TLDR: EthicAlly is an AI-powered prototype designed to assist social science and humanities researchers with ethics assessment and REC submission preparation. Developed using “Constitutional AI” and collaborative prompt engineering, it aims to bridge gaps in ethics support, reduce REC workload, and improve ethical research design by identifying potential issues and offering recommendations, without replacing human oversight.
Navigating the complex world of research ethics can be a daunting task, especially for researchers in the social sciences and humanities. Unlike biomedical fields where strict ethics review is standard, social science methodologies often involve nuanced socio-political and cultural considerations that don’t always fit neatly into existing ethics frameworks. This disparity has led to a significant gap in tailored ethics training, support, and access to specialized Research Ethics Committees (RECs) in many parts of the world.
Enter EthicAlly, an innovative prototype developed by Dr. Steph Grohmann at the Ethics & Diversity Hub of the Ludwig Boltzmann Gesellschaft in Austria. EthicAlly is an AI-powered system designed to bridge these gaps, offering much-needed support to social science and humanities researchers in designing ethical studies and preparing for REC submissions. Its primary goal is to help researchers identify and address potential ethical issues early on, thereby protecting research participants and streamlining the ethics review process.
Addressing a Long-Standing Challenge
For decades, the introduction of comprehensive REC oversight across all scientific fields, including social sciences, has been met with skepticism. Critics argue that ethics frameworks designed for biomedical science often don’t align well with the diverse and flexible research designs common in social sciences, particularly qualitative methods. This can lead to ethics review feeling like a bureaucratic “tick-box exercise” rather than a genuine ethical deliberation. Despite these critiques, ethics planning and review are increasingly mandatory for researchers globally, driven by institutional policies, journal publishers, and funding bodies.
This situation creates a disadvantage for social scientists, especially those in regions with limited infrastructure for ethics training and support. At the same time, RECs themselves are often overwhelmed by the sheer volume of applications and may lack the specific expertise to evaluate complex social science research. EthicAlly proposes that Generative AI can offer a meaningful contribution to alleviating these pressures without attempting to automate or replace human ethical oversight.
How EthicAlly Works
EthicAlly operates as a web-based prototype, allowing researchers to submit their proposals and contextual information for analysis. At its core, it uses a powerful AI model, Claude Sonnet 4 by Anthropic. The system takes the research proposal, user input (like scientific discipline and country), and a structured set of instructions, then returns a detailed report within 30-60 seconds. This report includes a summary assessment, an ethical risk score (from 1 to 5), and specific recommendations on how to address identified issues.
A key aspect of EthicAlly’s design is its reliance on “Constitutional AI.” Unlike traditional AI models trained primarily through human feedback, Constitutional AI is fine-tuned to adhere to high-level ethical principles, almost like a built-in moral compass. Claude’s constitution, for example, incorporates principles from frameworks like the Universal Declaration of Human Rights and includes safeguards against biases. This approach helps the AI balance competing moral principles and consider multiple perspectives, making it particularly suited for ethics-related tasks.
AI as a Co-Creator
One of the most unique aspects of EthicAlly’s development was the collaborative approach to prompt engineering. Instead of humans solely designing the instructions for the AI, the AI system itself helped write the initial prompt. This innovative method leverages the AI’s own understanding of what information it needs to perform effective ethics analysis, leading to a more optimized and effective system. This co-creation process suggests that AI can act as a valuable partner in designing its own operational conditions, potentially leading to better outcomes in various scientific applications.
Also Read:
- Assessing Ethical Risks in AI: A Fuzzy Framework for Decision Support
- The Silent Threat: How AI is Polluting Online Behavioral Research
Initial Successes and Future Vision
Initial testing of EthicAlly involved analyzing 25 fictional research proposals, some of which contained subtly hidden ethical breaches, including those related to scientific racism or harmful biases. EthicAlly successfully identified nearly all target issues, even reacting strongly to proposals reflecting harmful biases (though the system was later refined to maintain a professional tone). The system demonstrated sophisticated ethical understanding and in-depth domain knowledge, adapting to complex scenarios like ethnographic research with undocumented migrants by suggesting flexible consent procedures.
While EthicAlly is a promising proof-of-concept, its developers acknowledge certain risks and limitations. These include the potential for users to over-rely on the AI’s advice, the “black box” nature of proprietary AI models, and privacy concerns. However, the system incorporates clear disclaimers and measures to mitigate these risks. Future plans involve more extensive testing, exploring the development of an Open Source version for greater transparency and sustainability, and integrating features like institutional policy integration and literature review functions.
Ultimately, EthicAlly represents a significant step forward in leveraging AI to support ethical research practices. By providing accessible and tailored ethics guidance, it aims to reduce the burden on RECs, empower researchers (especially those in under-resourced environments), and contribute to more socially just global research. For more detailed information, you can refer to the full research paper: EthicAlly: a Prototype for AI-Powered Research Ethics Support for the Social Sciences and Humanities.
