The Latest from the Research Development and Strategy Team

There has been quite a buzz about this new Artificial Intelligence (AI) technology that can create uncannily human prose from a prompt – influencing everything from education to clinical care, to our research ecosystem. With great power, comes great responsibility. 

ChatGPT is a state-of-the-art natural language processing tool developed by OpenAI that is accessible via a web-based chatbot or application programming interface (API). Its unique ability to interpret natural language text and produce contextually relevant and coherent responses has made it a popular choice for various applications, including customer service, educational assistance, text summarization, and conversational interfaces. It can even be used to develop and correct open software code. With its power and availability, ChatGPT has the potential to revolutionize the way we communicate with machines and each other.  

However, it is essential to be aware of the security risks associated with using ChatGPT for confidential or personal information, and we must exercise caution when sharing sensitive data. While ChatGPT is a highly advanced and powerful language model,  its design also poses risks.  

First, the terms and conditions for ChatGPT and other AI services may allow inputted information to be used by the host companies for many purposes beyond a response to a user. Sharing sensitive information with ChatGPT or other such systems compromises that information, and can lead to additional dissemination to unauthorized parties. This is not a theoretical concern. It has already happened. ChatGPT’s own Frequently Asked Questions states: “No, we are not able to delete specific prompts from your history. Please don't share any sensitive information in your conversations.” Recent changes to the interface may allow some additional control over the reuse of information shared in the interface, but you must refrain from using ChatGPT or other chatbots on colleagues' or patients' confidential or personal information. Cutting and pasting personal or confidential information into ChatGPT or other web interfaces is a serious security and privacy breach. Even using AI-powered writing or coding assistants may expose confidential or private information since they invoke the same application programming interfaces (APIs), thus sending data to a company or third party.  

These algorithms, like any other technology, are not foolproof. They model patterns in human writing and can create incorrect but believable responses or reinforce biases introduced during their creation.  

We must constantly be vigilant and mindful of the information we share online or through the applications we use in our daily business. Sharing information with an AI requires extremely careful consideration and, while it may seem innocuous compared to sharing with a person, doing so may carry significant risks. We must follow best practices for safeguarding personal and sensitive information to protect those who entrust it to us.  

Resources: 

• ChatGPT Frequently Asked Questions 
• Article on security and privacy concerns with ChatGPT and other AIs 
• University of Colorado HIPAA Security Policy 
• Anschutz Medical Campus Acceptable Use of Information Technology Resources 

Contributors: Sean Davis, Casey Greene, Melissa Haendel, Jayashree Kalpathy-Cramer, Nikhil Madhuripan, Charlotte Russell 

On September 18, 2023, the Research Informatics Office (RIO) welcomed members from across the University of Colorado Anschutz community to hear from thought leaders and enthusiasts on the ever-evolving field of Artificial Intelligence (AI). The seminar featured 5 distinguished speakers that brought expertise and diverse perspectives to the audience. Presentations were followed by an engaging and thought-provoking panel discussion that invited the audience to weigh in on the ideas presented throughout the day. 

Session 1 -  AI: A (Very) Brief Introduction presented by Shawn O’Neil, PhD, MS
Shawn kicked off the seminar by sharing a brief presentation on the history of AI. Dating back to 1958, Shawn shared the major progressions of AI over the decades giving the audience a better understanding of how AI works and how far it has come. 

Session 2 - Implementing AI and Children’s Hospital Colorado presented by Sara Deakyne Davies
Sara described how CHCO is currently utilizing AI, the process of getting an AI project off the ground, and the long term maintenance and continuous-evaluation requirements of applications in clinical settings. Sara presented considerations for scoping AI projects, limitations, and future innovations. 

Session 3 -  AI Beyond ChatGPT with Jayashree Kalpathy-Cramer, PhD
With a presentation created with AI help, Jayashree reminds the audience that AI is ubiquitous in today’s society, and used daily by most of us, whether we know it or not. Jayashree explored some of the pitfalls of science within the AI research community, impacts to the clinical space, and future considerations.

Session 4 - Transparency, Bias, and Beyond with Arjun Krishnan, PhD
Arjun discussed sources of potential bias, misalignment, and inequity in AI research and applications. These include issues present in training data, how we evaluate AI-driven applications, and even how AI models are designed. Arjun also discussed potential misuses and security considerations, and the importance of diversity and inclusion in the field of AI research itself.

Session 5 - Security and Privacy Obligations with Charlotte Russell, John Heldens, and Lori Hopper
Charlotte, John, and Lori discussed a variety of security and ethical considerations for the use of AI at CU Anschutz. HIPAA and the role of Institutional Review Boards were discussed, as were security and intellectual property considerations when using AI models. 

We realized momentum towards curiosity, ideas, and solutions to some of the opportunities and challenges facing AI and LLM within our campus community and at large. The RIO anticipates additional opportunities for further collaboration and information sharing, including town halls and shareware presentations. Stay tuned for future AI focused events! 

You can view a recording of sessions 3-5 here. 
5:14 - Session 3
26:54 - Session 4
42:40 - Session 5

Full story