Guidelines and Resources
Artificial intelligence (AI) encompasses a wide range of technologies, each with it's unique capabilities and applications. From voice recognition on smartphones to advanced data analysis and trend predications, AI is permeating various aspects of our lives.
AI applications, such as ChatGTP, have immense potential for driving innovation, enhancing creativity, and solving problems. However, like any other platform or application, they come with risk.
It is important that all staff and students at JCU are responsible and careful when using AI technologies to prevent misuse.
The quality and diversity of the training data used to teach AI models like ChatGTP has a significant impact on the performance and effectiveness of it's outputs. If the training data has biases or is based on incorrect or contradictory information, it will inadvertently learn and potentially perpetuate those anomalies possibly leading to misinformation or discrimination. For example, if a piece of misinformation is repeated often enough, the neural network might recognise it as a valid pattern, leading to inaccuracies in chat responses. Always make a conscious effort to identify, critically interrogate, and mitigate these biases. AI models also have a knowledge cutoff date which can lead to outdated or no information on recent topics.
The storage and handling of data by generative AI mode is a risk that varies depending on factors such as the deployment environment e.g. is it deployed locally or hosted on a server/cloud infrastructure; data privacy policies; user consent and settings; and purpose of data collection e.g., is the data being collected to improve it's performance or the user's experience vs. for research or advertising.
Models such as ChatGTP are considered 'relatively safe', there are several factors to consider to protect your data and ensure compliance with privacy regulations.
Data Input, Use and Storage:
Only input data that you would share with external organisations, companies, and competitors. Do not input personal or confidential information and be aware of potential copyright breaches and implications of data storage by AI platform owners. Although the AI application may provide options to limit the retention and use of your inputs e.g., opting out of your inputs being used for the training model; or commitments to delete your inputs within certain timeframes, control is lost immediately information is uploaded and is at risk of reuse for unethical or illegal purposes such as theft, fraud, discrimination and misinformation.
Even if your inputs are not being stored or used for training purposes, you should deidentify data before entering into a chat.
- Real-time Processing: Some AI systems process or analyse input data in real-time, which means that even if it's not stored permanently, it could still be temporarily processed or accessed by the system.
- Logging and Monitoring: Chat systems may log or monitor interactions for various purposes such as debugging, performance monitoring, or compliance. While these logs might not retain data input indefinitely, they could still contain sensitive information if not properly deidentified.
- Accidental Exposure: There's always a risk of accidental exposure or leakage of data, even if the intention is not to store it – e.g., if there's a software bug or misconfiguration, sensitive information could be inadvertently exposed.
- Regulatory Compliance: Depending on the jurisdiction there may be legal or regulatory requirements to deidentify sensitive information before processing or sharing it, regardless of whether it's stored permanently.
Authorised Users of AI at JCU are legally and ethically responsible for complying with these principles and for adhering to all other requirements of the:
By following the guidelines outlined below, you will be actively contributing to the security and integrity of your data and interactions in AI platforms, protecting both personal and university information from potential risks and vulnerabilities.
Awareness is Your Responsibility
- Stay proactive and stay informed – this is a rapidly evolving landscape.
- Find out and understand the limitations and risks of the AI model you are using.
- Choose AI models that are driven by ethical values and norms e.g., ChatGPT was developed by the company OpenAI whose mission is to ensure artificial general intelligence is safe and benefits all of humanity.
- Be cautious and strategic with data inputs and critically evaluate outputs for inaccuracies and bias.
- Cross-check information from credible sources where possible.
- Be aware that generated outputs may include material that breaches copyright.
- Recognise the potential risks posed by generative AI, including the production of harmful or malicious data or content that could be used for unethical or illegal purposes such as theft, fraud, discrimination, and misinformation.
- Staff and students undertaking research should also refer to advice provided by the Centre for Education and Enhancement (CEE).
Security Measures You Can Take
- Only input data that you would share with external organisations, companies, and competitors.
- Do not input personal or confidential information and be aware of potential copyright breaches.
- Use the settings option to opt-out of your data inputs being used to train the AI model you are using.
- Use the settings option to opt-in to have your data inputs deleted within certain timeframes – usually 30 days.
- Access ChatGPT via its official site or applications provided by OpenAI, verifying the authenticity to avoid phishing risks and ensure your interactions are protected.
Privacy Protection
- Use Privacy-by-Design principles to embed privacy into the development process throughout the lifecycle of your data.
- Deidentify or anonymise personally identifiable information (PII) before using it in chat or any other public-facing application, even with chat history disabled.
- This typically involves removing or obfuscating any direct identifiers such as names, addresses, phone numbers, social security numbers, etc., from the data. Additionally, you may need to apply techniques like masking, tokenisation, or generalisation to further obscure potentially identifying information. Contact the Library.
- Never submit third-party copyrighted materials or materials that you don’t own or manage the rights for into commercial and external generative AI platforms or services.
- Immediately report to Technology Solutions if sensitive information is mistakenly shared or contact the IT Help Desk.
Password Management
- Regularly change your password to deter unauthorised access.
- Create a strong, unique password incorporating a mix of letters, numbers, and special characters, avoiding predictable personal information.
- Always use a secure, encrypted internet connection, signified by "https://" and a padlock icon in the browser's address bar, especially when accessing ChatGPT from work computers.
- Utilise a reputable password manager to handle complex passwords securely, minimising the hassle of password management.
- Ensure your password recovery options, such as email or phone number, are current and secure to prevent unauthorised account resets.