THE UNIVERSITY OF THE WEST INDIES(UWI) ARTIFICIAL INTELLGENCE(AI) POLICY GUIDANCE FOR RESEARCHERS WIKI

THE UNIVERSITY OF THE WEST INDIES(UWI) ARTIFICIAL INTELLGENCE(AI) POLICY GUIDANCE FOR RESEARCHERS.

GUIDANCE FOR RESEARCHERS.

There are at least four (4) categories of UWI researchers that can be defined by the way that they are impacted by Generative AI tools:

1. Researchers who develop or train generative AI tools or components of them.
2. Researchers who analyse generative AI tools and may create derivative artefacts from them.
3. Researchers who use Generative AI tools to help them conduct research effectively, but the topic of that research may be completely unrelated to AI.
4. Researchers who do not use Generative AI tools at all.

Below are guidelines for UWI researchers who fall in one or more of the first three (3).

ENHANCED RESEARCH TOOLS:

Generative AI tools offer a variety of applications to aid in research tasks such as preparing data for analysis, refining assessment tools, enhancing literature searches, summarising articles, and assessing their relevance to the existing body of literature. These capabilities are beneficial to researchers across the previously outlined categories. It is crucial, however, for researchers to verify the reliability and accuracy of the AI tools they employ. Despite the significant support these tools provide, researchers must remember that they bear ultimate responsibility for the information they publish, irrespective of the level of assistance received from AI technologies.

ETHICAL RESEARCH WITH AI:

The following Table 1 provides recommendations for each category of researcher who chooses to engage with Generative AI tools (the first 3 categories) how to do so ethically along certain dimensions:

Table 1

	Researchers who develop or train generative AI tools or components of them should:	Researchers who analyse generative AI tools and may create derivative artefacts from them should:	Researchers who use Generative AI tools to help them conduct research effectively, but the topic of that research may be completely unrelated to AI should:
Data Privacy	Ensure that any private data they have used in training or development cannot be reconstructed from only the model by its users	Be equally cautious with any private data that they may handle as part of their research. Furthermore, if their artefact may collect information from their users, the privacy of which cannot be guaranteed, this should be made very clear to those users	Ensure that any private data that may need to be fed to such tools are guaranteed by their creators to protect such privacy, otherwise the data should be anonymized before being submitted to those tools.
' Data Ownership	Ensure that their use of data for training, testing or evaluation of the AI model is both legal and ethical. Specifically, this means that such data either: is in the public domain, has been given explicit permission for such use by its creators/owners, or is used in accordance with Fair Use under Copyright Law	Give guidance to its users on the ownership of the data that their tools output to the user (e.g. many art-generating AI tools give copyright of the created works to the user)	Ensure that the data outputs of those tools are not encumbered in ways that may interfere with their research objectives
Bias	Strive to reduce bias in the data used in training, testing and evaluation of their tools	Be mindful of any known biases in those models and take them into account when presenting the capabilities and limitations of their artefacts	Be mindful of how any existing biases in those tools might interact with their research findings, and take steps to mitigate those effects. For example, when prompting the AI to recommend a selection among alternatives, to also provide the basis on which that choice should be made.
Transparency	Declare how the data used for training, testing and evaluation was obtained and curated, or at least be prepared to do so if requested.	Be able to disclose the role that those AI tools play within their artefact and to characterise the extent to which limitations of those AI tools impact their artefact	Document the role that such tools have played in their research process and make appropriate declarations according to the guidelines of their publishing outlets. Keep in mind that the work they publish adequately attributes ideas that they did not originate to their sources.