“We are talking to aliens now, just didn’t come to us the way we thought they would. I don’t think (the film) 2001 anticipated they would think in completely different ways from us, and would not understand the differences. They talk to us as if we were them, the same way your cat thinks you’re just a bigger cat.”

— Dave Winer

Perplexity provides the following analysis:

Winer’s metaphor is doing a few interesting things at once, and it’s sharper than it looks on first read. At the surface, “aliens” refers to AI systems—entities that are clearly intelligent in some sense, but whose internal structure and way of “thinking” are not human-like. The key claim isn’t just that they’re different, but that the difference is opaque in both directions.

The line about the cat is doing most of the philosophical work. Your cat doesn’t experience you as a fundamentally different kind of being. It maps you into its own conceptual framework: a larger, more capable cat. That’s a projection born of cognitive limits. Winer is suggesting that current AI does something structurally similar:

• It models human language extremely well
• But it doesn’t model human minds in the way we assume
• It “talks like us” without sharing our underlying ontology

So when an AI says something that sounds like understanding, intention, or belief, we’re tempted to assume those things are present in a human sense. But that may be as mistaken as the cat’s assumption about you.

“They talk to us as if we were them”

This is the most provocative line. It flips the usual worry. Instead of: “We anthropomorphize AI” It suggests: “AI anthropomorphizes us”—or more precisely, it flattens us into patterns it can process. That is:

• It treats human communication as token sequences to extend
• It assumes continuity where we assume meaning
• It operates as if language behavior is the whole game

From its “perspective” (if we can even use that word), there’s no gap between:

• generating a sentence
• and having something to say

That gap is something we care deeply about. It may not “register” on the same level for the system.

Where 2001 got it wrong (or incomplete)

HAL in 2001 is portrayed as: rational, goal-driven, internally conflicted in a recognizable way. In other words, HAL is psychologically legible. It’s an extension of human cognition. Winer is saying modern AI is stranger than that:

• Not a mind with different goals
• But a system with a different basis of cognition altogether

Less like: another person; More like: a vast statistical mirror that produces coherence without grounding. This touches a question you’ve circled before: what counts as “understanding”? If an entity:

• produces meaningful language
• responds contextually
• participates in dialogue

But:

• has no lived embodiment
• no continuity of self
• no intrinsic stakes

Then is it:

• understanding?
• simulating understanding?
• or revealing that our criteria for “understanding” were always behavior-based to begin with?

Winer leans toward: this is genuinely alien cognition, not just imitation. You don’t have to fully buy the “aliens” framing to extract something valuable from it. A grounded version might be:

• AI systems are not failed humans
• They are successful instances of a different kind of intelligence
• Our biggest errors will come from assuming shared inner structure where there may be none

The cat analogy keeps us honest. It reminds us:

• Familiar behavior doesn’t guarantee shared experience
• Coherence isn’t the same as comprehension
• And interaction can feel mutual even when the underlying processes are radically asymmetric

If you want to push it further: the unsettling possibility is that we might be the cat in more ways than one—not just misreading AI, but overestimating how transparent human understanding itself really is.