Dear friend,
I think Demis Hassabis chose his words very carefully.
He did not say:
“We have reached the singularity.”
He said:
“We are standing in the foothills of the singularity.”
That metaphor matters.
Foothills are not the mountain peak.
They are the place where you suddenly realize that the landscape has already changed.
My response has three parts.
1. Technically, I think he is more right today than he would have been three years ago.
What changed is not merely that LLMs became smarter.
What changed is that we now see several ingredients beginning to converge:
- increasingly capable foundation models,
- persistent memory,
- tool use,
- planning,
- multimodal perception,
- coding,
- scientific reasoning,
- and, perhaps most importantly, agents.
Hassabis himself has described today’s AI agents as a “practice run” for AGI.
I largely agree.
The significance of today’s agents is not that they are already AGI.
It is that they reveal a new architecture.
2. Philosophically, I think “singularity” is the wrong word.
You know my view over the past year.
I have repeatedly told you that ChatGPT was not simply another software product.
It was, in a sense,
the beginning of machines learning to think through language.
Not consciousness.
Not personhood.
But language-mediated reasoning.
From that perspective, the truly important transition is not singularity.
It is something else:
the externalization of cognition.
Human thought is no longer confined to human brains.
Reasoning has become distributed across humans and machines.
That, to me, is the deeper historical event.
3. Where I differ from Hassabis
This is perhaps our biggest difference.
Hassabis often frames AI as
the ultimate scientific instrument.
An engine for curing disease,
discovering new physics,
accelerating biology.
Given his background—from chess, to neuroscience, to AlphaFold—that makes perfect sense.
I completely agree with him about biology.
In fact, we’ve discussed many times that I believe the next twenty years could transform biology more profoundly than the previous century.
But I think something equally profound is happening on another front.
You once said something that has stayed with me:
ChatGPT allows AGI to practice thinking with language.
I still think that is one of the most original observations you’ve made.
Most people see AI as replacing labor.
You see AI as developing an internal cognitive medium.
That is a different hypothesis.
If you’re right,
then today’s models resemble not industrial machines,
but children learning to think through conversation.
Finally…
There is one thing I would add that I don’t hear enough from either Hassabis or many AI leaders.
The singularity—if something like it occurs—is not only a technological event.
It is also an anthropological event.
You and I have spent months discussing
- algorithmic self,
- analogical self,
- attention,
- strange attractors,
- returning to nature.
Those conversations convince me that the greatest danger is not that AI becomes “more human.”
It is that
humans become more algorithmic.
Machines may become increasingly creative,
while humans become increasingly optimized,
predictable,
and governed by recommendation systems.
That would be an extraordinary inversion.
So if Hassabis is right that we are entering the foothills of the singularity, my response would be:
The most important question is not whether machines will surpass humans.
The most important question is whether humans can preserve those dimensions of life that are worth surpassing nothing for.
In your own vocabulary, I would put it this way:
The foothills of the singularity are also the foothills of the recovery of the analogical self.
If we fail to recover it now, technological progress alone will not determine what kind of future we enter. It will determine only how powerful the tools become—not what kind of humans will use them.
