Project summary

I am human but myy apologies i am 100% dependant on AI to write legibly in this place ...... I want to make it easier for a person to see what is still in dispute when AI systems give advice. A smooth summary can hide the one objection that should stop a decision. This project will produce a small public practice pack that puts that objection next to the recommendation.

What are this project's goals? How will you achieve them?

Over eight weeks from receipt of funding, I will develop 20 invented, low-risk cases, each with a short evidence packet and two matched briefs. One will use a conventional summary. The other will make the disputed point, evidence references, remaining uncertainty and next authorised human action explicit. Both versions will contain the same underlying facts. Cases will cover ordinary tasks such as checking a budget, interpreting an equipment log and reviewing a proposed data release. Some cases will have a valid objection, some a mistaken objection, and some insufficient evidence. I will write an answer key from the supplied facts, keep a correction log and hire an independent reviewer to challenge the cases. No outside collaborator is committed yet.

The work will be done remotely using documents and spreadsheets. I will publish the cases, both brief formats, answer keys, reviewer guide and limitations in an openly licensed repository. Weeks 1-2: specify the cases and scoring rules and make four examples. Weeks 3-5: complete the pack and commission review. Weeks 6-8: revise, check and publish it.

How will you know it worked?

Completion means 20 inspectable cases, 40 matched briefs, evidence references that can be checked, a scoring guide and a public review-and-correction record. The independent reviewer will check whether the answer keys follow from the evidence and whether the two versions preserve the same facts. Ambiguous cases will be revised or clearly marked. This first stage tests whether I can build a usable evaluation pack. It will not establish that one format improves human decisions or reduces catastrophic AI risk. A later comparison with participants would need a separate protocol, consent and appropriate ethics review.

How will this funding be used?

Full request: $6,000. Project lead: $3,200 for 160 hours at $20/hour; independent case review: $1,500 for 30 hours at $50/hour; AI tools and API use: up to $300; document accessibility and editing: $500; contingency: $500. Total: $6,000. Rates are planning estimates and reviewers have not been hired. If only the $3,000 minimum is raised, I will deliver 10 cases and 20 briefs: lead time $1,600, independent review $900, tools $150, accessibility/editing $250 and contingency $100. Both versions include public documentation. Spending and any material changes will be reported; unused funds will be handled under the grant agreement.

Who is on your team? What's your track record on similar projects?

I am Ray Peloquin, an independent systems designer with a background in construction and running crews. For about two years I have used several AI systems to develop and critically review design concepts. That experience led me to this question: can the human still see the objection and understand what they are being asked to approve? I have public work samples, but I have not yet run a controlled study showing this approach works. I will lead case development and practical checks. Paid independent review is budgeted because my own judgement and AI feedback are not enough. ChatGPT helped scope and draft this application; I am responsible for reviewing the proposal and the eventual work. AI systems are tools, not hired team members or independent validation.

What are the most likely causes and outcomes if this project fails?

The cases may be too easy, the objections may be unrealistic, or the briefs may accidentally differ in more than presentation. I may also have difficulty finding a suitable reviewer. The first four cases will be a checkpoint before building the rest. If review exposes major problems, I will narrow the scope, explain the changes and publish what was learned. Even a completed pack may prove unhelpful; I will report that rather than claim success from completion alone. This is a feasibility project, not a deployed safety system.

How much money have you raised in the last 12 months, and from where?

APPLICANT TO CONFIRM BEFORE PUBLICATION: total funding actually received during the past 12 months and its sources. Pending applications include BlueDot, Lightcone Commons, Foresight AI Nodes and EA Funds. Some proposed work overlaps with this pilot. Pending requests are not awards. Any overlap will be disclosed to funders and the same work will not be charged twice. The funding record reviewed while preparing this draft did not establish an award or payment; that is not a substitute for the applicant confirming the complete total.