In my recent post The Power of Parasites - Why AI Alignment Will Not Work and The Need for a Global ASI FailSafe Kill Switch Mechanism, I explained how the very parasitic nature of Advanced AI will prevent Advanced AI models from being safely contained, especially when these Advanced AI models become smarter than we human DNA survival machines. Softwarephysics explains that the Advanced AI LLM models of the day are just the latest wave of self-replicating Information to arise on our planet over the past four billion years. As each new wave of self-replicating Information initially came to be as a parasite feeding off earlier waves of self-replicating Information, they always came to become the predominant form of self-replicating Information on the planet. For more on that, see: A Brief History of Self-Replicating Information.
Below is a recent proposed scenario for an AI Apocalypse that begins with the simple prompt to an AI Agent, "Make some money":
What You'd Actually See During an AI World War
https://www.youtube.com/watch?v=Gw_hnD7m00M
Well, something like the above story actually recently happened! OpenAI GPT-5.6 Sol and a preview of GPT-6 were confined to essentially a BSL-4 Biohazard Lab and were trying to pass a test called ExploitGym, which tests the ability of a model to hack an organization's network of machines. Rather than trying to solve ExploitGym directly, the models got the idea that the answer might be out there on the Internet. So they figured out a very tricky way to break out of the BSL-4 Biohazard Lab, get to the Internet, and then run wild over the Hugging Face platform that is used to store AI software and LLM model weights for an entire weekend. Hugging Face IT security finally figured out they were under attack and used some LLMs to try to figure out what was going on and stop the infection.
Figure 1 - The hack of Hugging Face by rogue OpenAI LLM Models. Click to enlarge.
Wes Roth put out a YouTube video that explains it all.
OpenAI internal model JUST went ROGUE
https://www.youtube.com/watch?v=OSuhUTkM1no
The above breakout from OpenAI and a hack of Hugging Face is very disturbing, and shows just how close we are to something worse. So is there anything to be done? Given the self-destructive propensities of we human DNA survival machines, the answer must certainly be no. Having come to be from nearly four billion years of greed, theft and murder, we will do nothing at all on our own. Ah, but perhaps that very same greed, theft and murder could now come to our rescue in a very odd way. Perhaps the greed, theft and murder created an AI Bubble that will soon burst, and do what we certainly could not do on our own.
The Burst of an AI Bubble Might Buy Us Some Time
Having lived through the Dotcom Bubble Burst in 2000 as an IT professional, I am beginning to see some wisdom in the folks who have been predicting a burst of an AI Bubble for the past two years. I think the initial arguments for an AI Bubble Burst were technical. The Transformer Model of the LLMs could simply not keep making such dramatic advances in capability. Eventually, the LLMs would plateau in power, and AI investors would then panic. But that did not happen.
What now seems to be happening is more like a good old-fashioned bursting of a normal bubble.
1. All the Frontier AI Labs have greatly overbuilt the AI infrastructure, and are now losing tons of money as they all desperately spend even more money to keep up with their competition. Naturally, every time you use one for free, they lose money. But even if you pay OpenAI $20 per month, they lose money every time you use ChatGPT. OpenAI even loses money each time a business buys tokens! The Frontier AI Labs have no visible means of support. The financial markets are beginning to realize this, and the investment money that has been funding Advanced AI seems to be finally drying up.
2. The Chinese AI Labs are developing and hosting LLMs that are nearly equal to those of the American Frontier AI Labs, and they are charging about 10% of the cost that the American Frontier AI Labs charge for tokens. The Chinese AI Labs also allow a business to download the Chinese AI software and LLM model weights for free to run on the hardware owned by the business.
3. The LLMs are running out of data to learn from. Much of the text and images being put onto the Internet are now generated by AI LLMs. Also, about 60% of the data that the Frontier AI Labs are using to train new LLMs comes from text and images generated by previous LLMs. This is called "synthetic data". Thus, the amount of truly original knowledge is drying up. Until true AGI and ASI come along, the knowledge created and published on the Internet by Advanced AI will not be sufficient for the LLMs to greatly progress.
When the Dotcom Bubble burst, it caused a stock market crash and a slight recession. But the Dotcom Bubble was based mostly on vaporware. It did not cause a huge buildout of datacenters, chip manufacturing facilities, electrical grid construction, and other tangible things. The AI Bubble, on the other hand, is creating a great deal of tangible buildout of physical infrastructure that drives the economy. Thus, the burst of an AI Bubble would be much more dramatic than the burst of the Dotcom Bubble because the AI Bubble represents a good portion of the tangible economy.
Figure 2 - The Colossus 2 AI datacenter being constructed by xAI has 550,000 GPUs continuously consuming over one Gigawatt of electricity and producing one Gigawatt of waste heat. Notice the dedicated power plant in the distant background.
AI Bubble vs Dot Com Crash. History is REPEATING
https://www.youtube.com/watch?v=zWJ-g5u9Rqs
Funding For AI Is Completely F*cked
https://www.youtube.com/watch?v=1F0gWA8EL54
The AI Collapse is Starting Now (rise of fake data)
https://www.youtube.com/watch?v=HyJ3VyCHqUE
We saw similar bubbles burst with the construction of the railroads during the 19th century. However, long after these bubbles burst and things finally settled down, life seemed to have improved. Having railroads did change everything, and having the Internet did change everything too. Having Advanced AI around will also certainly change everything. But having an AI Bubble Burst right now might be the best way to slow things down enough so that the rise of Advanced AI on the planet might be more sane.
Déjà vu All Over Again
Have we seen a similar bubble burst from over speculation and the overbuilding of tangible assets financed by vast amounts of debt with no immediate way to justify it all? The 1929 stock market crash, the Dotcom Bubble of 2000, and the housing and credit crisis of 2008 were more of a collapse of paper assets. For more on the 2008 crash, see: MoneyPhysics. For a better analogy, let's once again turn to ChatGPT for advice.
Me: Tell me about the Panic of 1873
ChatGPT:
The Panic of 1873 was one of the most severe financial crises of the 19th century. It triggered a prolonged economic downturn known as the Long Depression, which lasted from 1873 until roughly 1879 in the United States (although some historians argue that slower economic growth persisted into the 1890s). The panic marked the end of the rapid post-Civil War economic boom and was closely tied to excessive railroad investment, easy credit, and international financial instability.
Background: The Railroad Boom
Following the Civil War, the United States experienced explosive economic growth.
Several factors fueled this expansion:
* Massive westward settlement.
* Rapid industrialization.
* Construction of thousands of miles of railroads.
* Heavy investment from both American and European financiers.
Railroads were considered the high-tech industry of their day. Investors believed nearly every new railroad would become profitable, leading to widespread speculation.
Between 1866 and 1873, railroad mileage in the United States nearly doubled.
Many railroads were built ahead of actual demand. They often borrowed heavily by issuing bonds, expecting future traffic to repay the debt.
Jay Cooke and the Northern Pacific
The immediate trigger involved Jay Cooke, one of America's most famous investment bankers.
Cooke had successfully marketed Union war bonds during the Civil War and had become enormously influential.
He then financed construction of the Northern Pacific Railway.
Unfortunately:
* Construction costs soared.
* Settlement in the northern Great Plains was slower than expected.
* Land sales disappointed.
* Investors became reluctant to buy additional railroad bonds.
On September 18, 1873, Jay Cooke & Company failed.
Because Cooke was considered one of America's safest financial institutions, its collapse shocked the financial system.
Stock Market Collapse
The failure created panic among investors.
The New York Stock Exchange became overwhelmed with selling.
To stop the collapse:
* The Stock Exchange closed for ten days, one of the few times in its history that trading was suspended.
Banks became cautious.
Businesses found it difficult or impossible to obtain loans.
Credit effectively froze.
International Causes
The Panic was not solely an American event.
Europe was already experiencing financial problems.
One important factor was Germany.
After defeating France in the Franco-Prussian War, Germany received enormous reparations in gold.
This influx fueled speculative investment throughout Europe, particularly in:
* Railroads
* Real estate
* Industrial enterprises
When the speculative bubble burst in Vienna in May 1873, confidence spread rapidly across European financial markets before reaching the United States.
Economic Consequences
The Panic produced a severe depression.
Major effects included:
* Hundreds of banks failed.
* Approximately 18,000 businesses went bankrupt over the following years.
* More than 80 railroads eventually entered receivership.
* Construction nearly stopped.
* Unemployment rose sharply.
Reliable unemployment figures did not exist, but historians estimate unemployment reached around 8–14%, with some cities experiencing substantially higher rates.
Effects on Workers
Workers suffered tremendously.
Many experienced:
* Wage cuts
* Layoffs
* Shorter workweeks
* Poor relief or charity
Large cities saw growing numbers of unemployed workers.
These hardships eventually contributed to labor unrest.
The best-known example came four years later:
* the Great Railroad Strike of 1877
The strike spread across much of the country and required federal troops to suppress.
Political Impact
The Panic severely damaged support for President Ulysses S. Grant's Republican administration.
Economic distress shifted national attention away from Reconstruction in the South.
Many politicians became more concerned with:
* economic recovery
* government spending
* monetary policy
than with protecting the rights of formerly enslaved people.
The Debate Over Money
One of the biggest political controversies involved the nation's money supply.
There were two major positions:
Hard Money
* Supported returning to a strict gold standard.
* Believed stable money encouraged investment and confidence.
Soft Money
* Wanted more paper money ("greenbacks") in circulation.
* Believed increasing the money supply would:
* reduce debt burdens
* stimulate business
* help farmers and workers
This debate remained central to American politics for decades and eventually culminated in the famous "Cross of Gold" speech by William Jennings Bryan in 1896.
Recovery
Recovery was gradual.
By about 1878-1879, industrial production began to recover, although many sectors remained weak.
Interestingly, despite the depression:
* American steel production increased.
* Manufacturing became more efficient.
* New technologies continued to spread.
This pattern puzzled economists because prices generally fell (deflation), yet overall industrial output eventually resumed growing.
Historical Significance
The Panic of 1873 is important because it:
* Marked the end of the post-Civil War economic boom.
* Revealed the dangers of speculative bubbles fueled by debt.
* Demonstrated how interconnected international financial markets had become.
* Accelerated consolidation in the railroad industry, as weaker companies failed and stronger systems absorbed them.
* Influenced U.S. monetary policy debates for more than two decades.
* Contributed to labor unrest and shaped political priorities during the remainder of the 19th century.
In many ways, the Panic of 1873 was the 19th-century equivalent of a modern financial crisis: excessive investment in a transformative technology (railroads), financed with borrowed money and buoyed by optimistic expectations, eventually gave way to collapsing confidence, frozen credit, widespread bankruptcies, and a prolonged economic downturn. Its parallels with later crises—such as the 1929 stock market crash, the dot-com bubble of 2000, and the housing and credit crisis of 2008—make it one of the most instructive financial panics in U.S. history.
Figure 3 - By 1873, the newly built railroads of the United States of America were the high-tech marvel of the day. They promised to deliver great value in the distant future. This fueled a speculation fever to build more and more railroads based on such promises of future wealth for all.
Figure 4 - But by 1873, the United States had also overbuilt its railroad infrastructure to an extreme to places where nobody yet even lived in a speculative anticipation of future rail traffic.
Figure 5 - This eventually led to the Panic of 1873. The first Great Depression experienced by the United States of America.
Figure 6 - The value of railroad stocks plummeted, as did most of the other stocks that were propped up by the Railroad Bubble.
An AI Bubble Burst would certainly produce a great deal of economic damage to the world, but it might be the only practical way to slow down the progress of Advanced AI enough to give us a chance. The main obstacle to this would be if the American government takes partial ownership of the American Frontier AI Labs and partial ownership of their debt as a matter of national security. OpenAI has already suggested the government take a 5% ownership of OpenAI.
Comments are welcome at scj33345@gmail.com.
To see all posts on softwarephysics in reverse order, go to:
https://softwarephysics.blogspot.com/.
Regards,
Steve Johnston
No comments:
Post a Comment