BINDERS WON’T STOP BULLETS: WHEN MILITARY AI CAN’T SAY ‘NO’

History suggests that every revolutionary technology goes through a period of excessive optimism. Railroads did. Automobiles did. The internet certainly did. Many of those technologies ultimately changed the world, but many of the companies that investors believed would dominate those revolutions did not.

AI will likely provide more examples.

The first issue is productivity. AI creates lasting economic value only if it enables people or businesses to produce substantially more than they could before.

Many companies are now spending enormous sums on AI software. In some cases, those costs are approaching — or even exceeding — the salaries of the employees using the systems. That investment only makes financial sense if those employees become significantly more productive or if the products they create generate enough additional revenue to justify the expense.

So far, the evidence is mixed.

Programmers often write software faster with AI assistance. Customer service representatives can answer more questions more quickly. Researchers can sift through mountains of information in minutes rather than days. But it remains surprisingly difficult for many companies to measure whether these improvements translate into higher profits. Faster work is valuable only if it ultimately creates more value than it costs.

The second issue receives far less attention: Not every AI task requires the newest and most powerful models.

Consider automobiles. Most drivers would love to have the latest EV with all the bells and whistles, but millions of perfectly satisfied motorists continue driving reliable, gasoline-powered cars that are several years old. They get the job done at a much lower initial cost.

The same principle applies to AI.

Today’s most advanced AI systems are astonishingly capable, but they are also expensive to operate. Many routine business tasks don’t require the newest technology. Older AI models — or even specialized systems developed for a single purpose — may perform nearly as well while costing far less.

That creates a challenge for companies selling cutting-edge AI. If customers increasingly discover that “good enough” produces nearly the same results for a fraction of the price, revenue growth may eventually slow.

Some AI providers are already directing simpler tasks to less expensive models while reserving their most advanced systems for problems that genuinely require them. That makes perfect business sense for customers, but it also raises questions about whether demand for the newest and most expensive AI systems will continue to grow at today’s extraordinary pace.

Competition adds yet another layer of complexity and uncertainty.

Only a short time ago, many observers assumed that a handful of American companies would dominate AI indefinitely. That assumption now appears less certain. Chinese developers have introduced increasingly capable models, while open-source AI systems continue to improve at remarkable speed. Meanwhile, companies such as Apple are working to move more AI processing directly onto phones and personal computers rather than relying exclusively on distant data centers.

Competition almost always benefits consumers: It lowers prices, encourages innovation, and expands choice. But it can also compress profit margins.

That distinction is often overlooked. AI may become enormously valuable to society while proving less profitable than investors currently expect. Those are not contradictory outcomes.

Finally, there is the question of the costs of development.

The companies leading the AI race are spending staggering amounts of money on specialized computer chips, data centers, engineering talent, and electrical infrastructure. Speaking at the EPA headquarters on July 23, President Donald Trump said his administration is “insisting that AI data centers and big tech companies pay their own way” as electricity demand surges to support the rapidly growing industry.

Investors have largely overlooked those costs because they remain focused on rapid revenue growth. Many AI companies continue to be valued primarily on expectations of future dominance rather than current profitability.

Those expectations may prove correct. Every major technological revolution requires enormous upfront investment before generating substantial returns.

But investors should remember that creating revolutionary technology and building a profitable business are not always the same thing.

The challenge: AI must continue to produce entirely new applications that create enough economic value to justify the industry’s massive investment. Some of those breakthroughs almost certainly will arrive. AI may accelerate drug discovery, improve medical diagnosis, transform manufacturing, strengthen national security, and enable scientific advances that seem unimaginable today.

The question is whether those breakthroughs will arrive quickly enough — and generate enough economic value — to support today’s extraordinary expectations.

None of this means AI is a passing fad. On the contrary, artificial intelligence is almost certain to reshape the global economy for decades to come.

The real uncertainty lies elsewhere. Will every company — or even most companies — currently racing to build bigger models and larger data centers ultimately earn enough to justify the billions being invested today? Or will history repeat itself, with a handful of lasting winners emerging from a period of excessive enthusiasm?

WIKIPEDIA IS TRAINING GOOGLE AND AI TO LIE TO YOU

Technological revolutions often create enormous wealth. Investment manias often destroy it.

The challenge for investors is remembering that those two things can happen at exactly the same time.

Andrew I. Fillat spent his career in technology venture capital and information technology companies. He is also the co-inventor of relational databases. Henry I. Miller, a physician and molecular biologist, is the Glenn Swogger Distinguished Scholar at the Science Literacy Project. They were undergraduates together at M.I.T.