Presentation
WTF are AI Chatbots, and Should We Care? An Exploration of ChatGPT in Human Factors
SessionAI/Risk (MDD6)
Description*Background*
Artificial intelligence (AI) chatbots are taking the world by storm. From making organized grocery lists to planning vacations to giving (unofficial) medical diagnoses, AI chatbots are seemingly limitless tools that anybody with a computer and internet can use. ChatGPT is one of the most popular AI chatbots, becoming a workplace and schoolplace name similar to the way that “Kleenex” and “Band-Aid” have.
ChatGPT (Chat Generative Pre-trained Transformer) is an AI language model launched by OpenAI in 2020 that employs deep learning techniques to understand and generate human-like text responses. ChatGPT can engage in natural language conversations, answer questions, provide explanations, and help with a wide range of tasks. This natural language capability makes the chatbot particularly powerful, as it can help eliminate the middleman between human and machine and instead make the machine appear human. So much so, that it even helped us write this paragraph about ChatGPT for this submission! Don’t worry - the rest of the submission is human-powered unless otherwise stated upfront.
It was only a matter of time until us HF professionals began asking whether we should be using ChatGPT in our field. But exactly how can this natural language AI chatbot help us in our work? And, on the flip side, is it really all it’s cracked up to be, considering the risks and shortcomings?
*Topic and Application*
Some applications of AI in HF might seem obvious and ripe for replacing jobs (e.g., AI’s ability to create mock-ups, iterate on designs, and produce high fidelity renderings in seconds).
From our perspective, instead of being an AI tool that replaces us, ChatGPT could be an AI tool that helps us. Our interest in the intersection between ChatGPT and HF is how the tool can help us from the research, documentation, and process side of HF—particularly in a field as highly regulated as medical devices and pharmaceuticals—but also understanding the risks.
Should our submission be accepted, we will introduce ChatGPT to unfamiliar attendees, and then investigate ChatGPT’s capabilities to understand its full capacity for medical device HF. Through our presentation, we hope to enlighten attendees on how ChatGPT has the potential to supplement our work—rather than replace it (as may be the case in other industries)—and draw attention to potential risks of using ChatGPT in the workplace. Our presentation will aim to answer the following questions:
• What types of tasks could ChatGPT perform to lighten our workload? When we asked ChatGPT, it said “it can crunch data faster than you can say ‘Human Factors & Ergonomics Society Health Care Symposium’ and synthesize it into actionable insights.” But, we’re still wondering: can we use ChatGPT to perform relatively simple tasks like creating a task analysis, rewriting instructional content to be at a lower reading level, and generating open-ended, non-biasing interview questions? And should we?
• What is the quality and accuracy of ChatGPT’s work, and how can the quality of our human inputs impact the quality of ChatGPT’s outputs (e.g., garbage-in, garbage-out)?
• What is the extent of ChatGPT’s capabilities? For example, can it correctly identify critical tasks based on the applicable FDA definition of critical tasks (i.e., CDER or CDRH), or develop user needs/product requirements that are measurable and verifiable?
• What are the potential issues, disadvantages, and risks of using ChatGPT? For example, what are the repercussions of making decisions based on ChatGPT’s outputs, or enabling complacent “HFers” to lead HF activities with false confidence provided by ChatGPT? Or, what about confidentiality concerns?
To answer these questions in an engaging way, we expect to include live ChatGPT demos as frequently as possible. Our hope is to include audience input (e.g., topics of interest, specific questions) to generate real-time ChatGPT outputs, but we recognize that factors outside of our control might make this impractical (e.g., ChatGPT server is down or overcrowded). In case we cannot access ChatGPT live during our presentation, we will pre-record conversations with ChatGPT that we can show to the audience as examples instead.
Additionally, we are learning from other HF professionals how they currently use ChatGPT in their work. We distributed a survey on LinkedIn over the summer and continue collecting responses that give us high-level insights from respondents, and several respondents have indicated an interest in sharing their experience with us in more detail during an interview. Such learnings would be captured in our presentation as well.
Artificial intelligence (AI) chatbots are taking the world by storm. From making organized grocery lists to planning vacations to giving (unofficial) medical diagnoses, AI chatbots are seemingly limitless tools that anybody with a computer and internet can use. ChatGPT is one of the most popular AI chatbots, becoming a workplace and schoolplace name similar to the way that “Kleenex” and “Band-Aid” have.
ChatGPT (Chat Generative Pre-trained Transformer) is an AI language model launched by OpenAI in 2020 that employs deep learning techniques to understand and generate human-like text responses. ChatGPT can engage in natural language conversations, answer questions, provide explanations, and help with a wide range of tasks. This natural language capability makes the chatbot particularly powerful, as it can help eliminate the middleman between human and machine and instead make the machine appear human. So much so, that it even helped us write this paragraph about ChatGPT for this submission! Don’t worry - the rest of the submission is human-powered unless otherwise stated upfront.
It was only a matter of time until us HF professionals began asking whether we should be using ChatGPT in our field. But exactly how can this natural language AI chatbot help us in our work? And, on the flip side, is it really all it’s cracked up to be, considering the risks and shortcomings?
*Topic and Application*
Some applications of AI in HF might seem obvious and ripe for replacing jobs (e.g., AI’s ability to create mock-ups, iterate on designs, and produce high fidelity renderings in seconds).
From our perspective, instead of being an AI tool that replaces us, ChatGPT could be an AI tool that helps us. Our interest in the intersection between ChatGPT and HF is how the tool can help us from the research, documentation, and process side of HF—particularly in a field as highly regulated as medical devices and pharmaceuticals—but also understanding the risks.
Should our submission be accepted, we will introduce ChatGPT to unfamiliar attendees, and then investigate ChatGPT’s capabilities to understand its full capacity for medical device HF. Through our presentation, we hope to enlighten attendees on how ChatGPT has the potential to supplement our work—rather than replace it (as may be the case in other industries)—and draw attention to potential risks of using ChatGPT in the workplace. Our presentation will aim to answer the following questions:
• What types of tasks could ChatGPT perform to lighten our workload? When we asked ChatGPT, it said “it can crunch data faster than you can say ‘Human Factors & Ergonomics Society Health Care Symposium’ and synthesize it into actionable insights.” But, we’re still wondering: can we use ChatGPT to perform relatively simple tasks like creating a task analysis, rewriting instructional content to be at a lower reading level, and generating open-ended, non-biasing interview questions? And should we?
• What is the quality and accuracy of ChatGPT’s work, and how can the quality of our human inputs impact the quality of ChatGPT’s outputs (e.g., garbage-in, garbage-out)?
• What is the extent of ChatGPT’s capabilities? For example, can it correctly identify critical tasks based on the applicable FDA definition of critical tasks (i.e., CDER or CDRH), or develop user needs/product requirements that are measurable and verifiable?
• What are the potential issues, disadvantages, and risks of using ChatGPT? For example, what are the repercussions of making decisions based on ChatGPT’s outputs, or enabling complacent “HFers” to lead HF activities with false confidence provided by ChatGPT? Or, what about confidentiality concerns?
To answer these questions in an engaging way, we expect to include live ChatGPT demos as frequently as possible. Our hope is to include audience input (e.g., topics of interest, specific questions) to generate real-time ChatGPT outputs, but we recognize that factors outside of our control might make this impractical (e.g., ChatGPT server is down or overcrowded). In case we cannot access ChatGPT live during our presentation, we will pre-record conversations with ChatGPT that we can show to the audience as examples instead.
Additionally, we are learning from other HF professionals how they currently use ChatGPT in their work. We distributed a survey on LinkedIn over the summer and continue collecting responses that give us high-level insights from respondents, and several respondents have indicated an interest in sharing their experience with us in more detail during an interview. Such learnings would be captured in our presentation as well.
Event Type
Oral Presentations
TimeTuesday, March 261:30pm - 2:10pm CDT
LocationContinental Ballroom, Lobby Level
Medical and Drug Delivery Devices