Log in or create a free Rosenverse account to watch this video.
Log in Create free account100s of community videos are available to free members. Conference talks are generally available to Gold members.
This video is featured in the Ed's Test playlist playlist.
Summary
Artificial Intelligence (AI) tools like large language models present opportunities — and risks — for people working with digital content. Can AI help with tedious tasks, like encoding and categorizing documents, or rewriting text snippets? Will AI be so good at these tasks that Information Architects (IAs) are no longer needed? In this session, Jeff and Karen provide an overview of what "AI" means for IAs, explaining the differences between natural language processing, machine learning, and large language models. They also dig into a real-world example of using different systems to categorize web content from Reddit. IAs will come away from the session reassured that the robots pose no threat to their jobs.
Key Insights
-
•
Natural language processing is about mathematically modeling words and their relationships but struggles with context and nuance.
-
•
Machine learning models detect statistical patterns but do not truly understand content or reason about it.
-
•
Large language models like ChatGPT work by predicting what text should follow based on massive internet-scale training data.
-
•
Tokenization breaks text into meaningful units, filtering out filler words to create useful representations for AI.
-
•
Choosing the right AI model and customizing prompts is more important than using the 'biggest' or most popular model.
-
•
Using AI for content categorization requires iterative refinement, precise instructions, and ongoing validation.
-
•
AI tools often generate plausible but overlapping categories, which complicates achieving mutually exclusive, exhaustive taxonomies.
-
•
Cost, time, and carbon footprint are significant considerations for scaling AI workflows, making simpler tools sometimes preferable.
-
•
Visualization techniques like Sankey diagrams and proximity charts help interpret AI categorization results and spot issues.
-
•
Human expertise in information architecture is more vital than ever to effectively harness AI in complex content tasks.
Notable Quotes
"AI is not going to save us. It is a tool that we use with significant human support and understanding."
"Trying to map all the possible connotations and contextual interrelationships between words is mindbogglingly computationally expensive."
"The risk with LLMs is that they invent new categories or start writing Reddit posts instead of just categorizing."
"The best large language models we tested only achieved about 47% accuracy in categorizing Reddit posts."
"Breaking down tasks into one thing at a time dramatically improves AI performance and reliability."
"Choosing a tool designed for your specific problem matters way more than just picking the smartest AI available."
"We ended up using garbage ID numbers for categories to keep the LLM from inventing new ones."
"It wasn’t obvious the difference between AI generating category ideas and validating categories until we tried both."
"Humans still excel at mental modeling of topics and nuance that AI struggles to replicate automatically."
"Foundational IA and metadata infrastructure are the meat and potatoes supporting scalable AI-driven experiences."
Or choose a question:
More Videos
"The 1500 are designers, we don't distinguish strictly between UX or visual or industrial—they all bring design to the org."
Adam Cutler Karen Pascoe Ian Swinson Susan WorthmanDiscussion
June 8, 2016
"Knowing your subject matter as a UX leader is just as important as knowing your craft."
Peter MerholzThe Trials and Tribulations of Directors of UX (Videoconference)
July 13, 2023
"If you don’t tune it properly, sometimes you just don’t get what you want."
Lisa WelchmanCleaning Up Our Mess: Digital Governance for Designers
June 14, 2018
"Sustainable practices are not just a luxury; they are a necessity for our survival."
Vincent BrathwaiteOpener: Past, Present, and Future—Closing the Racial Divide in Design Teams
October 22, 2020
"The greatest sign of success for a teacher is just to be able to say the children are now working as if I did not exist."
Brenna FallonLearning Over Outcomes
October 24, 2019
"Linear perspective created this myth that we can see reality from a single perspective, which is the source of all misunderstanding."
Tricia WangSpatial Collapse: Designing for Emergent Culture
January 8, 2024
"If you don’t bring stakeholders into the research journey, they won’t believe or use the data."
Edgar Anzaldua MorenoUsing Research to Determine Unique Value Proposition
March 11, 2021
"We want to build teams with diverse skill sets so we can create a full picture during the knowledge creation phase."
Designing Systems at Scale
November 7, 2018
"The conversion design process creates collective knowledge, which gets reinfused to strengthen future experiments."
Erin WeigelGet Your Whole Team Testing to Design for Impact
July 24, 2024
