AIRBDS: Assessing AI readiness in bioscience datasets

The capability of artificial intelligence (AI) technology is advancing rapidly, bringing both tremendous opportunities and risks. In the biosciences, AI promises to accelerate understanding of fundamental biology and facilitate advances in medicine, agriculture, sustainability, and biotechnology. However, the use of inappropriate, poorly documented or intentionally manipulated datasets can lead to inaccurate and unreliable results, undermining the integrity of the scientific record. Data is the crucial foundation of AI, directly influencing the capabilities and biases of any research or model, and modern bioscience research generates expansive data in a diversity of repositories, formats, and standards. To expedite opportunities and mitigate risks, it is necessary to define and measure the suitability and readiness of data resources for use with AI. To this end we have developed AIRBDS Core, a method of assessing the AI-readiness of datasets. It is designed to be objective, accompanied by clear and explicit annotations; to be applicable to any dataset and application; and to produce a quantitative result. AIRBDS Core also functions as a checklist, generating implicit suggestions for improvement. We have also developed prototype agentic skills to help support human assessments using AIRBDS Core. We present the approach and the metric here, with the aim of initiating community feedback to ensure broad applicability and ultimately encourage uptake.