Dr Aaron Snoswell and Dr Kevin Witzenberger led an interactive and lively discussion for the Faculty on Generative AI — its capabilities, challenges, and implications for our work and society.
Key points included:
-
Understanding GenAI: The distinction between analytical AI (rule-based, predictive systems) and generative AI (systems that create new content such as text, images, and video).
-
Emerging Capabilities: GenAI models are becoming capable of independently performing tasks for minutes to hours (knowing when to check in for human oversight or direction) and of stronger reasoning abilities, including the ability to document its steps of reasoning, and “self-reflect” (spot mistakes and correct them or adjust course). These tools are increasingly capable of performing complex tasks, from writing reports to managing emails and even coding.
-
Improving Productivity: GenAI tools are estimated to improve productivity by ~20%, comparable to the 20-30% improvement brought on by generalised computing. Attendees discussed how they currently use GenAI tools, and how they might want to use them in the future.
-
Critical Reflections: The session also addressed the limits and risks of GenAI, including:
-
Political economy of AI must be considered, when leading models (such as OpenAI, Microsoft, Anthropic models) are proprietary, not open-source
-
Training data can be biased, and GenAI may be inherently conservative as it can only generate based on past data
-
Environmental and operational costs can be incredibly high
-
Operational logic of GenAI tools is determined by who develops it – inbuilt assumptions of the “right” way to do things may not align with what practices are actually valuable in a given industry
-
-
Responsible Use: Attendees mapped potential uses of GenAI along axes of responsible/irresponsible and human-centered/efficiency-centered use.