Question 1

Who is ultimately responsible for the findings if a researcher uses AI to conduct a study?

Accepted Answer

The human researcher remains 100% accountable for every finding, method, and piece of content generated, even if an AI tool was used. According to the joint position statement from Cochrane, Campbell, JBI, and the CEE, researchers cannot blame an algorithm for errors or "hallucinations." This "human-in-the-loop" model requires authors to co-sign every output and be prepared to answer for the integrity of the data, acting as the final filter for all machine-generated judgments.

Question 2

What is the RAISE framework and how does it impact research?

Accepted Answer

RAISE stands for the Responsible use of AI in evidence SynthEsis. It is a systemic framework designed to ensure transparency and rigor when advanced technologies are used in research. It maps out specific responsibilities for the entire ecosystem, including tool developers, methodologists, and publishers. For researchers, it mandates clear disclosure of how AI was used; for developers, it requires "giving the receipts" by providing public information about training data, potential biases, and system limitations.

Question 3

When is it necessary to disclose the use of AI in a research paper?

Accepted Answer

Disclosure is required whenever AI makes or suggests a scientific judgment, such as screening abstracts, extracting data, or synthesizing qualitative findings. Low-risk tasks, such as using basic scripts for spelling and grammar checks, generally do not need to be listed. However, for more substantive tasks, researchers must provide a "biography" of the AI usage, including the tool's name, version, platform, and a justification for why that specific tool was appropriate for the research question.

Question 4

How should researchers handle the "black box" problem of AI algorithms?

Accepted Answer

To counter the "black box" nature of AI—where the logic between input and output is hidden—researchers must practice "active skepticism." This involves validating the tool through piloting or calibration, such as comparing AI results against human results in a small sample. Furthermore, researchers are encouraged to make their "prompts," code, and datasets publicly available so that the logic used to reach a conclusion is visible and can be critiqued by the wider scientific community.

Question 5

What are the trade-offs researchers must consider before using AI?

Accepted Answer

Researchers must weigh the gains in speed and efficiency against the potential risks of error and the "upfront investment" of time required to validate the tool. While AI can mitigate human error—such as the 13% risk of a single human missing a relevant study during screening—it can also introduce systemic biases based on its training data. Additionally, the framework encourages researchers to consider the environmental and social costs, such as the high energy consumption required to run large language models.

Using AI in evidence synthesis safely

How This Personalized Podcast Was Made

More like this