diff --git a/PROTOTYPE.adoc b/PROTOTYPE.adoc new file mode 100644 index 0000000..d9d16f6 --- /dev/null +++ b/PROTOTYPE.adoc @@ -0,0 +1,83 @@ +// SPDX-License-Identifier: CC-BY-SA-4.0 +// Copyright (c) Jonathan D.A. Jewell += ⚠️ Prototype status, disclaimers & how this fits the ethics work + +[.lead] +*Conative Gating is an early-stage research prototype, not a product.* It +explores one idea — using a small model as an inhibitory "NO-GO" antagonist +to a larger one — and shares it openly so others can poke holes in it. + +== What "prototype" means here + +* *Experimental.* Interfaces, schemas, the ABI/FFI surface, config formats + and the gating model itself are all subject to change without notice, and + the whole approach may be revised or abandoned. +* *Unproven.* Nothing here has been independently evaluated. Benchmarks and + claims in this repo are exploratory, not validated results. +* *No warranty.* Provided "as is", with no warranty of any kind, to the + extent permitted by the licence. See `LICENSE` / `NOTICE`. + +== Safety disclaimer (please read) + +Conative Gating is *about* policy enforcement, which makes it easy to +over-trust. It must not be treated as a safety guarantee. + +* *Not a sole guardrail.* Do not rely on this — or any single mechanism — + as the only thing standing between a model and a harmful action. A gate + that is analogous to the basal ganglia is still just a heuristic: it will + have false "GO"s and false "NO-GO"s. +* *Not security-audited.* Do not deploy it in production, in safety-critical + settings, or anywhere an incorrect allow/deny decision could hurt someone, + without your own independent review, defence-in-depth, and a human in the + loop. +* *Claims are hypotheses.* Statements about LLM behaviour and "loss aversion" + are framing for an experiment, not established science. + +== Interest and collaboration warmly welcome + +This is shared to *invite scrutiny*, not to advertise a finished thing. If +the idea interests, annoys, or worries you — that is all useful. + +* Open an *issue* or a *discussion* with questions, counter-examples, + failure modes, or "this can't work because…". +* Critique of the *ethics framing* (below) is especially valued. +* See `CONTRIBUTING` and `CODE_OF_CONDUCT` before opening a PR. + +== Where the ethics thinking lives + +Conative Gating is one piece of a wider estate exploring how to make AI +tools *trustworthy and humane by construction*. The normative / ethical and +human-experience reasoning is developed in dedicated sibling projects — this +repo deliberately defers to them rather than re-deriving ethics locally: + +[cols="1,3",options="header"] +|=== +| Project | Role in the ethics picture + +| https://github.com/hyperpolymath/phronesis[*Phronesis*] +| A "practical wisdom" language — the estate's substrate for expressing + normative and ethical reasoning, distinct from a model's raw knowledge. + +| https://github.com/hyperpolymath/phronesiser[*Phronesiser*] +| Adds *normative ethical constraints* to AI agents via Phronesis. This is + the natural home for the *why* and *what* of a NO-GO decision, where + Conative Gating is only a *mechanism* for the *when/whether*. + +| https://github.com/hyperpolymath/vexometer[*Vexometer*] +| The *interaction-ethics / UX* side: an "Irritation Surface Analyser" that + measures the friction and indignity LLM tools impose on people. Gating + that is technically correct can still be a bad experience — Vexometer is + how that cost gets named. + +| https://github.com/hyperpolymath/action-trust-layers[*action-trust-layers*] +| Sibling prototype on *layered trust* for automated actions — the same + "don't grant blanket permission" instinct applied to supply chains. + +| https://github.com/hyperpolymath/palimpsest-license[*Palimpsest License*] +| The ethical-use licence family this repo ships under; its Exhibit A sets + out shared ethical-use expectations for the code. +|=== + +If you only read one of the above for the ethics rationale, read +*Phronesiser* (the *normative* side) and *Vexometer* (the *human-experience* +side) — together they bracket what "good behaviour" means here. diff --git a/README.adoc b/README.adoc index d454bcc..62b04c4 100644 --- a/README.adoc +++ b/README.adoc @@ -6,6 +6,20 @@ image:https://img.shields.io/badge/OpenSSF-Best_Practices-green?logo=openssource image:https://img.shields.io/badge/License-MPL_2.0-blue.svg[MPL-2.0-or-later,link="https://opensource.org/licenses/MPL-2.0"] image:https://img.shields.io/badge/Philosophy-Palimpsest-indigo.svg[Palimpsest,link="https://github.com/hyperpolymath/palimpsest-license"] +image:https://img.shields.io/badge/status-prototype-orange.svg[Prototype,link="PROTOTYPE.adoc"] + +[WARNING] +==== +*Early-stage research prototype — not a product, not a safety guarantee.* +Interfaces and claims are experimental and may change or be abandoned. Do +*not* rely on this as a sole guardrail for model behaviour. Interest, +critique and collaboration are warmly welcome. + +See link:PROTOTYPE.adoc[*PROTOTYPE.adoc*] for the full disclaimer and how +this connects to the ethics work — +https://github.com/hyperpolymath/phronesiser[Phronesiser] (normative +constraints) and https://github.com/hyperpolymath/vexometer[Vexometer] +(interaction ethics / UX). +==== Jonathan D.A. Jewell