Every morning, I wake up, check the news, and discover that artificial intelligence has apparently moved one step closer to destroying civilization before I have finished my coffee.
By breakfast, AI has eliminated every job. By lunch, it has become conscious. By dinner, it has manipulated an election, written a screenplay, impersonated someone’s grandmother, crashed the stock market, replaced the family doctor, and developed a private ambition to turn humanity into decorative furniture.
Then I open an AI application and ask it to format a table.
It forgets a column.
This is the strange moment we are living through. Artificial intelligence is simultaneously presented as an all-knowing digital god and an unreliable intern who needs detailed instructions, constant supervision, and three attempts to complete a basic task. It is going to cure cancer, end employment, personalize education, abolish creativity, make everyone rich, make everyone obsolete, democratize expertise, and concentrate all power in the hands of six companies.
Apparently, nuance was the first job AI replaced.
I understand why people are anxious. AI is improving quickly, arriving everywhere at once, and being promoted by corporations that tend to describe every new feature as a historic turning point. The technology is already changing how people write, code, search, create images, analyze data, serve customers, and make decisions. Some workers are watching parts of their jobs become automated in real time. Artists see systems trained on enormous collections of human work. Parents wonder what education means when a student can generate an essay in seconds. Voters face synthetic voices and fabricated videos. Ordinary people are asked to trust tools whose internal operations even their creators cannot always explain clearly.
Those are legitimate concerns.
But concern is not the same thing as hysteria, and lately we seem determined to confuse the two. We have turned AI into a screen onto which everyone projects a favorite fantasy or fear. Executives see infinite productivity. Investors see infinite growth. Doomsayers see extinction. Consultants see billable hours. Politicians see a speech topic they can discuss confidently without understanding. Social-media personalities see twelve alarming videos before lunch.
I see a powerful, uneven, frequently impressive technology surrounded by an industrial quantity of theater.
No, I do not think the sky is falling. I do think several roof tiles are loose, the weather forecast looks unpleasant, and the people selling umbrellas have purchased most of the television advertising.
That is a different problem.
We Have Seen This Movie Before, Although This Version Has Better Graphics
Human beings have a reliable response to transformative technology: we exaggerate what it will do tomorrow, misunderstand what it is doing today, and eventually forget how strange it once seemed.
The printing press threatened established authority and supposedly encouraged intellectual chaos. Trains were accused of moving at speeds the human body could not tolerate. Radio was going to elevate civilization and rot the public mind, sometimes in the same week. Television would educate the world and destroy attention. The internet would eliminate gatekeepers, spread universal knowledge, liberate human connection, and usher in a borderless golden age.
Then it gave us password-reset emails, conspiracy forums, targeted advertising, and arguments with strangers who have cartoon frogs for profile pictures.
Technological revolutions rarely arrive as pure salvation or pure catastrophe. They redistribute power, alter habits, create opportunities, eliminate familiar roles, and generate problems nobody included in the launch presentation. Society adapts badly, then partially, then so completely that the new arrangement begins to look inevitable.
AI will probably follow that pattern, but “probably” is doing important work. History offers perspective, not immunity. Saying earlier generations worried about earlier technologies does not prove current fears are foolish. Sometimes the alarm is connected to an actual fire. Nuclear weapons were not merely a more efficient printing press.
Still, historical memory should protect us from believing every breathless forecast. Whenever someone tells me AI will replace all human labor within eighteen months, I remember that businesses still use fax machines. There are companies that cannot make the “contact us” button work. A large organization can spend millions on digital transformation and somehow produce a website that asks customers to call during business hours.
The future may be arriving at extraordinary speed, but it still has to pass through procurement.
The Marketing Department Would Like a Word With Your Nervous System
Part of the panic comes from the way AI is sold. Technology companies need investors, customers, talented employees, political influence, and favorable regulation. Modest descriptions do not attract those things.
No chief executive earns a magazine cover by announcing, “Our new system may improve selected office tasks under appropriate supervision.” The preferred message is that everything has changed, nobody understands how much, and anyone who fails to buy immediately will be left behind with the other historical losers.
Fear is not an accidental byproduct of AI marketing. It is one of the products.
The industry benefits when businesses believe adoption is unavoidable. Consultants benefit when managers feel unprepared. Media outlets benefit when readers feel alarmed. Influencers benefit when every model update becomes either the birth of a new species or the final insult to human creativity. Even some prominent warnings can raise the perceived importance of the technology and, by extension, the people building it.
If I announce that my toaster may eventually threaten civilization, you might question my toaster. You might also conclude that I have built an exceptionally powerful toaster.
This does not mean warnings are insincere or risks are imaginary. It means incentives matter. I become cautious whenever the same people asking society to admire the power of their systems also ask society to trust them to define the danger, design the solution, and remain in charge afterward.
The rhetoric creates a convenient double identity. When seeking investment, the model is revolutionary. When accused of harming someone, it is merely a tool. When regulators approach, it is too complex for ordinary rules. When copyright holders object, it learns just like a human. When users expect accuracy, the disclaimer explains that it may make mistakes.
That is quite a résumé. It possesses world-changing intelligence and the legal responsibility of a decorative lamp.
The Robots Are Coming for Our Jobs, Starting With the Junior Ones
Employment is where public anxiety becomes personal. People can tolerate abstract discussions about machine consciousness. Tell them software may eliminate the job paying their mortgage and philosophy quickly becomes a budgeting problem.
The most dramatic forecasts imagine mass unemployment arriving like a tidal wave: accountants, programmers, writers, paralegals, designers, analysts, teachers, customer-service representatives, and perhaps everyone who uses a keyboard swept away at once.
I do not dismiss the threat. Stanford’s 2026 AI Index reports that AI’s labor effects are appearing unevenly, with the greatest pressure visible in hiring pipelines and among younger workers in exposed occupations. It notes a sharp decline in employment for software developers ages 22 to 25 since 2024. It also reports that roughly one-third of surveyed organizations expect AI to reduce their workforces in the coming year, even though broad economy-wide job losses have not yet appeared.
That is not the end of work. It is not nothing, either.
The early-career problem deserves special attention. Companies often imagine that AI can perform the routine assignments previously given to junior employees. That may save money now, but it creates a delicious management puzzle later: where will experienced employees come from if nobody hires beginners?
You cannot automate the bottom rung of a career ladder and then act surprised when nobody can climb it.
I expect AI to replace tasks faster than it replaces entire occupations. Most jobs are bundles of activities: some repetitive, some social, some physical, some judgment-heavy, and some so oddly specific that no software company will bother automating them. A tool might draft the report without attending the difficult meeting, noticing the client’s hesitation, defending the conclusion, accepting responsibility, or remembering that the chief financial officer hates the color orange.
But task automation still affects employment. If AI allows ten people to do the work once performed by fifteen, the five missing jobs do not feel reassured by a lecture on occupational bundles. Productivity gains can create new demand and new roles, but those benefits may arrive in different places, at different times, and for different people.
The correct response is neither “every job will disappear” nor “technology always creates more jobs.” Both statements replace analysis with slogans. The real questions are which tasks change, who owns the tools, who receives the gains, how quickly workers can adapt, and whether society helps people through the transition or sends them a motivational webinar.
It Is Not Conscious Just Because It Used a Semicolon
One of the stranger features of AI panic is how easily fluent language convinces us there must be a mind behind it. Humans are built to detect intention. We see faces in clouds, personalities in cars, and emotional betrayal in printers. Give a machine a coherent voice and we immediately begin wondering what it wants.
Current language models can produce startlingly humanlike responses. They can explain concepts, imitate styles, revise arguments, solve many problems, and sustain conversations that feel meaningful. That experience should not be trivialized. Language is central to how humans recognize intelligence in one another.
But polished language is not proof of consciousness, desire, self-awareness, or secret plans. A system can generate the sentence “I am afraid” without experiencing fear, just as a novel can contain grief without the paper becoming sad.
I am open to the possibility that future systems may force us to rethink consciousness. I am less impressed by people who claim the question has already been settled because a chatbot sounded moody at two in the morning.
The danger of anthropomorphism cuts both ways. We may grant a machine too much authority because it sounds wise, or imagine motives where there are only patterns produced by training and design. Calling an AI “evil” can distract from the human institutions deploying it. The mortgage algorithm does not need hatred in its heart to discriminate. It does not have a heart. The company using it still has obligations.
Sometimes “the AI did it” is a modern version of “mistakes were made”—a grammatical fog that allows every responsible person to quietly leave the room.
The Models Really Do Fail, and They Fail With Excellent Grammar
Anyone attempting to calm AI fears should resist the temptation to pretend the technology is more dependable than it is. Current systems can fabricate sources, misunderstand instructions, reproduce bias, expose private information, assist fraud, and produce incorrect answers with impressive confidence.
The 2026 AI Index reports that documented AI incidents rose from 233 in 2024 to 362 in 2025. It also describes wide variation in hallucination performance across leading models and warns that responsible-AI evaluation is not keeping pace with capability.
That does not prove we are approaching extinction. It proves we are deploying imperfect systems faster than we are learning to govern them.
Context matters. If AI invents a fact while helping me brainstorm names for a fictional tavern, no civilization-level emergency has occurred. If it invents a legal precedent, medical dosage, financial disclosure, or structural-engineering calculation, the error becomes serious.
Risk is not a property of the model alone. It is a relationship among the tool, the task, the user, and the consequences of failure. We do not need the same safeguards for generating birthday-card ideas and allocating medical treatment.
This sounds obvious, which is why policy discussions often avoid it.
We need risk-based rules, independent evaluation, incident reporting, security testing, meaningful privacy protections, and clear accountability. The National Institute of Standards and Technology’s AI Risk Management Framework offers a practical foundation by emphasizing the need to govern, map, measure, and manage risks across a system’s life cycle. It is less cinematic than arguing about robot uprisings, but it has the advantage of being useful.
The public conversation spends too much time imagining whether AI will develop a desire to conquer humanity and too little asking whether companies can be held responsible for harms occurring next quarter.
The hypothetical superintelligence receives all the dramatic lighting. The automated benefits system quietly denies someone’s claim.
Deepfakes: Because Regular Lying Was Apparently Too Labor-Intensive
AI lowers the cost of producing convincing false material. Images, audio, and video can be generated or altered with increasing realism. Fraudsters can imitate voices. Political operatives can manufacture clips. Harassers can create intimate images without consent. A fabricated event can travel around the world before a careful correction has located its reading glasses.
This is a real problem, but even here the popular imagination sometimes runs ahead of human behavior. People did not begin believing nonsense when image generators arrived. They believed edited photographs, misleading headlines, forged documents, partisan rumors, and the confident person at the bar.
AI does not invent deception. It industrializes it.
The greatest danger may not be that everyone believes every fake. It may be that people stop believing anything. Once synthetic media becomes common, authentic evidence can be dismissed as generated. A politician caught on a real recording can simply announce that it is fake. This “liar’s dividend” turns uncertainty into protection for the guilty.
The answer is not to panic whenever an image looks unusual. It is to strengthen authentication, provenance, platform response, media literacy, law enforcement, and penalties for harmful impersonation. We must learn to treat viral media as a claim requiring verification rather than a divine revelation delivered in vertical video.
That cultural adjustment will be irritating. It will also be necessary.
What About the End of Humanity?
Now we reach the deluxe package of AI anxiety: extinction.
Some researchers and industry leaders argue that sufficiently advanced systems could eventually become difficult to control, pursue goals in unexpected ways, enable catastrophic weapons, or outmaneuver human institutions. These warnings deserve serious study. The consequences are too large to dismiss merely because the scenarios sound like science fiction. Aviation safety investigates unlikely failures precisely because crashes are costly.
But uncertainty should be admitted in both directions. Nobody knows whether current approaches will produce a system capable of autonomous, strategic, civilization-scale action. Nobody knows the timeline. Nobody knows which proposed safeguards will remain effective as systems improve. Confidence—utopian or apocalyptic—is often performing far beyond the evidence.
I reject two equally lazy positions. The first says extinction risk is ridiculous because present systems make silly mistakes. A technology can be limited now and dangerous later. The second treats speculative catastrophe as a settled forecast and every new benchmark as another trumpet from the Book of Revelation.
We can take low-probability, high-impact risks seriously without living inside a permanent emergency trailer.
That means funding alignment and safety research, requiring evaluations for frontier systems, securing powerful models against theft and misuse, coordinating internationally, and establishing clear thresholds for heightened oversight. It also means refusing to let distant hypothetical dangers eclipse immediate harms involving labor, surveillance, discrimination, fraud, copyright, energy use, and political power.
If someone insists the only AI issue worth discussing is extinction, I start wondering which present-day business model they would prefer I not inspect.
Panic Is a Terrible Regulator, but So Is Faith
Public fear can produce bad policy. Legislators under pressure may write sweeping rules they do not understand, protecting established companies that can afford compliance while crushing smaller competitors. Schools may ban useful tools instead of teaching responsible use. Organizations may avoid beneficial systems because leaders fear headlines more than failure.
Blind optimism produces bad policy too. Companies may deploy systems before testing them, collect data without meaningful consent, and call voluntary promises “governance.” Governments may treat national competition as a reason to remove every speed limit. Employers may install AI simply because shareholders expect the word to appear in presentations.
The sensible position is annoyingly conditional.
Use AI where evidence shows it helps. Require human review where errors matter. Test systems in the environments where they will operate. Inform people when automated decisions affect them. Give them a way to appeal. Protect personal data. Audit high-impact applications. Report serious incidents. Assign legal responsibility before harm occurs. Update rules as evidence changes.
None of that fits neatly on a protest sign or venture-capital podcast.
It may still work.
My Personal Rule: Follow the Incentives, Then Check the Evidence
When I encounter an extraordinary AI claim, I ask four questions.
First, who benefits if I believe it? A company announcing human-level performance may be raising money. A consultant warning that every business must transform may sell transformations. A public figure predicting doom may gain attention, influence, or authority. Incentives do not automatically invalidate a claim, but they tell me where to examine it closely.
Second, what exactly happened? “AI passed a test” is less informative than it sounds. Was the test public? Was the model trained on similar material? Were humans allowed tools? How were failures counted? Can independent researchers reproduce the result?
Third, what is the baseline? If AI makes errors, how often do qualified humans make comparable errors under real conditions? If automation eliminates jobs, what is happening across the wider labor market? If a model uses significant energy, how does that usage compare with alternatives and what improvements are underway?
Fourth, what would change my mind? If the answer is nothing, I am not evaluating evidence. I am defending an identity.
This approach is slower than panic and less emotionally satisfying than certainty. It also saves me from treating every demo as destiny.
The Sky Is Still There—For Now
I do not believe AI is fake, trivial, or merely another technology cycle. It is already useful, disruptive, and capable of altering institutions faster than those institutions usually adapt. The gap between AI capability and responsible governance is real. So are the incidents, job pressures, security threats, and power imbalances.
But the loudest version of the story is rarely the most accurate.
AI is not taking every job tomorrow. A chatbot producing fluent sentences has not proved it possesses a soul. Every generated image is not the death of art. Every model update is not a step toward human extinction. And every concern is not ignorant resistance from people who failed to appreciate innovation.
Reality is messier. AI will eliminate some roles, create others, and transform many more. It will help people produce remarkable work and help other people produce remarkable quantities of garbage. It will expand access to expertise and manufacture convincing misinformation. It will distribute power in some contexts and concentrate it in others. It will reflect human intelligence, human bias, human ambition, and the human tradition of releasing a product before customer support is ready.
The sky is not falling.
It is changing color, and a great many people are trying to sell us their interpretation of the weather.
I intend to keep looking up. I also intend to check who owns the forecast, who funded the satellite, who profits from the storm warning, and whether the person screaming about the apocalypse has a premium survival course linked in the description.
We do not need panic. We need attention.
We do not need worship. We need scrutiny.
We do not need to choose between racing blindly into the future and hiding under the nearest desk. We can experiment, regulate, adapt, challenge exaggerated claims, and demand evidence. We can enjoy what the tools do well while refusing to hand them authority they have not earned.
That position will disappoint the evangelists and the prophets of doom, which is one reason I like it.
If the machines eventually become conscious, they may even appreciate the nuance.
Assuming they remember all the columns in the table.
Sources and Further Reading
Stanford Institute for Human-Centered Artificial Intelligence, The 2026 AI Index Report.
Stanford HAI, Economy: The 2026 AI Index Report.
Stanford HAI, Responsible AI: The 2026 AI Index Report.
National Institute of Standards and Technology, AI Risk Management Framework.
This article is opinion and analysis.