Transcription
We just spent 285 billion dollars building what people are calling a digital god that can solve the International Math Olympiad problems in seconds. And yet, show it a wall clock, and it gets it wrong about half the time. That should worry you. Not in a “this is interesting” way… but in a “we might not actually understand what we just created” way.
Because right now, there’s a delusion spreading through Silicon Valley. 73% percent of insiders think AI will change everything. Only 10% of the public wants it. So which is it? The greatest breakthrough in human history… or a multi-billion dollar misunderstanding of what people actually need?
This is why AI trust is collapsing. Ever since AI burst onto the scene with bold promises, it’s been a strange and uneven journey for early adopters. It works as a highly advanced Mechanical Turk, programmed to accomplish specific tasks - but whether it passes muster depends specifically on the success conditions. In tasks with a highly specific and objective condition, it’s incredibly good. The software has been shown to pass the bar exam - an incredibly challenging legal quiz that would-be lawyers study for months for - with ease, simply scanning all its accessible data for the answers.
But for lawyers who use it to create depositions and legal papers, they’ve often been in for unpleasant surprises. AI has been found to hallucinate fictional cases to make the case for its side - leading more than one lawyer to be reprimanded by the judge or even threatened with disbarment for trying to present a fraudulent case based on AI.
So why does AI get so many things right - and so many wrong? AI experts have studied the system’s ups and downs, and discovered some shocking shortfalls. It fails to identify simple facts like counting the number of “r”s in the word “strawberry”, or incorrectly identifies the time shown on an analog clock - a common error made 50% of the time by models. And sometimes, these mistakes can be dangerous. Google’s AI summaries responded to a joke query of “How many rocks should I eat?” by telling people to eat one rock a day, based on an article from satire website The Onion. AI gets the big things right, but often gets tripped up on the little things. It’s called the “Jagged Frontier” and it calls everything into question.
As of February 2026, $285.9 billion had been invested in AI in the US alone, seeking to enhance the technology and compete in what’s essentially becoming a tech arms race. A yearly investment of $25.2 billion in Generative AI is the cost of building 17 Burj Khalifa’s. But the progress for AI hasn’t been smooth. Huge gains are countered with sudden setbacks. Ethan Mollick, the writer of the paper identifying this problem, argued that AI simply doesn’t learn the same way humans do. It makes the same mistake repeatedly. And as people charge full-speed ahead into an AI-driven future, those unexpected shortfalls could be dangerous. If the tech powers have their way, it’ll be in every single part of our lives.
But now, people are saying “enough”. Talk to anyone about AI and you’ll hear a divided story. On one side are the early adopters, the people who can’t stop talking about how it’s transformed their work, their creativity, even their daily lives. On the other, a quieter but rising group is beginning to push back. People who are proud to have never used it, accusing every query of stealing art and poisoning the environment. Like every controversial issue, both sides have vocal activists and it seems evenly split. But the reality of the data paints a very different picture.
The Stanford AI Index has been tracking people’s views of artificial intelligence both in and out of the industry since 2017. Every year a more dramatic shift emerges. Right now, there’s one group that believes fully that AI is on the upswing, and the future only means good things. The problem is… those are the people involved in AI. 73% of experts expect a positive impact from further investment in AI. They have been consistently pitching it to companies both in and out of the tech sector as the solution to their problems.
But the general public paints a very different picture. When you head into the larger audience, AI support doesn’t just fall, it drops off a cliff. Only 23% of Americans see a positive impact from AI, with most worried about a massive loss of jobs as AI automates one industry after another. Only 10% say they’re more excited than worried. That’s lower than the number of people who think the moon landing is fake. This isn’t just a difference of opinion. It’s the biggest disconnect between the public and financial elites since 2008, when blind confidence helped trigger the worst crash since the Great Depression. And the experts still aren’t worried. They believe everything will fix itself… once AI hits its next milestone. But this isn’t just optimism. It’s an echo chamber.
When you spend enough time looking at stories about the economy, markets, and where everything might be headed, it’s easy to get stuck in this mindset where you feel like you need to have everything figured out immediately. Like if you can’t solve the whole picture at once, you just stay stressed, overthink it, and don’t know where to start. And I think that’s something a lot of people can relate to, because life is like that too. Sometimes the pressure isn’t one huge dramatic thing, it’s just the constant weight of uncertainty, stress, and feeling like you’re supposed to have all the answers right now. That’s one reason therapy can be so valuable. It gives you space to slow down, sort through what you’re feeling, and take one step at a time instead of treating everything like it has to be all-or-nothing. Therapy has helped me take what feels like a huge problem and break it down into something a lot more manageable. It’s helped me focus on what I can do right now and make things less overwhelming. And that’s why I’m glad BetterHelp is sponsoring this video. BetterHelp makes starting therapy easier. You can take a quick quiz and get matched with a licensed therapist, and if it doesn’t feel like the right fit, you can switch anytime at no extra cost. You can also communicate in whatever way feels most comfortable for you, whether that’s phone, video, or text. Sometimes progress isn’t about fixing everything at once. Sometimes it’s just about having the support and perspective to move forward a little more clearly. So if you’ve been thinking about trying therapy, click the link in the description or go to betterhelp.com/infographics to get 10% off your first month of therapy.
AI experts point to impressive feats like AI ripping its way through a math olympiad, solving PHD-level science questions, or analyzing statistical data. And in these areas, AI has been showing incredible progress. AI agents handling cybersecurity issues showed a 93% success rate in 2025, up from only 15% the previous year. But just because AI is performing, doesn’t mean it’s thinking. Tech experts love to say the doubters are just stubborn. Resistance is just temporary and everyone falls in line once it becomes unavoidable. But what if that assumption is wrong? What if this time… is different?
From day one, AI companies have been selling a vision. Not just better tools, but a quantum leap in intelligence. They tap into something people already recognize… the version of AI we’ve seen in movies. An all-powerful system. Fully autonomous. Capable of reasoning, deciding, even running the world better than humans. And at the center of all this is one idea: the singularity. The moment AI surpasses human intelligence entirely. And everything hinges on it. The believers say we have to push harder, invest more, because on the other side is a utopia. But the skeptics see something else. They see all the moments AI fails. But what if neither is right? What if AI is simply an illusion of intelligence?
A June 2025 Apple study sent ripples through the tech world when it cast doubt on the entire driving argument for AI. Apple was in a strange position during the AI boom. It was still one of the most powerful tech companies in the world, but its own AI efforts weren’t really delivering results. There were several false starts, and after pressure from investors, it shifted direction. Instead of trying to lead the race alone, it began partnering with other companies to bring AI into its devices. And that gave it something rare in the tech world… Impartiality.
Apple researchers dug into the primary current AI models and experimented to see which problems it could solve and which it ran into trouble with. Their studies confirmed what most people already knew. AI is amazing at determining the right answer from a collection of data, and is able to sort through it faster than any human. But it relies on statistical analysis, and when there isn’t a preponderance of information, its ability to determine truth decreases. And the more steps, the more trouble it finds. One technique the researchers used was to present familiar mathematical problems in new formulations. They didn’t give the AI the chance to use models that were already out there. There was a significant decrease in the AI’s ability to determine the next step, indicating that it relied heavily on pattern recognition rather than on actual knowledge and thinking. And the more steps it had to take to solve a problem, the harder it became for it to stay on track. This became clear when it was tested on the Tower of Hanoi puzzle, a task where you move a stack of discs from one rod to another under strict rules. As the number of discs increased, the AI started to struggle. It hit repeated stumbling blocks… and eventually, it seemed to give up altogether. And that might indicate that the promise of AI isn’t just overstated. It could be an outright lie.
To reach the singularity, AI needs to be able to solve problems like a human can. Researchers pull together data from dozens of sources, compare them against each other, and hold multiple variables in mind while building theories and testing them step by step. As new information comes in, those theories shift and adapt in real time. Chess grandmasters do something similar. They develop hundreds of strategies not just for each formulation of the board, but for each opponent they play. A human’s reasoning is never quite complete, evolving with the moment. But AI isn’t reasoning. It’s simply matching.
Those who think AI is merely a shell game point out that we’ve seen it before… and we didn’t like it. AI is simply a much more advanced version of the autocomplete that tries to write your e-mails for you before you get to type; or the automated customer service agent that makes you go through all the steps before you can talk to an agent. That’s because AI predicts what the next step is likely to be, and then chooses an output based on probability. It draws from enormous datasets across the internet, analyzes patterns in how language is used, and generates responses that statistically fit the prompt. Most of the time, this will resemble a baseline of an acceptable answer. But it doesn’t take much to throw it off.
One of the best illustrations of the problem with AI right now can be seen in a very different kind of AI system. Robotic dogs, used in crowd control, can move with incredible precision when everything goes as expected. Every step is calculated, every motion tightly controlled. But that precision comes with a weakness. Even small disruptions can essentially render them useless until reset. And that’s AI right now. It’s incredibly efficient, convincing autocomplete that’s surprisingly easy to confuse. And many say that’s not worth it… especially given all it costs.
Despite these lingering concerns, there is still a massive appetite for investment in AI. Those pushing the funding narrative have convinced many that the next update, the next model, will fix the current limitations. And that belief comes with a huge cost. Each new generation of models doesn’t just require top-tier researchers, it needs massive amounts of computing power. Even before it rolled out to the public, Google’s Gemini Ultra cost $191 million just to train. That’s the cost of a fleet of two modern F-35 fighter jets, just on one model. That’s just scratching the surface of the cost.
It’s not just that this never-ending project costs money, it also requires an enormous infrastructure investment. The internet and all its associated services rely on data centers to keep them up and running. The onset of AI, and the mass adoption of these services, led to a massive surge in demand. In 2024, there was a 690% increase in need for AI-driven data services, and it’s expected to continue rising by about a third each year. And that’s not just reshaping our internet, it’s reshaping our world.
Data centers are cropping up around the world, taking up a huge amount of available real estate. While the majority of new data centers are abroad, the US holds a total of 43% of the world’s facilities. And these centers use a massive amount of energy, juggling millions of queries a day. That requires heavy electricity and water use, leading to the communities where they’re built seeing heavier emissions, poorer air quality, and even some water shortages. Anti-AI activists have been pushing for a moratorium on data centers, citing the environmental impact. But the tech sector has pointed out that each AI query has a tiny impact, far less of a carbon footprint than eating a hamburger or driving a car. The problem is, no one is making just one query. The system has become a constant companion for many that it adds up quickly. In fact, the cost of training Grok 4 was estimated to be equivalent to the emissions of 17,000 cars over a full year.
And it’s not the only place where people are feeling the pain. Advocates against AI expansion may have an unusual ally… gamers. Despite usually favoring advances in technology, video game fans are experiencing the most direct consequences of the AI surge. The cost of DDR4 has shot up by over 2000% over the last year, due to the massive demand from AI firms. Even if gamers can afford it, they may not be able to buy it. Some of the top companies that make RAM chips have gone out of the individual sales business all together. There’s a growing perception that Americans are being asked to make major sacrifices on a personal and environmental level in order to win the AI arms race.
But the thing is, we might not even be doing that. The top AI firms in the world are all American, including Open AI and Anthropic, as well as larger tech conglomerates like Google and X. When it comes to investments, there isn’t even a close second place. The United States private sector invested over $470 billion in AI research between 2013 and 2024, with the numbers each year skyrocketing after that. In 2024 alone, the US invested $109 billion, compared to China’s $9.3 billion. And all that might not even land the US in the top twenty.
In terms of where new AI technology is coming from, the United States is still at the front of the pack. Countries like China, Japan, and South Korea are competing heavily, but US firms have maintained a clear lead in both capability and scale. Other countries are now actively looking to partner, license, or invest in American AI companies just to keep up with how fast the technology is advancing. But they might wind up being a better investment in the long run, because the US lags well behind the pack in terms of acceptance.
Only 28.3% of the US population say they use AI regularly in their work duties. According to the Stanford AI Index, which tracks AI diffusion based on the share of people regularly using generative AI at work, the United States actually ranks around 24th. And that puts them well behind the pack. Across the world, other countries incorporate AI into their daily routines far faster than Americans. Some of the top countries include Ireland, Norway, and France, with rates in the 40s. Singapore hovers at 60.9%, while the oil-rich United Arab Emirates continues its big tech push with a 64% adoption rate. These countries see the promise in AI, and generally view it much more positively than Americans.
Which raises the question for tech executives - what’s going wrong? It’s not a lack of access. All the core AI technology is concentrated in the United States. Nor is it lack of affordability, as AI has been offered to the public for free ever since the debut of Chat-GPT in 2022. While most now offer premium subscriptions for heavy users, especially those who want to generate video and images, most functions are available for free at the click of a button. But there might be a cultural divide that’s far harder to overcome.
In Europe, some of the strongest adoption of AI has emerged in countries with a deep tradition of strong unions and labor protections. In those places, automation is often framed differently. Rather than being seen purely as a threat to jobs, it’s increasingly viewed as a tool that could reduce workload, streamline repetitive tasks, and ultimately improve working conditions. In tech-oriented cultures like South Korea, the media is primed to portray new technology in a positive way. However, in the United States, the picture looks a lot less rosy. And people aren’t just worried about it stealing their future. The United States has a large population and a comparatively weaker social safety net than much of Europe. As a result, many people tend to view AI less as a productivity boost and more as a direct threat to their livelihood. That concern has only intensified as companies experiment with replacing workers - especially in customer service and support roles - with AI chatbots. In several cases, those rollouts have delivered mixed results, with firms later scaling back the automation and rehiring human staff to fill the gaps.
There’s only one thing worse for a new tech innovation than looking like a problem… looking like a loser. And to many ordinary Americans right now, AI looks like both. Many investors warn that we might be reaching “peak AI”. Heavy investment is likely to continue, but the public isn’t being won over. A minority of Americans is increasingly into AI, but they haven’t shown the ability to branch out beyond that audience. And as more comes out about the technology’s limitations, the fear of the “jagged frontier” may start to chill investment. After all, is any major company going to invest their infrastructure in a technology that can’t think, only predict according to what it thinks the answer is? Maybe not… if they’re actually listening.
The problem facing the AI market right now is that many of its top proponents are within a bubble. They’re spending so much time with their fellow believers, and with the AI technology itself, that they can’t see the warning signs. And that means they’ve built a house of cards that could collapse at any time. A bad report on the progress of a new AI model, or a disaster caused by bad advice from one of the models, could lead to a mass sell-off that would see the AI bubble burst. Countless companies lose everything, leading to the worst economic crisis since 2008.
But that might not be the worst-case scenario. What keeps AI skeptics up at night is a world where AI adoption continues full speed ahead. Companies and governments heavily incorporate it into essential infrastructure, trusting it to make millions of decisions a day. And then something goes wrong. A stumbling block turns into a cascading series of events that could potentially crash the economy, the internet, or the power grid as the AI tries to unscramble its flawed logic. And this is an all-too-realistic situation. After all, would you trust your country’s power grid to a technology that still can’t read an analog clock? Some sectors aren’t waiting for AI to be “perfect” before deploying it. They’re pushing ahead anyway. Governments are integrating AI into military planning and the results are ominous. Find out.