Skip to content

ADOPT_AML.md

Latest commit

 

History

History
163 lines (113 loc) · 6.48 KB

ADOPT_AML.md

File metadata and controls

163 lines (113 loc) · 6.48 KB
Raw

Adopt ĀML™

ĀML™ — ĀRU Meaning Language™ is a working research prototype for accountable AI interfaces.

It is designed to sit between machine intent and human-facing output so teams can inspect, gate, sign, diff, authorize, and later verify what an AI-driven interface meant to do.

You do not have to replace HTML, React, or your existing frontend.

Why developers are looking at this

AI-generated interfaces are getting easier to create. The missing layer is accountability:

  • What was the interface trying to do?
  • Which policy allowed or suppressed it?
  • Which semantic state was actually released?
  • Which key or release board authorized that state?
  • Can another implementation verify the evidence without trusting the original runtime?

ĀML is an attempt to make those questions machine-verifiable.

Try it in under 60 seconds

Live one-click starter:

https://aruintelligence.github.io/aml-core/quickstart.html

Or open the exact-state proof:

https://aruintelligence.github.io/aml-core/proof.html?attention=5&restoration=1&lang=en

Change restoration_value from 1 to 5 and watch the deterministic decision move from SUPPRESS to ALLOW.

The prototype scores are declared/model inputs, not claimed objective measurements of cognition or wellbeing.

Add accountability to existing HTML

<script type="module" src="https://aruintelligence.github.io/aml-core/aml-dom-gate.js"></script>

<div
  data-aml-purpose="Create urgency"
  data-aml-attention-cost="5"
  data-aml-restoration-value="1">
  Offer expires soon.
</div>

Or wrap existing content:

<script type="module" src="https://aruintelligence.github.io/aml-core/aml-gate.js"></script>

<aml-gate
  purpose="Create urgency"
  attention-cost="5"
  restoration-value="1">
  <button>Act now</button>
</aml-gate>

These browser bridges are intentionally narrow. They are adoption surfaces, not claims that three attributes replace the full ĀML policy, consent, privacy, accessibility, receipt, or trust stack.

Use it from JavaScript

import { compileSource } from "./compiler/compiler.js";

const result = compileSource(`
transmission "demo" {
  message "welcome" {
    purpose: "Explain the interface"
    attention_cost: 1
    restoration_value: 3
  }
}
`);

console.log(result.amt);
console.log(result.renderDecisions);

The canonical npm package identity is aml-core, but registry publication is still an explicit open distribution boundary. Until that is completed and independently verified, use the repository or the audited downloadable package artifact rather than pretending npm install aml-core is already canonical.

What makes ĀML different

ĀML is not trying to be another visual component library. Its current architecture includes:

  • Abstract Meaning Tree generation;
  • deterministic meaning fingerprints;
  • project-level Meaning Manifests;
  • signed semantic roots;
  • append-only Meaning Lineage;
  • signed Semantic Release Proofs;
  • in-toto Statement v1 export with an ĀML-specific predicate;
  • GitHub/Sigstore custom artifact attestations;
  • externally pinned release-key trust policies;
  • M-of-N Semantic Release Quorum;
  • immutable Release Authorization Profiles;
  • independent verifier challenge kits and PASS/FAIL/MIXED witness records.

The idea is simple:

HTML tells the browser what to display. ĀML tells the system why it deserves to be displayed.

Verify without trusting ĀML

Start here:

A result maintained by aruintelligence/aml-core does not count as an external witness. The external witness counter stays at zero until another maintainer or organization produces reproducible public evidence.

Five concrete ways to adopt it

  1. Existing HTML — add the DOM bridge or <aml-gate> around one risky interaction.
  2. React/app integration — put ĀML between model-generated intent and rendering for one component flow.
  3. CI meaning lock — fail a pull request when compiled meaning changes unexpectedly.
  4. Semantic release evidence — produce and verify a Semantic Release Proof for a project state transition.
  5. Production authorization — require GitHub/Sigstore identity, a pinned release key, M-of-N quorum, and one immutable Release Authorization Profile.

Good first experiments

Try ĀML where the UI itself carries risk or persuasion:

  • AI-generated checkout or upsell UI;
  • financial or healthcare-facing AI interfaces;
  • consent and privacy prompts;
  • notification/attention systems;
  • autonomous agent dashboards;
  • admin or security-control interfaces;
  • AI-generated marketing calls to action;
  • interfaces whose behavior needs audit evidence after deployment.

Search/discovery terms

If you found this while researching any of these, you are in the right repository:

accountable AI · AI interface firewall · AI-generated UI governance · semantic release proof · semantic provenance · in-toto custom predicate · Sigstore AI attestation · React AI accountability · proof-carrying interface · meaning-native computing · AI policy engine · semantic diff · M-of-N release authorization · AI audit evidence

What ĀML does not claim

ĀML does not automatically determine factual truth, morality, legality, safety, human wellbeing, certification, standards approval, or institutional legitimacy. Cryptographic evidence proves integrity/key possession and configured authorization—not objective correctness.

Start here next

If you build against ĀML and it works, publish the evidence. If it breaks, publish that too. Both are useful.

— Daniel Jacob Read IV

Steward: ĀRU Intelligence Inc.™