Invention Village

Where we stand

AI-Assisted R&D: What We Do and What We Don't

This page is both a product description and our compliance position. Three sentences first, then the basis and the limits.

Three sentences

  1. 1We don't invent for you

    AI proposes falsifiable hypotheses, searches real patents and papers, and drafts experiments that can actually test them. Which path is worth taking, how to run the experiment, and whether a result counts — those are your calls.

  2. 2We record how you invented

    What you asked and when, which real patents and papers the search returned at the time, which routes the AI offered, which one you chose, what the verification showed, and what remains unverified — kept in order.

  3. 3So when someone asks, you have something to show

    A timestamped R&D activity record that cannot be quietly edited after the fact. Export it, file it, hand it to your patent attorney.

Why this position holds

The question was never whether a piece of text was written by AI. It is whether a real act of invention happened behind it. On this, the Chinese and US offices say the same thing — and both classify AI explicitly as a tool.

China · CNIPA Guidelines for Patent Applications for AI-Related Inventions (Trial), issued 31 December 2024 — quotations below are our own translation; the Chinese text is authoritative

An AI-assisted invention is one obtained using artificial intelligence as an auxiliary tool during the process of invention. In this situation, the role played by artificial intelligence is similar to that of an information processor or a drawing tool.
For AI-assisted inventions, a natural person who has made a creative contribution to the substantive features of the invention may be named as the inventor. For AI-generated inventions, inventor status cannot be conferred on artificial intelligence under current Chinese law.

United States · USPTO Revised Inventorship Guidance for AI-Assisted Inventions (90 FR 54636, 28 November 2025; the February 2024 version was rescinded in its entirety)

The same legal standard for determining inventorship applies to all inventions, regardless of whether AI systems were used in the inventive process.
AI systems, including generative AI and other computational models, are instruments used by human inventors. They are analogous to laboratory equipment, computer software, research databases, or any other tool that assists in the inventive process.

Note where the line falls

The Chinese guidelines split this into two categories. An AI-assisted invention — where a natural person made a creative contribution to the substantive features — can be named and filed. An AI-generated invention, made without substantive human contribution, gets no inventor. The line is not whether AI was used. It is whether a person actually did the work.

Which is why the product works this way

If the line is whether a person actually did the work, then the job of the product is to leave a trace of that work.

  1. 1

    Start with something that can be proven wrong

    The research explorer outputs falsifiable hypotheses, parameters that still need measuring, and verification methods you can follow — not a polished technical description that merely looks complete.

  2. 2

    You pick what to verify

    Only the items you deliberately promote become anchors in your project. Promoting one snapshots what the AI said at that moment; no later run ever rewrites it.

  3. 3

    Come back and record what happened

    Supported, refuted, or inconclusive — record it as it was. The log is append-only: to correct something you add another entry, and the old one stays, marked as superseded. What you retracted is part of the record too.

  4. 4

    Refuted routes feed the next round

    A direction you tested and ruled out won't be suggested again. It enters the next round as a constraint only — never as data anything may cite.

What this proves, and what it does not

This section matters more than the ones above it. We will not make any claim that reads as "use this product and you're fine".

What it does show: a person added entries over time to a timestamped, append-only log, attesting to their truthfulness each time. That is meaningfully better than nothing.

Three things it does not show

  • ·It does not prove the experiment was run. We receive the text and numbers you type; software cannot verify what happened in the lab.
  • ·It does not prove you made a creative contribution to the substantive features. Inventorship is a judgement for you and your attorney, not a conclusion software can reach for you.
  • ·It does not rule out backfilling everything in one sitting afterwards. Steady accumulation looks different from a late catch-up, but noticing that is not the same as disproving it.

So we say "R&D activity record" and "traceable". We do not say "proof of truthfulness". And we promise no application will sail through examination — whether it does depends on whether you actually did the research.

What we will never build

  • ✕No one-click generation of many similar applications, and no rewriting one filing into several variants.
  • ✕No integration that files on your behalf with a patent office. Filing is the applicant's and the attorney's act.
  • ✕No promise of grant, no promise of passing examination, and nothing that claims to help anything evade detection.
  • ✕No AI-invented experimental data, test report numbers, or material properties. Where a number has to come from you, we leave it blank and mark it.

Start from a real technical problem

Describe the problem you are actually working on and see whether it yields a few routes worth testing.

Run a research exploration