AI Regulations

Countries worldwide are starting to react the potential huge risks related to AI, by designing and implementing AI governance legislation. If you are a user or developer of AI tools, it is important to be aware of current developments in laws and regulations to ensure compliance. Resources like the IAPP Artificial Intelligence Topic Page have useful resources, like their Global AI legislation tracker of current AI legislative policy and related developments.  Of EU relevance, passage the EU AI Act is is expected by the end of 2023 or in early 2024. In brief, some of the proposals being debated include measures that:

• Prohibit certain types of AI systems, i.e., those that are manipulative, exploitative, perform social scoring or use “real-time” biometric ID systems.
• Classify high-risk AI systems and set out requirements with which these systems and their providers must comply.
• Delegate regulatory and enforcement authorities.
• Prescribe standards of conformity which may be applied by high-risk systems.
• Ensure transparency obligations for AI systems which interact with natural persons.
  
• Provide innovation measures.
  
• Set out a governance framework divided between the EU AI Board and national authorities.
  
• Create a database for high-risk systems.
  
• Requires an iterative process of review after-market deployment.
  
• Prompt the formation of codes of conduct for AI systems which are not high-risk.
  
• Stipulate confidentiality requirements for authorities dealing with proprietary data.
  
• Afford the power to adopt delegated acts to the European Commission.

Bottom line, if you are working on or planning to work on the development of artificial intelligence systems, specially if it involves the creation of datasets involving personal data used for their learning, be sure to contact the data team and the faculty privacy officer.
Guidelines for Generative AI

Data Protection

• Do not upload personal or sensitive data to any third-party service.
  • We also discourage this even with AI models that run locally (such as Llama from Meta) unless you are capable of auditing the code and ensuring that no data is leaving your device.
  • Grammarly, a popular software for assisting with improving the language used in documents, has recently released a cloud-based AI product, be sure that you are not sharing personal or sensitive information when using this product.
• We strongly discourage uploading your research documents and data to third-party services.
  • Research Documents can include but are not limited to: grant application, budget specifications, data management plan, privacy scan, contracts, project plan, scientific or data publications
• Ensure that your use of generative AI complies with local, national, and international laws and regulations, including copyright and intellectual property laws. If there are not any, ask relevant experts for advice.
• Be mindful about any privacy related concerns, when using AI generated images and media. Make sure that no personal/sensitive information is included in the generated content.
• Be alert for AI-enabled phishing, generative AI has made it easier for malicious actors to create sophisticated scams at a far greater scale. Continue to follow the UU security guidelines on phising and report suspicious activities or messages to CERT.

Responsible Use

• When using AI-generated content, be aware that you have to take accountability and responsibility for the content itself and its integrity.
• Scientific journals only accept works which are “original”. Complete works written by AI programs will not be accepted and labeled as plagiarism.
• An AI program cannot be considered or stated as a text author.
• Authors should acknowledge the use of AI, language models, machine learning, or similar technologies in their research. In the Acknowledgment or Methods section, provide a brief description of the content created or edited and specify the tool or model used, including its version and manufacturer (e.g., “Generated by OpenAI’s GPT-3.5”). Basic grammar, spelling, and reference-checking tools are excluded.
• We asked ChatGPT 3.5 to revise this point to be more concise. If you like to use AI generated content in publications, please include a clear description of the content that was created and the name of the model or tool, version and extension numbers, and manufacturer into accompanied documentation.
• Be cautious about potential infringement of someone else’s intellectual property rights such as copyrights and trademarks. Ensure that you have the appropriate permission to use such materials.
• Ensure that the generated content is used ethically and responsibly. Avoid generating any misleading and harmful content that might have potential consequences in others’ lives.
• Clearly define the context and intent of the generated media, to avoid any misunderstanding or misuse.
• Bear in mind that many generative AI models are trained on large static datasets, and may not have up-to-date information, such as ChatGPT 3.5 which only has information up to 2021.
• Develop or use AI models which have been trained and fine-tuned to be fair and open especially if the generated content is intended for a wide audience.
• Always implement quality check/control to ensure that the generated outputs are of sufficient quality for your intended use.
• Be aware of potential biases in your generated content, such as gender, religious, racial biases and take proper actions/measures to mitigate them.
• AI models like ChatGPT lack genuine emotional understanding. While they can generate empathetic responses, they don’t truly understand emotions. Therefore, it is strongly discouraged to use AI tools for decision making in which human emotions are involved.
• The rapid advancement of AI technology has raised legal and regulatory challenges. Laws and regulations regarding AI are still evolving, which can create uncertainty. Before using the output of AI tools, check for available legal basis or get advice from relevant experts.

Do you have more tips or guidelines, please submit them here: