Generative AI in Test Development: Debates, Legalities, and Security Insights

Watch highlights from our latest L.A.B.S. webinar “Getting Unstuck: Ask-Me-Anything on AI in Test Development” featuring Marc J. Weinstein , Bridget Herd , Liberty Munson , and Pat Ward .

Welcome to the exciting and ever-evolving era of generative AI in test development! As you heard from Marc, Bridget, Liberty, and Pat, generative AI is a cutting-edge technology that enables machines to create content, such as exam items or rationales, that mimics human-like creativity. But hold on, what other things should be considered before testing this for yourself? Let’s dive in and explore more about what generative AI means for test development.

The Debate Around Generative AI

There’s currently a lack of clear structure and guidelines surrounding AI, and our understanding of the technology is also still evolving. This means that there are debates on how to navigate and standardize this emerging technology:

• Cheating has become a contentious topic as the accuracy of AI-generated content raises concerns about the authenticity and integrity of high-stakes assessments.
• Prompt creation raises issues like biases, manipulation, and unintended consequences.
• Questions around the ownership and confidentiality of exam content and sources of training data fuel legal, security, and accuracy concerns.

These debates are crucial to the ongoing exploration of generative AI's potential impacts.

Legal Considerations

When it comes to legal considerations in the realm of generative AI, it's like navigating a complex maze filled with essential checkpoints. Marc outlined these checkpoints for us:

1. Ensure the confidentiality and security of generative AI prompts and outputs used to create test content by employing enterprise-level solutions that (a) do not use your prompts or outputs for training data, (b) meet rigorous data security standards, and (c) offer a comprehensive agreement with the platform provider that includes meaningful confidentiality and data protection commitments.
2. Mitigate risks of copyright infringement by opting for platforms that only use licensed training data from known sources.
3. Carefully document the iterative process of human subject matter experts modifying and revising test content generated by AI, to enable the possibility that some form of copyright protection may be afforded to the content (if it is substantially revised).
4. Clearly define the obligations of subject matter experts through contractual agreements that explicitly specify what test development tools they are permitted to use and whether the use of generative AI to create content is allowed or prohibited, fostering clarity and accountability.

Taking these necessary precautions and seeking legal opinions will help you navigate the legal complexities.

Security Risks

While generative AI brings new opportunities, it also opens the door to innovative ways to cheat, making robust security measures vital. We can start with the following:

• Pat suggests we reimagine security measures as AI technology evolves by redesigning items that minimize vulnerability to AI-based cheating. ?We need to create comprehensive exam items.
• We need to safeguard item data and exam blueprints to prevent unauthorized access and maintain the integrity of assessments.

By embracing these changes, we can navigate the security landscape surrounding generative AI with resilience and ensure the integrity of our assessments.

Accuracy Concerns

While generative AI can create compelling and seemingly accurate content, it's crucial to remember that it's a narrator, not a reliable source of truth. Here’s what Liberty and Bridget suggest we keep in mind:

• Subject matter experts (SMEs) still play a vital role in upholding accuracy, particularly in high-stakes assessments.
• We have to maintain a healthy dose of caution and skepticism when evaluating AI-generated content.
• We need clear policies and guidelines to serve as our guiding beacons, ensuring content accuracy, consistency, and adherence to rigorous standards.

Join the Conversation

In this exciting journey through the world of generative AI in test development, we've explored the debates, legal considerations, security risks, and concerns about content accuracy. Want to continue the conversation? Send us a note at [email protected] or check out a recording of our recent webinar on this fascinating topic. Together, let's unlock the full potential of generative AI while ensuring the fairness and reliability of assessments.

About the Author

Maya Spence is a Marketing Specialist at ITS with over a decade of experience in sales and marketing. Her creative flair for storytelling and her love for data-driven campaigns has led her to constantly innovate and elevate marketing strategies. Outside of ITS, she enjoys baking cookies and staying active through long-distance running.