How AI contribution is evaluated

Overview

There have been recent claims that AI tools are now capable, or are becoming capable of doing research itself autonomously, with "minimal" human intervention. WorkSigned's mission is to help researchers prove that they are, and always will be the driving force behind innovation. While AI technology may continue to advance, WorkSigned strongly believes that humans are the ones that provide the direction and expertise to conduct meaningful research activity.

Using AI as a helpful tool

WorkSigned and many researchers recognize AI as a useful tool. That's why WorkSigned uses AI to detect the extent of research activity that is AI driven. When researchers use a web-based AI tool, they have no expedient way of proving "I directed the AI to do a minor task for me, one that simply saves me time and energy.", rather than "I delegated important strategic decisions to the AI.". Most platforms do not provide this kind of service. WorkSigned addresses this by applying a strict grading scale to gauge how much AI contributed to some task, like brainstorming or drafting.

The growing importance of attribution

Most AI providers have a strong claim to the inputs that are provided to their services. If the inputs were only used to train and improve their AI service, that would be a small problem. However, if the AI service starts to make the claim that their AI can produce innovation autonomously, a serious conflict of interest arises. How can a party who ingests intelligent input from others, claim credit for intelligent output without giving credit to those who provided the inputs?

The protection WorkSigned provides

WorkSigned's view is that modern ways to protect one's work are currently limited. Patent law is one of the strongest tools, but is restricted in what types of intellectual property it can protect. Research is especially exposed in that it has the dual requirement of needing open collaboration and public exposure, but also definitive, appropriate credit to those who produced the research. WorkSigned's approach is to prove the most - what was done, how it was done, when it was done, and who did it - with the least amount of reasonable doubt. This puts the researcher and institution in a very strong legal position.

WorkSigned's grading categories for research items with respect to AI generation

Note that human-assisted work is not a category here. Researchers already have human collaborators under existing authorship frameworks. These categories exist because AI is a new type of contributor that does not fit that framework.

Human generated. Researcher produced this artifact by hand, with no AI assistance in its creation (no AI drafting, coding, or idea generation). It is entirely their own work. The Researcher must also attest they did not use external AI services which are outside the scope of the WorkSigned application.

Human-driven AI generation. The Researcher is the driving force behind the entire idea and the artifact: Researcher supplies the direction, the requirements, and the decisions, and an AI tool executes the mechanical work (e.g. AI coding where the Researcher wrote the spec and the AI wrote the code, or AI drafting where the Researcher dictated the structure and the AI filled in the rough text.

AI-assisted generation. The Researcher gave the AI a creative task and let it propose solutions/options on its own; Researcher then selected the best one (or several) from what it produced. The AI generated the candidates; Researcher made the selection. The Researcher is still completely responsible for the ultimate result.

AI-only generation. The Researcher gave the AI a very general research goal (usually a nontrivial primary goal, e.g. 'produce an analysis of X') and it carried it out on its own with little or no human input or decision making along the way. The artifact is basically the AI's own work. The distinction with AI-assisted generation here is the Researcher only made the initial input.