Special Focus Session
SF 20a - Building trust: pitfalls in artificial intelligence in daily practice
Lectures
1
Chairperson's introduction
05:00Daniel Pinto Dos Santos, Frankfurt / DE
2
What should I know before buying an AI tool?
15:00Patrick Omoumi, Lausanne / CH
3
AI in practice: adoption, aversion and automation bias
15:00Mohammad Hosein Rezazade Mehrizi, Amsterdam / NL
4
The dark side of explainable AI: adversarial attacks
15:00Andrea Ponsiglione, Naples / IT
5
How sure is the AI tool: reporting together with AI certainty scores
15:00Natalia Alves, Nijmegen / NL
6
Panel discussion: Are we ready to trust AI?
25:00Panel discussion: Are we ready to trust AI?
5 min
Chairperson's introduction
Daniel Pinto Dos Santos, Frankfurt / Germany
15 min
What should I know before buying an AI tool?
Patrick Omoumi, Lausanne / Switzerland
- To review the main categories of points to consider before purchasing an AI solution.
- To provide a framework for the assessment of commercial AI solutions in clinical practice.
- To briefly discuss relevant regulations that apply to AI solutions in clinical practice.
15 min
AI in practice: adoption, aversion and automation bias
Mohammad Hosein Rezazade Mehrizi, Amsterdam / Netherlands
- To identify consequential biases and traps in using AI solutions in practice.
- To distinguish between healthy and unhealthy human-AI configurations.
- To reflect on the challenges and opportunities of developing expertise in using AI solutions in practice.
15 min
The dark side of explainable AI: adversarial attacks
Andrea Ponsiglione, Naples / Italy
- To learn about different technical approaches for explainable AI.
- To appreciate how adversarial examples can challenge AI systems.
- To discuss strategies to mitigate adversarial attacks and enhance the resilience of explainable AI models.
15 min
How sure is the AI tool: reporting together with AI certainty scores
Natalia Alves, Nijmegen / Netherlands
- To learn about the two main types of uncertainty affecting AI algorithms and their differences.
- To appreciate three methods for uncertainty quantification in AI algorithms and how they differ from model calibration.
- To learn about three applications of uncertainty quantification in radiology workflows and discuss strategies for effectively communicating uncertainty.
25 min
Panel discussion: Are we ready to trust AI?
Daniel Pinto Dos Santos
Frankfurt / Germany
Daniel Pinto Dos Santos
Frankfurt / GermanyPatrick Omoumi
Lausanne / SwitzerlandMohammad Hosein Rezazade Mehrizi
Amsterdam / NetherlandsAndrea Ponsiglione
Naples / ItalyNatalia Alves
Nijmegen / Netherlands