Shaping Science with Nick Dirks
Nov 2, 2026 · 41 min · 12 segments
Ian Bogost, PhD, is a philosopher, author, game designer, and professor whose work explores the intersection of technology, media, and culture. Currently the Barbara and David Thomas Distinguished…
Nick DirksHost
On the one hand, that's what science is, is building models that isolate and identify certain kinds of issues, problems, questions.

at a different kind of level than some direct reference to a real world per se, even though obviously the assumption is that it will apply to the world.

But you have on the one hand this capacity to think about what came to be called AI as building these kinds of simulations that were also places where you could test certain kinds of relationships, whatever, on the one side.

On the other side, and that, again, continues to the present day, albeit I think some of you know, some of that history has been forgotten.

But what has also made that history, I think, not only forgotten in a way, but almost impossible to remember is the progress, as we've seen earlier this afternoon with LLMs.

What the, you know, what this, what these language models are doing by virtue of however much this is a part of fiction and however much this is not related to something that might be called true, it is fundamentally related to what we think of as human.

And the human propensity for language and the like is precisely what makes powerful language models begin to invite questions about whether superintelligence will develop at this new kind of level, which is to say to be, and this I think is a perversion of the way in which Alan Turing talked about the Turing test, but it's nevertheless a way of thinking about will machines become not just more intelligent, but will they take over from humans? Now, both of those end up collapsing this distinction between representation and reality in the end.

And I guess the question would be, is that going to matter when machines get to a certain point of power? That, in a way, representation does become reality.
Well, I suppose if you have an all-encompassing enough representational rationale that everyone believes and accepts it and there are no alternatives, then it's effectively reality.
So if we have a system that uses a complicated probabilistic model to produce stochastic textual output based on effectively the sum total of human knowledge produced under any circumstances, that doesn't make it less a take on the world, it's just a particular one that will color the way that we experience the world by means of it.
It has assumptions and it's decided on our behalf what kind of ways of producing, whether it's textual or whether it's synthetic analytic knowledge, or even whether it's visual information, that we take to be the important parts of reality, right? Those are sort of social, cultural, economic questions of power, they're questions of capital, and so on.

But let me ask a little bit about what is happening, whether it's to reality or just to language.

In the rush to see these models is everything from ways to write your term papers to sending out memos in organizations that are non-profits, for example.

In your paper, you talked about the way in which language models are great because they can produce great narratives, great stories, and those stories will help one decide what one thinks one wishes to test if one has the capacity and a second machine to test the world.

But the experience that I've had and that has been reported by many of even the best new LLMs, is that it does a lot of work.

On the one hand, that's what science is, is building models that isolate and identify certain kinds of issues, problems, questions.

at a different kind of level than some direct reference to a real world per se, even though obviously the assumption is that it will apply to the world.

But you have on the one hand this capacity to think about what came to be called AI as building these kinds of simulations that were also places where you could test certain kinds of relationships, whatever, on the one side.

On the other side, and that, again, continues to the present day, albeit I think some of you know, some of that history has been forgotten.

But what has also made that history, I think, not only forgotten in a way, but almost impossible to remember is the progress, as we've seen earlier this afternoon with LLMs.

What the, you know, what this, what these language models are doing by virtue of however much this is a part of fiction and however much this is not related to something that might be called true, it is fundamentally related to what we think of as human.

And the human propensity for language and the like is precisely what makes powerful language models begin to invite questions about whether superintelligence will develop at this new kind of level, which is to say to be, and this I think is a perversion of the way in which Alan Turing talked about the Turing test, but it's nevertheless a way of thinking about will machines become not just more intelligent, but will they take over from humans? Now, both of those end up collapsing this distinction between representation and reality in the end.

And I guess the question would be, is that going to matter when machines get to a certain point of power? That, in a way, representation does become reality.
Well, I suppose if you have an all-encompassing enough representational rationale that everyone believes and accepts it and there are no alternatives, then it's effectively reality.
So if we have a system that uses a complicated probabilistic model to produce stochastic textual output based on effectively the sum total of human knowledge produced under any circumstances, that doesn't make it less a take on the world, it's just a particular one that will color the way that we experience the world by means of it.
It has assumptions and it's decided on our behalf what kind of ways of producing, whether it's textual or whether it's synthetic analytic knowledge, or even whether it's visual information, that we take to be the important parts of reality, right? Those are sort of social, cultural, economic questions of power, they're questions of capital, and so on.

But let me ask a little bit about what is happening, whether it's to reality or just to language.

In the rush to see these models is everything from ways to write your term papers to sending out memos in organizations that are non-profits, for example.

In your paper, you talked about the way in which language models are great because they can produce great narratives, great stories, and those stories will help one decide what one thinks one wishes to test if one has the capacity and a second machine to test the world.

But the experience that I've had and that has been reported by many of even the best new LLMs, is that it does a lot of work.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.