Some of the largest companies in the world are spending like they’ve already seen the future — a future where machines think, reason, and create as well as any human. Microsoft is pouring billions into AI data centers. Google is buying chips by the ton. Enormous electrical grids are being built to power the computer. Nvidia’s valuation has soared past $3 trillion on the belief that its processors are the keys to artificial intelligence.
The problem is, there’s no evidence that any of this is actually true. Let me explain.
The AI industry, centered primarily in San Francisco, believes in a simple premise: keep scaling the existing systems — feed them more data, give them more processing power, train them for longer — and they will eventually develop human-level reasoning and be able to think like humans. That’s the entire assumption behind today’s gold rush and all the predictions of human-level intelligence by much of the industry. It’s the premise for huge valuations, biliions upon billions in investments, and the million dollar salaries being offered to some engineers.
But what if scaling doesn’t work?
In the current issue of The New Yorker, David Marcus, an NYU professor, asks the question Silicon Valley has ignored: “What if A.I. doesn’t get much better than this?”
The most advanced models today — GPT-5, Claude, Gemini and others— are more polished than their predecessors, but are still prone to the same hallucinations, gaps in logic, and failures to understand context.
AI today is more like a huge search engine with access to an increasing amount of data. It does a great job in searching but can’t reason. As an example, ChatGPT 5, just released to claims of having some intelligence, has failed to meet expectations; it still makes things up and still shows no ability to reason like humans.
Ed Zitron, a relentless critic of AI hype, explains: “Reasoning models are not actually reasoning,” he told MarketWatch. “They do not have human thought, or any thought. They are just large language models [LLMs] that spit out answers based on what the user wants.” In his view, the scaling narrative is “a marketing ploy” designed to keep the huge investments coming.
This viewpoint presents a big issue for companies already deep into the most aggressive and expensive tech buildout in history. Microsoft, Meta, Amazon, Google, and others, are pouring billions into building the needed infrastructure: data centers, fiber, and even nuclear plants.
The fact is that the design of today’s AI products haven’t changed: they still don’t “know” anything. They still work the same way: predict the next word in a sequence based on patterns in training data. They have no grounding in the physical world, no truth-checking, no real comprehension, no verification process. As Zitron points out, “They’re not reasoning — they’re regurgitating.”
Should scaling fail, the fallout could be staggering. Instead of creating a new form of intelligence, we may have constructed one of the most expensive illusions in history — an infrastructure empire built on an assumption that never materializes.
Nevertheless, AI software like ChatGPT and Gemini are still great products for accomplishing tasks that don’t require human intelligence. We can use them like a powerful search engine or a research vehicle that can plow through data and provide us what might take a researcher hours to do.
And now for those that use Gmail and Google Calendar, there’s a new benefit when you allow Google Gemini access to your email, calendar, and documents. You can ask Gemeni things that are in your email, on your calendar, and use these resources to solve problems.
For example, I asked it to look through my email to find all correspondence with a company that I sell my wine maps to (www.wine-maps.com). It looked through my mail over the fours year of the business relationship and provided a report summary, from the initial agreement to every time we corresponded about new orders and invoices. This is what it provided:
Here is a summary of your email correspondence with XXXX:
Business Relationship
- You and XXXX, who works for a business called “YYYYY,” have an ongoing business relationship involving the sale of “Napa wine posters”.
- You have offered her an exclusive deal to sell the posters.
- Orders and Invoices
- Most Recent (May 2025): You emailed her to check if she had sufficient posters in stock before you went on a two-week trip starting June 3, 2025.
- October 2024: YYYYY asked to place her normal reorder and inquired about an invoice from August, which you confirmed was paid by check.
- May 2024: You checked if she was still selling the posters, and she responded that she would like to and wanted to start with 12 posters. You later sent an invoice for a June 2024 order.
- September 2022: YYYYY reordered 24 posters, and you sent her an invoice.
- July 2022: She placed a reorder for the “same quantities” and you sent an invoice for 24 posters that were shipped on July 15, 2022.
- April 2022: She requested a reorder, and you confirmed an order for 24 posters.
- January 2022: She notified you that one of the posters she received was damaged, and you offered to send a replacement.
- August 2021: You shipped 24 posters and sent an invoice for $470. After a reminder, she confirmed payment in September 2021.
- June 2021: You confirmed an order for 24 posters to be shipped around June 29.
- April 2021: You shipped posters and sent an invoice.
- Other topics
- There were several instances where you followed up on outstanding payments and he confirmed he had paid or was about to pay.
Gemini can also respond to natural enquiries such as “Find the contract email from Vendor X sent in 2019 with the updated terms.” Within seconds, Gemini can show you the most likely emails, highlighting excerpts that match your query, such as:“From: Vendor X, Sent: March 12, 2019, Subject: Updated Contract Terms 2019, Attachment: Contract_VendorX_2019.pdf”.
You can then ask Gemini to summarize the key points of the contract, extract deadlines, or even create a calendar event based on contract milestones. Essentially, Gemini acts as a smart, AI-powered search assistant that can comb through your entire Gmail history without requiring exact keywords, making it much faster than traditional manual search.
In another clever use of AI, Google just announced that it’s testing a new AI-powered Flight Deals tool, a tool that let’s you plan a trip to a place based on your budget. You describe the kind of destination you are interested in, how long a trip and your preferred activities. Google then uses AI to to find cheap flights and hotels that match your goals.
There is no doubt that AI is here and available to help us in many different ways. It has just not progressed to being able to perform human reasoning or human-like thinking. That should be a big concern for the AI industry, but maybe it’s best for the rest of us.