AI Is Bayesian Evidence That We Live in a Simulation

jamesbaker11 pts0 comments

AI Is Bayesian Evidence That We Live in a Simulation

James’s Newsletter

SubscribeSign in

AI Is Bayesian Evidence That We Live in a Simulation<br>Artificial intelligence does not prove that we live in a simulation. But it should update our priors.

James Baker<br>Jul 15, 2026

Share

Let:<br>\(S=\text{we live in a simulation}\)

\(B=\text{we live in base reality}\)

Bayes’ theorem gives:<br>\(\frac{P(S\mid E)}{P(B\mid E)}

\frac{P(S)}{P(B)}<br>\cdot<br>\frac{P(E\mid S)}{P(E\mid B)}\)

AI matters because it shows that increasingly general intelligence can arise inside an artificial computational system.<br>A language model begins by minimizing prediction error:<br>\(\mathcal{L}(\theta)

-\sum_t \log P_\theta(x_t\mid x_{

Adjust billions of numbers, repeat the process, and the result can write software, produce mathematical proofs, and model parts of the world. In 2025, Gemini Deep Think achieved gold-medal-level performance at the International Mathematical Olympiad.<br>This does not prove that AI is conscious. The relationship between computation and consciousness remains deeply disputed; the computational theory of mind is a serious philosophical position, not an established fact.<br>But AI still establishes something important: behavior that looks like intelligence can exist inside a world we constructed.<br>Thanks for reading James’s Newsletter! Subscribe for free to receive new posts and support my work.

Subscribe

We are also beginning to build AI world models in which artificial agents learn, predict, and act inside generated environments. Some agents can even be trained within a model’s imagined world and then transfer what they learned back into the external environment.<br>The trajectory is increasingly clear:<br>\(\text{computation}<br>\rightarrow<br>\text{world}<br>\rightarrow<br>\text{intelligence within that world}\)

This strengthens Nick Bostrom’s simulation argument. If advanced civilizations eventually create large numbers of simulated worlds containing conscious observers, then simulated observers could vastly outnumber biological observers in base reality.<br>\(N_{\text{simulated}}\gg N_{\text{base}}\)

Under ordinary assumptions about observer selection, we should then expect to be among the simulated majority.<br>AI does not establish those assumptions. We do not know whether machines can be conscious, whether civilizations survive long enough to create ancestor simulations, or whether they would choose to run them.<br>But AI makes the premises less speculative. We now know that minds can create artificial agents, generated environments, and agents that model the environments they inhabit.<br>That evidence seems less surprising under a universe containing simulations than under one in which simulated minds are impossible:<br>\(P(E\mid S)>P(E\mid B)\)

Therefore:<br>\(P(S\mid E)>P(S)\)

A simulation would not necessarily contain visible pixels or glitches. Its inhabitants would discover internally consistent laws and call them physics. As David Chalmers argues, even a simulated world could still be a genuine reality to the beings living inside it. We are building minds inside machines and worlds inside computers. The provocative question is no longer whether a reality like ours could be simulated. It is why we are so confident that ours was not.

Share

Discussion about this post<br>CommentsRestacks

TopLatest

No posts

Ready for more?

Subscribe

© 2026 James Baker · Privacy ∙ Terms ∙ Collection notice<br>Start your SubstackGet the app<br>Substack is the home for great culture

This site requires JavaScript to run correctly. Please turn on JavaScript or unblock scripts

simulation text world simulated live inside

Related Articles