talljeff68
Key distinction I believe: Inference alone can't get us there because there is a difference between Agents [properly] inferencing decisions and humans actually endorsing / authorizing the presumed decisions. So, what the post gets right is that humans have to do something explicit to articulate key decisions guiding the work and therefore we need some verifiable and traceable artifact back to the decisions, who made those decisions, and also ideally why.
Beyond that, this structure should really exist outside the silo of the agents doing the work in some kind of standardized form accessible to all parties involved in the work (people and agents)...which I think the post also got right.
I have been thinking about this for several years as the problems here are not specific to agents. It happens anytime people are coordinating to get things done...it's just that humans are better at, and used to, working around this via various ad hoc approaches.
In fact, the broader concept this related to - referred to as "Articulation Work" dates back to the 1980's in the area of Computer Supported Coordination Work (CSCW) research.
There is a seminal paper from 1992 by the researchers Kjeld Schmidt and Liam Bannon titled "Taking CSCW Seriously: Supporting Articulation Work." and I posted about it from different angle recently on our own company blog.
Ie: the Windows OEM license costs less than the total amount of money that the other software publishers are paying for distribution and if you reject windows, your are rejecting things that subsidize the price of the PC itself.
The key to understanding the watermark technology is to realize that the model is/was already using randomness to select among the top most probable tokens, often randomly picking between choices of [nearly] equal weight. The watermarking does NOT change the distribution of the random number generation nor does it affect the range of probabilities for which tokens are being considered. Instead, it only drives the sequence of the random numbers such that they form a cryptographically generated known ordering pattern that is determined by the secret key generating the pseudo-random sequence.
As a result of the approach, assuming inference is being done with all other parameterization of token selection being the same, there should be NO impact on the output quality....the amount of variation of output is within sample of the variation in output that already exists run-to-run of the same prompt. FWIW too: Google has confirmed this experimentally as well through full scale tests and evaluation of online Gemini output in search result pages.
ie: At one point "hyperlink" and "Uniform Resource Locator" were made up words. No different that say "web log" (blog) and "Client-Server Architecture" were being made up as well.
I contend that both "Artipoint" and "Cortex Layer" have similar parallels in the sense they refer to concepts newly formulated or at least are captured into a formalism with a label for the first time!
The work is still in draft form, but substantial: multiple detailed specifications are included, and we’re actively iterating across the grammar, security model, and synchronization layer. It’s not “finished,” but it’s well beyond half-baked.
We welcome contributions and feedback from collaboration-tool builders, protocol designers, distributed-systems engineers, and security/cryptographic reviewers who are interested in advancing an open, vendor-neutral standard for durable shared context.
The Agents Shared Cognition Protocol (ASCP)
I think of framing AI as having two fundamental problems:
- Practical problem: They operate in contextual and emotional "isolation" - no persistent understanding of your goals, values, or long-term intent
- Ethical problem: AI alignment is centralized around corporate values rather than individual users' authentic goals and ethics.
There is a direct parallel to social media's failure - platforms optimized for what they could do (engagement, monetization) rather than what they should do (serve user long term interests).
With these much more powerful AI systems emerging, we're at a crossroads of repeating this mistake...possibly at catastrophic scale even.