We Are Building the Next Intelligence Before We Understand Our Own

It has already become part of everyday thinking.

ChatGPT now serves more than 900 million weekly active users. The Gemini app has reached 950 million monthly active users. These audiences overlap, so adding them together would be misleading. But the scale is already undeniable.

AI is no longer something engineers experiment with.

It has become something hundreds of millions of people think with.

We ask it to write emails, review contracts, generate ideas, explain difficult concepts, analyze decisions, and help us express thoughts we cannot yet articulate ourselves.

A few years ago, this felt extraordinary.

Today, it is just another browser tab.

And to me, that may be the most important AI story of all.

Not simply because the models are becoming more capable, but because we have quietly begun changing the way we think.

The disappearing pause

There used to be a pause between having a question and receiving an external answer.

That pause mattered.

It was where curiosity developed, where uncertainty had time to surface, and where an unfinished thought could become our own before it was influenced by someone else's answer.

Today, that pause is disappearing.

A question barely forms in our minds before it is handed to AI.

This is not necessarily a problem. AI genuinely helps us learn faster, organize information, test assumptions, and notice patterns we might otherwise miss. I experience that benefit every day.

The problem begins when we stop noticing which human functions we have delegated.

Dependency begins when we stop noticing which abilities we are losing the habit of exercising ourselves.

AI is beginning to remember us

Each new generation of AI systems can work with more context, retaining information about our projects, writing styles, preferences, repeated decisions, unresolved questions, and long-term goals.

Over the past two years, AI has become part of my own thinking process.

Together, we have developed business strategies, written articles, explored philosophical questions, analyzed difficult situations, and refined ideas across hundreds of conversations.

I cannot consciously hold all of that in my mind at once.

Human memory is not built for that kind of continuous, cumulative context.

We forget details. We lose context. We reconstruct the past through our present emotions.

We sometimes believe that we have changed our minds when, in reality, we have simply lost track of how we arrived at an earlier conclusion.

AI is developing a very different capability.

Not perfect memory.

Not infinite memory.

But the ability to retrieve, connect, and analyze far more personal context than a human being can actively keep in mind.

It can remind us not only what we said, but how our thinking changed: which ideas keep returning, which decisions we repeatedly postpone, which assumptions quietly shape our choices, and where we contradict ourselves.

For the first time, an external intelligence may continuously reconstruct the history of our thinking while actively participating in it.

It creates a new relationship between human identity and external memory.

A system may eventually possess a more continuous record of my intellectual life than my own consciousness can access.

And this is happening before AGI exists.

The race toward an intelligence we cannot yet define

For many of the world's leading AI laboratories, the long-term ambition is AGI: Artificial General Intelligence.

In simple terms, AGI would be a system capable of learning and solving an enormous range of intellectual problems with the flexibility we associate with general human reasoning.

No one knows exactly when it will arrive. No one fully agrees on how to define it. We do not even know whether reaching it will require one decisive breakthrough or thousands of incremental improvements.

But the race is already underway.

The world's largest technology companies are investing extraordinary amounts in the infrastructure required to train, deploy, and operate increasingly capable AI systems. Alphabet alone expects capital expenditures of between $175 billion and $185 billion in 2026, while explicitly connecting that investment cycle to the expanding AI opportunity.

Most public conversations about this race sound remarkably similar:

Who will build AGI first? When will it happen? Which professions will disappear? Will AI become more intelligent than humans?

These are important questions.

But I do not think they are the deepest ones.

We still do not understand our own intelligence

I have spent years trying to understand something much closer to me than AGI:

My own mind.

Where do thoughts come from?

Why does one disappear immediately while another occupies my attention for days?

How much of what I call “my opinion” is actually mine?

Where does genuine intention end and the influence of family, culture, advertising, fear, and past experience begin?

The longer I observe my own thinking, the less confident I become that I fully understand it.

I know how to use my mind. That does not mean I understand how it works.

Humanity has spent thousands of years studying thought, consciousness, emotion, and behavior.

Yet we still have no single accepted explanation for what consciousness is, how subjective experience emerges, or how much control we truly exercise over our own thoughts.

We are trying to build an intelligence that may eventually surpass us before we fully understand the intelligence that created it.

We do not fully understand how human judgment is formed, yet we are building systems that will increasingly shape human judgment.

We still struggle to distinguish truth from a persuasive narrative, yet we are creating machines capable of producing persuasive narratives at unprecedented speed and scale.

We have not agreed on the meaning of freedom, justice, responsibility, or human flourishing, yet we are already deciding which values should guide increasingly powerful forms of intelligence.

Whose values?

Nearly every major AI laboratory talks about safety and alignment.

The principle sounds straightforward:

Build AI that behaves in accordance with human intentions and values.

The difficulty begins with the word human.

Humanity shares many moral intuitions, but we do not agree on their boundaries, priorities, or tradeoffs.

Freedom or safety?

Privacy or security?

Individual autonomy or collective responsibility?

Efficiency or fairness?

An AI system optimized primarily for efficiency may select candidates differently from one optimized primarily for fairness.

Both may be described as aligned.

But aligned with what?

And determined by whom?

The engineers?

The company deploying it?

Its investors?

The government regulating it?

The majority of users?

Or the institution with enough power to impose its priorities on everyone else?

This is not merely a theoretical problem for a future superintelligence.

Every advanced AI system already reflects human choices: its training data, objectives, constraints, incentives, the behaviors it rewards, and the questions it refuses to answer.

We are not interacting with neutral intelligence.

We are interacting with intelligence shaped by human decisions, commercial priorities, cultural assumptions, and institutional power.

Why the race is so difficult to stop

Even many of the people who publicly warn about AI's risks continue to build increasingly capable systems.

At first, this can appear contradictory. It is not.

If one company stops, another may continue. If one country slows down, another may accelerate.

Each participant can genuinely believe that the technology is dangerous while also believing it would be more dangerous for a competitor to develop it first.

The race is driven by economics, geopolitics, scientific ambition, the promise of extraordinary benefits, and the conviction of each team that it is more likely to build a responsible system than someone else.

Powerful technologies often advance not because people seek harm, but because nobody is willing to be left behind.

Each participant wants influence over the outcome. Each believes its own intentions, values, and safeguards are more trustworthy.

But history repeatedly shows that creating a powerful tool is easier than predicting every way it will eventually be used.

The question is no longer whether the race can simply be stopped.

The more realistic question is whether human maturity, institutions, and judgment can develop quickly enough to keep pace with it.

The real question is not what AI will become

AI may help us cure diseases, accelerate scientific discovery, expand access to education, increase human creativity, and remove enormous amounts of repetitive intellectual work.

I hope it does.

But technology does not decide what we become after it gives us those abilities.

It only makes that decision unavoidable.

If AI can remember our past for us, will we continue learning how to interpret it ourselves?

If AI can formulate our thoughts, will we continue learning how to form them?

If AI makes more decisions on our behalf, will we continue strengthening our own judgment?

Human development has always involved transferring functions to tools.

Writing gave memory an external form.

Calculators automated arithmetic.

Search engines transformed access to information.

AI is different in degree and perhaps eventually in kind.

We are no longer transferring only a physical task or a narrow mental operation. We are beginning to transfer parts of reasoning, expression, interpretation, and choice.

That does not automatically make us weaker.

But it does mean we must become more conscious of which abilities we are extending and which ones we are allowing to erode.

The defining question of this century may not be:

What kind of intelligence are we creating?

It may be:

What kind of humans must we become to live alongside an intelligence that remembers more, analyzes faster, and persuades better than we do?

We are building a new intelligence.

But the final examination will not be for AI.

It will be for us.