Webinar: What AI Can and Cannot Do: Intelligence Augmentation in Practice with Michael Bernstein

Value of the Information & Strength of the Argument

The talk provides a valuable, actionable framework that goes beyond generic AI hype. The argumentation is logically sound: it uses clear analogies (spam filters, customer service robots), distinguishes between technical capability and deployment tolerance, and builds a predictive model based on accuracy thresholds. The distinction between sharp-edged and rough-edged problems is compelling and supported by concrete examples from industry and academia. The speaker avoids overpromising and offers practical advice for organizations, such as converting sharp problems into rough ones and setting realistic expectations. The reasoning is accessible yet grounded in research, making it a useful resource for decision-makers.

Scientific Rigor, Source Quality, Title Accuracy

The scientific rigor is high: the speaker references peer-reviewed work (e.g., a Princeton study on agent reliability, an accessibility study from Germany) and his own research background, lending credibility. However, specific citations are not displayed, so viewers cannot easily trace the sources. The title matches the content well, accurately reflecting the session’s scope on AI limitations and intelligence augmentation. The webinar effectively conveys expert opinion and research-informed insights, but it would benefit from more explicit references for further reading.

The title accurately reflects the webinar's core theme: analyzing what AI can and cannot do, and how to leverage intelligence augmentation in practice.

8/10

The speaker is a Stanford professor with recognized expertise in human-computer interaction and AI. The talk presents a coherent, useful framework (sharp-edged vs. rough-edged problems) and references real research and examples, although full citations are not provided on screen.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 02:00Introduction to the webinar topic and the three curves of AI progress, products launched, and successes.
• 06:00Introduction of the sharp-edged vs. rough-edged problem framework.
• 10:00Examples of rough-edged (writing, video generation) and sharp-edged (bug fix, hospital readmission) problems.
• 15:00Agentic AI defined; explanation of how sharp edges compound in multi-step workflows.
• 20:00Discussion of tolerance for error using spam filters and virtual assistants as historical examples.
• 25:00Legal hallucinations and Princeton research on AI agent reliability.
• 35:00Prediction of future model improvements; turning sharp-edged problems into rough-edged ones.
• 45:00Transition to the importance of human-AI collaboration, illustrated with robotics examples.

• Stanford Online AI Courses and Programs — Mentioned in the video description as a resource for exploring AI concepts further.
• Stanford Agentic AI Program — Linked in the description as a program for building intelligent AI agents.