Generative Artificial Intelligence (AI) Guidance

Generative Artificial Intelligence (AI) Guidance

Definition and Purpose

Artificial Intelligence refers to the broad field of developing machines or software that can perform tasks that would typically require human intelligence, including learning, reasoning, problem-solving, perception, language understanding, and more.

Generative Artificial Intelligence (AI) refers to a subset of AI technologies that can create new content, data, or information that resembles human-generated output. These technologies utilize machine learning models and deep learning algorithms to analyze and learn from existing data, thereby generating novel outputs that can include, but are not limited to, text, images, audio, and video.

Scope of AI

This guidance document outlines the ethical, responsible, and secure use of Artificial Intelligence (AI) within the University. This guidance aims to support the University’s mission of education, research, and community service while protecting all stakeholders' integrity, privacy, and rights. It covers the use of AI tools, including but not limited to machine learning models, natural language processing systems, and AI-based search and administrative tools.

This page focuses on guidance for ÃÛÌÒ½´students, faculty, and staff. It is divided into the following sections:

  1. Key Considerations
  2. Educator Guidance
  3. Student Guidance
  4. Practical AI Applications in Research
  5. Intellectual Property Guidelines
  6. International Travel Considerations
  7. Procurement and Contracting

Key Considerations

  • Privacy Violations: While powerful, AI models can inadvertently reveal sensitive information when trained on personal data, leading to privacy breaches. Additionally, synthetic data generated by AI can sometimes be reverse-engineered to identify individuals, underscoring the need for caution and awareness.
  • Intellectual Property and Copyright Issues: Generative AI can produce similar content or directly copy existing works, potentially violating intellectual property rights. This can result in legal disputes and challenges in determining the ownership of AI-generated content.
  • Privacy Concerns: Generative AI, with its ability to create realistic synthetic data that includes personal information, raises significant privacy concerns. This synthetic data, if not handled with caution, can be misused for identity theft, surveillance, or other malicious activities.
  • Security Risk: Generative AI can create sophisticated phishing scams, deep fakes, and other forms of cyberattacks. These can be challenging to detect and defend against, posing significant security risks.
  • Bias: AI models can inherit biases in the data they are trained on, leading to unfair or discriminatory outcomes.
  • Accuracy: AI users should always validate the accuracy of created content with trusted first-party sources. Users are accountable for the content, code, images, and other media AI tools produce. They should be wary of potential “hallucinations” (e.g., citations to publications or materials that do not exist) or misinformation.
  • Code Development: Generative AI can assist software developers with writing code. However, caution should be taken when using AI for computer code because the resulting code may be inaccurate, lack security precautions, and potentially damage software systems. All code should be reviewed, ideally by multiple people.
  • Institutional Data: The acceptable use and institutional data governance policies govern ÃÛÌÒ½´institutional data. These policies disallow the uploading of institutional data into non-SMU-sanctioned AI products. In addition, AI users should exhibit great care in what data is uploaded into a product.
  • Business Process: AI offers many potential benefits and efficiencies for business process improvements and process automation. Tools, processes, and outputs should be reviewed for institutional data's reliability, accuracy, consistency, and privacy.

Caution:

AI Hallucination in Generative AI Usage

When employing generative AI, it is crucial to be vigilant about the phenomenon known as "AI Hallucination." This refers to instances where AI generates false or misleading information, which can occur even in well-trained models. To mitigate this risk:

  • Verification Protocol: Always cross-verify AI-generated data with trusted sources before use.
  • Awareness and Training: Educate users on the signs of AI hallucination, empowering them to identify and question implausible outputs.
  • Limit Use for Critical Tasks: Avoid relying solely on AI for decision-making in high-stakes scenarios where misinformation could lead to significant consequences.
  • Iterative Review: Implement a multi-stage review process that involves both AI outputs and human oversight to ensure accuracy and reliability.

This caution is integral to maintaining the integrity and reliability of academic work where AI tools are utilized.

Recommended Do and Don't

The following shares a list of dos and don'ts when considering generative AI tools and subscriptions for university business.

  • DO cooperate with a risk review for new AI tools and existing tools that have not been approved by OIT.
  • DO regularly review and comply with AI guidance and key data policies (i.e., information security and data governance).
  • DO ensure vendor contract terms align with safe and responsible AI governance.
  • DO report any potential data incident or breach immediately to OIT.
  • DON'T enter any sensitive, protected, regulated or confidential data into AI tool.
  • DON'T assume that public data is free of intellectual property.
  • DON'T purchase or subscribe to AI tools that have NOT been fully reviewed by OIT.

Generative AI & Related News