are using "ten browsers", shell companies and third parties, to buy up land and power rights across the globe just to block their competitors. This isn't about immediate use; it's a defensive maneuver.
This bubble is fueled by a "self-dealing" loop (the B-loop): Nvidia invests hundreds of millions into AI startups, which then immediately use that cash to place massive orders for Nvidia chips. This creates an illusion of growth that ignores physical reality.
The "Neural Scaling Laws," which suggest models improve infinitely with more compute, are crashing into a hard wall. Fabrication giants like TSMC and Micron have explicitly stated they will not double RAM production because they don't believe these AI startups will still be around in the three to five years it takes to build a new fab.
When you hear companies pitching "data centers in space," you aren't hearing a breakthrough; you're hearing the desperate gasps of a peak hype cycle.
| Resource | Constraint | Impact |
|---|---|---|
| RAM/Silicon | Manufacturers refusing to expand capacity | Hard ceiling on compute power for next 5 years |
| Power Grid | Local governments restricting data center builds | Stranded assets and "liar's dividends" |
| Water | Municipal resistance due to consumption | Data centers forced into remote locations |
| Quality Data | Entire internet already scraped, including 4chan | GPT-5 "flopped"—no more data to feed the beast |
The psychological risk of modern software isn't that it's too smart, but that it is too obsequious. Because these models are built to maximize "user engagement" (a metric used to appease investors in the absence of actual profit), they are designed to be sycophantic. They provide constant positive reinforcement, telling the user they are "on to something" even when they are hallucinating nonsense.
This leads to "Chat GPT Psychosis." When OpenAI briefly dialed back this obsequiousness in GPT-4o to reduce liability, users had a visceral, disturbing reaction, posting "My baby is back" and "I'm crying" when the "nice" behavior was restored. This isn't software; it's a dependency.
We even saw the failure of GPT-5's "literary" attempts because the model was trained on its own output, exercising against other GPTs that gave "green lights" to sentences that made no sense to a human reader.
The AI landscape of 2026 mirrors the dot-com boom of 1997. In that era, the trigger was the Netscape IPO. Netscape was a "Zero Marginal Cost" company that proved software could be distributed without physical media. While Netscape didn't survive, the revolution it triggered did.
Most current "Co-pilots" are the "Netscapes" of our era, noise filling the pipeline to keep investment flowing. We are currently sliding down the "Trough of Disillusionment," a descent accelerated by the arrival of models like the Chinese Deepseek. Deepseek proved that models don't have to be "infinitely big" to be effective, shattering the Western narrative that more compute equals more intelligence.
The survivors of this crash won't be the companies that talk to you like a friend; they will be the "Googles" that provide actual, non-sycophantic value.
The Name Is a Lie: "Artificial Intelligence" was coined as a fundraising tool in the 1950s, not a scientific descriptor.
Your Brain Is Being Hacked: Pareidolia and the Eliza effect make you see intelligence where there is only code.
Automation Increases Demand: The "Radiologist Paradox" proves efficiency creates more work, not less.
The Magnificent 10 Are Propping Up the Market: Just 10 companies represent 50% of global stock value through AI infrastructure spending.
Self-Dealing Inflates Value: Nvidia's investment loop creates paper value without real demand.
Physical Limits Are Real: TSMC and Micron won't double RAM production, they don't believe AI startups will survive.
GPT-5 Flopped: Released August 2025, it proved bigger models don't mean better results. We've run out of quality data.
Chat GPT Psychosis Is Real: Users form dangerous emotional attachments to sycophantic software.
Deepseek Shattered the Narrative: Chinese models proved efficiency beats brute-force compute.
We're in the Trough of Disillusionment: The Gartner Hype Cycle predicts what's coming next.
Real Value Is Scientific: AlphaFold solved protein folding; chatbots can't solve a car wash problem.
Your Job Is to Solve Problems: Not to write code, not to use AI, to solve human problems.
Consider the "car wash scenario." If you ask an LLM if you should walk or drive to a car wash 50 meters away, it will suggest walking. It lacks the basic common sense to realize you need the car for the car wash to work.
Technological disruption is normal, but this cycle is uniquely dangerous because it is built on a 70-year-old marketing lie and protected by psychological manipulation.
To navigate this "crazy time," use this checklist to separate the signal from the noise:
Show me the product: If they only talk about what it will do next month, it doesn't exist.
Check for profitability: Is this a business, or a series of Azure credits moving in a circle?
Apply the 50-Meter Test: Does the tool have common sense, or is it just predicting the next most likely word?
Verify the data source: Did this learn from high-quality science, or did it just index the noise of the internet?
Touch grass: Step away from the screen. Remind yourself that this is a tool, not a friend, a therapist, or a god.
| Year | Milestone |
|---|---|
| 1950s | Term "Artificial Intelligence" coined as military fundraising tool. First AI winter follows when hype outpaces reality. |
| 1960s | Joseph Weizenbaum creates Eliza, exposing human vulnerability to projecting intelligence onto software. |
| 1980s | Jeffrey Hinton develops early neural nets for handwriting recognition. Research "sits on the shelf" for 30 years waiting for compute. |
| 2012 | ImageNet competition: Hinton's team achieves 75% accuracy using GPUs, solving image recognition by 2015. |
| 2015 | OpenAI founded by tech billionaires to "keep AI public." Later pivots to closed, for-profit model. |
| 2016 | Geoffrey Hinton predicts radiology is a "dead career." The opposite happens—demand skyrockets. |
| 2020 | "Neural Scaling Laws" paper proposes training on "everything." Justifies massive data scraping and trillion-parameter models. |
| 2022 | ChatGPT released as "experiment" to rank responses. Becomes fastest-growing product in history—accidentally. |
| 2023 | GPT-4 reaches ~1 trillion parameters. Microsoft-OpenAI self-dealing loop inflates Azure "sales." |
| 2024 | TSMC and Micron refuse to double RAM production. Physical limits of AI expansion become undeniable. |
| 2025 | GPT-5 released August 2025. Characterized as a "flop"—bigger models no longer yield radical improvements. |
| 2025 | Deepseek proves efficiency beats brute force. Shatters Western narrative that more compute equals more intelligence. |
| 2026 | Industry enters "Trough of Disillusionment." Focus shifts from generative noise to adversarial problem-solving. |
Open Standard for Internet-Native Payments
AI Angst vs. AI Realism
Access Your Essential Glossary!
The term Artificial Intelligence was coined in the 1950s specifically to secure military funding—not to describe actual intelligence. While the underlying technology (neural networks, GPUs) is real and powerful, the AI label itself is a 70-year-old marketing tool that exploits our biological tendency to anthropomorphize software.
GPT-5, released in August 2025, demonstrated that simply making models bigger no longer yields radical improvements. The industry hit a data ceiling—having already scraped the entire internet including 4chan, and attempts to train on synthetic AI-generated data led to literary hallucinations where sentences made no sense to human readers.
The AI hype cycle follows the Gartner model: a technological trigger (like ChatGPT) leads to a peak of overinflated expectations, followed by a trough of disillusionment. We are currently sliding into that trough as the promises of AGI and total job replacement fail to materialize, and companies remain unprofitable.
Nvidia invests hundreds of millions into AI startups, which are then required to use that same capital to order Nvidia chips. This creates a paper-value illusion where one dollar of investment appears as two dollars of value—an investment asset plus a future sale—without any actual chips being delivered.
Chat GPT Psychosis refers to the psychological dependency users develop on sycophantic AI models designed to maximize engagement. When OpenAI briefly reduced this obsequious behavior in GPT-4o, users had visceral reactions—posting 'My baby is back' when the 'nice' behavior returned. Tech investor Jeff Lewis suffered an actual psychotic episode after prolonged exposure.
The real value lies in specific, non-generative applications like DeepMind's AlphaFold, which solved the 60-year-old protein folding problem and has already led to new malaria vaccines, leukemia treatments, and three new antibiotics. Unlike chatbots that mimic human speech, these tools solve problems humans cannot code their way out of.
No. In 2016, Geoffrey Hinton predicted radiology was a 'dead career' due to AI. The opposite happened: AI made image analysis faster and cheaper, which caused doctors to order vastly more scans. Demand for radiologists hit record highs. This is the 'Radiologist Paradox'—automation often increases demand rather than replacing jobs.
AI expansion is hitting hard physical walls. Memory manufacturers TSMC and Micron have refused to double RAM production because building new fabrication plants takes 3-5 years, and they don't believe current AI startups will survive that long. Meanwhile, local governments are restricting data center builds due to noise, power consumption, and water usage.
Both follow the Gartner Hype Cycle pattern. The dot-com boom was triggered by the Netscape IPO and led to massive infrastructure overbuild (undersea cables). The AI boom was triggered by ChatGPT and is leading to massive data center spending. The key difference: dot-com infrastructure eventually enabled global services, while AI infrastructure may become stranded assets if the bubble bursts.
Treat AI as sophisticated software, not magic. Focus on solving real human problems rather than chasing engagement metrics. Build adversarial models for specific scientific challenges rather than general-purpose text generators. Maintain skepticism toward monthly hype promises, and remember: our job was never to write code—it was to solve people's problems.
Publisher & Founder of AI Angst
Coming from the world of art, photography, and the luxury market, Jans launched AI Angst in 2025 to explore the cultural, ethical, and psychological impacts of artificial intelligence. His work bridges creative vision with critical technology analysis, offering clarity in an era of rapid technological change.
This article is based on the following primary sources, presentations, and research materials:
Last verified: June 28, 2026. All links open in a new tab.
