We Don’t Know AI’s Limits. But the Race Isn’t Waiting.

On superintelligence, the global competition to build advanced AI, and why safety needs to keep up.

Introduction

I disagree with several of Newport’s conclusions.

I think he’s underestimating how quickly AI is advancing, how much more capable these systems could become, and the economic and global forces driving their development.

Here are three points explaining why.

Point #1 — We don’t know how far AI can go.

Newport argues that OpenAI and Anthropic don’t currently know how to create superintelligence and that many technical breakthroughs are still needed.

Sure, but how can we be so confident about how far away those breakthroughs are?

Look at how much AI has improved already. We have systems that outperform humans in specific areas, and AI is already helping researchers develop better AI.

Even if we need completely new technology, who’s to say AI itself can’t help us discover it?

And yes, AI isn’t the same thing as LLMs. Of course it isn’t. That’s like saying your nervous system or skeleton isn’t your entire body.

The model is just one component. Combine it with tools, memory, research capabilities, and other technologies, and the entire system can do things the model couldn’t do alone.

We don’t have general superintelligence today, and none of this guarantees we’ll achieve it. But technology keeps evolving, and AI is increasingly helping drive that progress.

So I think calling superintelligence a fairy tale is premature. We don’t know how far this can go or how quickly we might figure out what’s missing.

Point #2 — The race toward advanced AI is about much more than ideology.

Newport argues that an ideological obsession with superintelligence is driving reckless behavior at companies like OpenAI and Anthropic.

I think he’s underestimating the economic and geopolitical forces involved.

Look at how much money is being invested, how fiercely companies are competing, and how important AI has become to national security.

Look at Nvidia and Jensen Huang. Nvidia’s board has opened an investigation into reports that he was targeted by a suspected Chinese intelligence operation. The allegations haven’t been conclusively established, but they illustrate the kinds of intelligence concerns surrounding this industry.

And have you noticed how many major companies and services have been experiencing outages lately?

Obviously, we can’t assume every outage is a cyberattack. But cybercrime and espionage are real problems. Chinese state-backed hackers have already been documented gaining access to American infrastructure networks.

Countries are trying to gain advantages by accessing each other’s technology, research, companies, and systems.

We know the US government conducts surveillance and intelligence operations. Why would we assume China isn’t trying to do the same to us?

This is an international competition involving governments, intelligence agencies, corporations, and entire economies.

It’s not just a few researchers obsessed with superintelligence.

Even if OpenAI and Anthropic slowed down, other companies and countries would still have powerful incentives to keep developing AI.

I’m not saying slowing down is impossible. I’m saying it’s incredibly difficult to coordinate globally when everyone has so much to gain.

Point #3 — AI safety needs to advance alongside AI itself.

This is where I actually agree with Newport.

We need regulation, better security, stronger safeguards, and more responsible experiments.

And yes, sometimes slowing down a dangerous experiment or deployment makes sense.

But I don’t think slowing down a few companies solves the larger problem.

The global competition isn’t going away just because some laboratories decide to change direction.

So why aren’t we pouring significantly more money into AI safety research, cybersecurity, testing, and securing these systems?

We should be investing heavily in making AI safer as it becomes more capable.

And given the competition between the US, China, and other countries, I think we should also be focused on maintaining our lead.

That doesn’t mean ignoring safety to win. It means making safety a major part of how we compete.

Newport himself says he isn’t advocating abandoning AI research or giving up American leadership to China. I agree with that.

I just think the problem is much bigger than the behavior of OpenAI and Anthropic.

We should regulate irresponsible development, slow down when necessary, and invest far more into safety and security. But we also have to recognize that AI development is being driven by global forces that no single company controls.

If increasingly powerful AI systems are going to keep being developed, we need to be prepared to make them as safe and secure as possible.

Sources & context (verify each before publishing)