Reading across these 80 projects, I see something larger than a collection of AI experiments. Cosmos appears to be funding pieces of an emerging epistemic infrastructure: systems for tracing claims, exposing provenance, testing corroboration, detecting confabulation, preserving uncertainty, surfacing dissent, protecting human judgment, and improving collective deliberation.
Projects such as Claimgraphs, Sextant, OpenEvidence, Anti Lethe Machine, DissentCheck, Language Diffusion, Confabulation Detector, Context Engine and others approach different parts of the same underlying problem: how do humans and machines establish what they have reason to believe in an environment saturated with synthetic information?
What is especially interesting is the movement away from a simple “fact-checking” model. Truth increasingly appears here as a relational problem: claims connected to evidence, counterevidence, provenance, uncertainty, historical change, expertise, disagreement and institutional process.
In other words, Cosmos may be funding many of the instruments required for truth-seeking at scale.
The question I come away with is: where do all of these epistemic instruments eventually meet?
One way to answer that question is through a layered-Web architecture.
By a layered Web, I mean a shared digital layer above existing webpages in which contextual information and computation can appear in relation to whatever a person is currently looking at, without requiring the underlying website to provide it.
Cosmos is building many of the instruments needed to uncover truth, but a layered Web could provide the shared civic surface where those instruments become interoperable, contextual, governable, and collectively intelligent at Web scale.
Imagine encountering a disputed claim in an article. A Sextant-derived smart tag could show whether the claim is independently corroborated. A Claimgraphs overlay could expose the surrounding evidence and counterevidence. An Anti Lethe Machine tag could reveal how major AI systems have answered the question over time. A DissentCheck signal could warn that apparent consensus may mask convergence dynamics rather than genuine independent agreement.
None of these systems needs to become the universal arbiter of truth. Their signals can coexist around the same claim.
In the Metaweb architecture introduced in "The Metaweb: The Next Level of the Internet", a “smart tag” is essentially programmable context attached to a piece of Web content. The broader Meta-Layer Initiative is exploring how such tags, overlays, shared presence, trust signals and governance could operate as interoperable infrastructure above today’s Web.
What I don't yet see in the portfolio is a common place for the outputs of these systems to accumulate and become collectively useful.
Suppose ten Cosmos projects can evaluate the same claim. Where do their assessments meet? How are contradictory signals displayed? Who determines which are relevant in a medical context versus a political or scientific one? How can a community challenge a signal? How does reputation travel? How does new evidence update old conclusions? And how does the result appear to the next person who encounters that claim elsewhere on the Web?
At some point the truth-seeking problem becomes less an AI-model problem and more an Internet-architecture problem.
A layered Web provides one possible answer. Instead of requiring everyone to visit a dedicated truth platform, epistemic services can come to the person's existing point of attention. The article, video, paper or AI output becomes the common coordinate around which provenance, evidence, disagreement, people and machine reasoning can gather.
There is another reason the layer matters. Once many trust signals exist, the problem changes from signal creation to signal coordination. Which signals matter? Who governs them? Can communities adopt different evidentiary standards? Can minority interpretations remain visible? Can participants inspect why a particular warning appeared?
The Meta-Layer work deliberately avoids proposing one universal truth score. Instead, it treats the interface as a place where plural trust signals can be contextualized, compared and governed. Its emerging Desirable Properties are intended to constrain such an infrastructure around agency, provenance, interoperability, privacy, accountable governance, collective sensemaking and safe AI.
Could Cosmos' grantee ecosystem evolve from a collection of truth-seeking tools into an integral part of an interoperable epistemic layer for the Web?
Fellow truth-seeking Cosmonauts, I’d love to go deeper on that possibility with you here:
Congrats to all of the grantees!
Reading across these 80 projects, I see something larger than a collection of AI experiments. Cosmos appears to be funding pieces of an emerging epistemic infrastructure: systems for tracing claims, exposing provenance, testing corroboration, detecting confabulation, preserving uncertainty, surfacing dissent, protecting human judgment, and improving collective deliberation.
Projects such as Claimgraphs, Sextant, OpenEvidence, Anti Lethe Machine, DissentCheck, Language Diffusion, Confabulation Detector, Context Engine and others approach different parts of the same underlying problem: how do humans and machines establish what they have reason to believe in an environment saturated with synthetic information?
What is especially interesting is the movement away from a simple “fact-checking” model. Truth increasingly appears here as a relational problem: claims connected to evidence, counterevidence, provenance, uncertainty, historical change, expertise, disagreement and institutional process.
In other words, Cosmos may be funding many of the instruments required for truth-seeking at scale.
The question I come away with is: where do all of these epistemic instruments eventually meet?
One way to answer that question is through a layered-Web architecture.
By a layered Web, I mean a shared digital layer above existing webpages in which contextual information and computation can appear in relation to whatever a person is currently looking at, without requiring the underlying website to provide it.
Cosmos is building many of the instruments needed to uncover truth, but a layered Web could provide the shared civic surface where those instruments become interoperable, contextual, governable, and collectively intelligent at Web scale.
Imagine encountering a disputed claim in an article. A Sextant-derived smart tag could show whether the claim is independently corroborated. A Claimgraphs overlay could expose the surrounding evidence and counterevidence. An Anti Lethe Machine tag could reveal how major AI systems have answered the question over time. A DissentCheck signal could warn that apparent consensus may mask convergence dynamics rather than genuine independent agreement.
None of these systems needs to become the universal arbiter of truth. Their signals can coexist around the same claim.
In the Metaweb architecture introduced in "The Metaweb: The Next Level of the Internet", a “smart tag” is essentially programmable context attached to a piece of Web content. The broader Meta-Layer Initiative is exploring how such tags, overlays, shared presence, trust signals and governance could operate as interoperable infrastructure above today’s Web.
The Metaweb book: https://metawebbook.com
Meta-Layer Initiative: https://themetalayer.org
Desirable Properties of a Layered Web: https://desirableproperties.org
What I don't yet see in the portfolio is a common place for the outputs of these systems to accumulate and become collectively useful.
Suppose ten Cosmos projects can evaluate the same claim. Where do their assessments meet? How are contradictory signals displayed? Who determines which are relevant in a medical context versus a political or scientific one? How can a community challenge a signal? How does reputation travel? How does new evidence update old conclusions? And how does the result appear to the next person who encounters that claim elsewhere on the Web?
At some point the truth-seeking problem becomes less an AI-model problem and more an Internet-architecture problem.
A layered Web provides one possible answer. Instead of requiring everyone to visit a dedicated truth platform, epistemic services can come to the person's existing point of attention. The article, video, paper or AI output becomes the common coordinate around which provenance, evidence, disagreement, people and machine reasoning can gather.
There is another reason the layer matters. Once many trust signals exist, the problem changes from signal creation to signal coordination. Which signals matter? Who governs them? Can communities adopt different evidentiary standards? Can minority interpretations remain visible? Can participants inspect why a particular warning appeared?
The Meta-Layer work deliberately avoids proposing one universal truth score. Instead, it treats the interface as a place where plural trust signals can be contextualized, compared and governed. Its emerging Desirable Properties are intended to constrain such an infrastructure around agency, provenance, interoperability, privacy, accountable governance, collective sensemaking and safe AI.
Could Cosmos' grantee ecosystem evolve from a collection of truth-seeking tools into an integral part of an interoperable epistemic layer for the Web?
Fellow truth-seeking Cosmonauts, I’d love to go deeper on that possibility with you here:
https://gometa.substack.com/p/truth-is-too-important-to-leave-to