When AI Gets Cheaper, “Wild AI” Will Multiply Overnight—The Future of a Market Big Tech Will Not Touch

Using today’s AI sometimes makes me think:

It has become remarkably well-behaved.

AI offered by major companies such as OpenAI, Google, and Anthropic comes with all kinds of restrictions.

The moment a request enters territory that is difficult for a company to handle—dangerous activities, adult content, copyright, security, and so on—the AI suddenly becomes cautious.

That is only natural for companies providing services around the world.

But it also creates a distortion in the market.

What major companies prohibit does not necessarily match what people want to do.

And money flows into that gap.

The Same Thing Happened with the Internet

New technologies are not used only for respectable purposes.

The internet did not develop through corporate websites and e-commerce alone. There was also demand for things that were difficult to discuss in public, including adult content, piracy, and dubious message boards.

People do not suddenly become well-behaved just because they are given a new technology.

AI will probably be no different.

The more major AI services improve their safety measures, the more demand remains outside them from people who say:

“I do not need those restrictions.”

The question is who will serve that demand.

Open-Weight Models Change the Situation

This is where open-weight AI becomes interesting.

If a model’s weights are publicly available and it can run in your own environment, the situation is different from using a major company’s cloud AI.

There is no central service provider to stop every request by saying, “I cannot answer that question.”

I have taken to calling this kind of thing “Wild AI.”

For now, however, barriers to entry remain.

Running a high-performance model for many users requires a substantial amount of GPU capacity, electricity, and infrastructure. It is not yet cheap enough for individuals to start casually.

But that barrier will eventually fall.

The Day Anyone Can Become an “AI Provider”

What matters is how much money is required to enter the market.

Advances in quantization, distillation, and inference chips will make the necessary equipment cheaper.

We moved from an era when owning a web server was a major undertaking to one in which anyone could publish a website using an inexpensive hosting service.

The same thing will happen with AI.

Once someone can launch a capable service with equipment costing a few hundred thousand to a few million yen—roughly $2,000 to $20,000 at ¥150 to the dollar—small businesses and individuals will be able to enter the market.

Naturally, some people will have this idea:

“We will provide what the major companies prohibit.”

That is where the market for “Wild AI” begins.

The Frightening Possibility of Repeating the Winny Case

In Japan in the 2000s, the file-sharing software Winny became a social issue after it was used for copyright infringement and other activities.

The developer himself was charged with assisting violations of copyright law. He was convicted at his first trial, then ultimately acquitted by the Supreme Court.

In other words, a developer who was eventually found not guilty was arrested and went through a long criminal trial.

That weighs heavily on engineers.

You create a new technology.

Someone misuses it.

Then even the developer gets arrested.

Even if you may eventually be acquitted, that does not exactly make you think:

“All right, let’s confidently build the next technology.”

The same issue will arise with Wild AI if a major incident occurs.

Will the authorities pursue “the person who misused it”?

Or will they also pursue the developer, arguing:

“The person who created this kind of AI bears responsibility too”?

If they choose the latter, there will be no need to ban Wild AI itself.

Engineers will become too afraid to build it.

No one develops technology on the assumption that they can simply get arrested and later win an acquittal from the Supreme Court.

In a sense, that is more powerful than banning something by law.

All that is required is to make engineers think:

“If I build this, I might be arrested too.”

What began as an effort to stop AI crime could end with only AI development in Japan becoming paralyzed by fear.

Then, several years later, Japan will pay high prices to use AI made overseas and wonder:

“Why did no AI companies grow in Japan?”

Perhaps Japan’s AI strategy will be complete when, after frightening its engineers into inaction, it starts demanding that the country produce “the next GAFA.”