Artificial Intelligence in Dental Education: Opportunities and Challenges
Sergio Uribe, Assoc Prof
Riga Stradins University, Riga, Latvia
Ludwig-Maximilian University, Munich, Germany
ABSTRACT
Artificial intelligence (AI) and generative AI models have the potential to revolutionize dental education.In this presentation, we will analyze recent research from the perspective of dental educators worldwide, examine current guidelines for AI implementation, and propose a core AI curriculum to equip dental students with the skills necessary for a technology-driven future.We will also explore potential applications of AI in dental education, discuss its benefits and limitations, and provide examples of successful AI integration. By the end of the presentation, attendees will have a comprehensive understanding of the current state of AI in dental education and practical insights into how they can use AI to enhance teaching and learning.
Learning Objectives
Declaration of interests or financial relationships
I have no interests or financial relationships to disclose in connection with the subject matter of this presentation.
Sergio Uribe
Contents
The Context
We talked about what we'd learned, about families, everything. He was not being melancholy, it was very forward-looking, saying we haven't really improved education with technology...
AI and Generative AI
Traditional AI
Supervised Learning
Generative AI
Unsupervised Learning
Traditional AI: Age Estimation in Radiograph
Traditional AI
Traditional AI
Supervised Learning
Generative AI
Unsupervised Learning
Generative AI
Generative AI
Generative AI
Traditional AI
Generative AI
One caveat:
BIAS
Meteorological
data
Botanical
data
Cosmological
data
BIAS
Biased data Biased tuning
Dental Educators View about GenAI
AIM: Investigate dental educators' views on AI chatbots (e.g., ChatGPT) in education.
METHOD: Global survey (May-June 2023) using 31-item online questionnaire.
Not at all
Greatly
Believe Generative AI can Enhance Dental Education?
📈 31% of respondents already use AI tools, with 64% recognizing their potential in dental education.
⚠️ However, 53.9% expressed concern about AI reducing human interaction.
📋 What are the perceived barriers to using AI chatbots in dental education? �
📋 What is the dental educators perception of AI chatbots' impact?
Ramezanzade et al., 2023. J. Dent. 10.1016/j.jdent.2023.104732
🔍 Dental educators have a positive yet cautious view of AI chatbot integration in dental curricula.
📄Clear implementation guidelines are needed.
Start
Does it matter if the output is true?
Safe to use Chatbots as ChatGPT/Gemini/Anthropic
Do you have experts to verify that the output is accurate?
Are you able and willing to take full responsibility for the use of the content?
Possible to use Chatbots as GenAI
Unsafe to use Chatbots as GenAI
No
No
Yes
Yes
Yes
by Aleksandr Tiulkanov. Creative Commons license.
Implementing AI in Dental Education
Core AI Concepts for Dental Students
Item | What Students Should Learn |
Basic AI Concepts | Understand AI, machine learning, and how AI makes predictions from data. |
AI Applications in Dentistry | Learn AI use cases in dental imaging, diagnostics, and treatment planning. |
Evaluating AI | Know how to assess AI tools using metrics like accuracy, sensitivity, and impact on patient outcomes. |
Ethical and Governance Issues | Be aware of AI challenges (bias, explainability) and understand the importance of ethical use and oversight. |
AI, like any novel technology, �must adhere to established standards to ensure safe clinical use:
randomized controlled trials.
Key Responsibilities for Dental Educators �in Teaching AI
Key Takeaway | What Educators Should Focus On |
AI Literacy | Ensure educators understand basic AI concepts. |
Critical Appraisal of AI | How to evaluate AI trials, focusing on metrics like accuracy, bias, and clinical impact. |
Curriculum Integration | Implement the core AI curriculum |
Ethical AI Usage | Ethics in AI, including issues like explainability, bias, data privacy, and human oversight. |
How to implement
Item | Actionable Point |
Clear AI Policies | Develop and implement clear policies for AI use. |
AI Courses | Integrate AI courses into dental curriculum and staff education. |
AI Infrastructure Investment | Invest in AI infrastructure. |
AI Multidisciplinary Committee | Create a committee to oversee AI integration. |
Regular AI Policy Review | Regularly review and update AI policies. |
Integrating Generative AI in Dental Education: A Scoping Review of Current Practices and
Recommendations. Uribe E; Maldupa I, Schwendicke F. Working paper. (2024)
AI in Dental Education:
Opportunities and Challenges
sergio.uribe@rsu.lv
Sergio Uribe, Assoc Prof
Riga Stradins University, Riga, Latvia
Ludwig-Maximilian University, Munich, Germany
EXTRA SLIDES
Learn More
Effective GenAI Use
Craft Prompt
Analise response
Use Output
Review Prompt
Context
Role
Task
Output
Weak Prompt | Improved Prompt |
| I am writing a research paper… |
| You will act as an expert… |
Do this… | Your task is to… |
| and your output will be… |
The Role of AI in Solving Translational and Implementation Research Challenges in Dentistry
Sergio Uribe, PhD, MSc, DDS
Assoc Prof
Riga Stradiņš University. Riga, Latvia
Universidad Austral de Chile, Valdivia, Chile
“There is no function the computer cannot do in radiology”
Gwilym S. Lodwick, MD
Use of non-human evaluators in diagnostic healthcare
Pathology | Sensitivity (95% CI) | Specificity (95% CI) | Source | |
Sars-CoV 2 | 98% (95-100) | 99% (95-100) | Guest et al. J Travel Med. 2022;29. | |
Breast Cancer | 99% (97-100) | 98% (97-100) | Kure et al. Biology. 2021;10. | |
Colorectal cancer | 97% | 99% | Sonoda et al Gut. 2011;60:814–9. | |
Cancer biomarkers | discriminate between cancerous and healthy cells and between two cancerous lines, 99% accuracy | Piqueret et al. iScience. 2022;25:103959. | | |
(these tools)... achieve equal or better diagnostic performance for the detection of complex pathologies…
…however, the biggest challenge is translating what we see in the research setting into an operational setting
Photopoulos. Nature. 2022;606:S10–1.
Use of non-human evaluators in diagnostic healthcare
Pathology | Sensitivity (95% CI) | Specificity (95% CI) | Source | |
Sars-CoV 2 | 98% (95-100) | 99% (95-100) | Guest et al. J Travel Med. 2022;29. | |
Breast Cancer | 99% (97-100) | 98% (97-100) | Kure et al. Biology. 2021;10. | |
Colorectal cancer | 97% | 99% | Sonoda et al Gut. 2011;60:814–9. | |
Cancer biomarkers | discriminate between cancerous and healthy cells and between two cancerous lines, 99% accuracy | Piqueret et al. iScience. 2022;25:103959. | | |
Hinton 2016
“We should stop training radiologists now, it’s just completely obvious within five years (2021) deep learning is going to do better than radiologists.”
Hinton 2022
“The transition is slightly slower than I hoped but well on track for AI to be better than most radiologists at interpreting many different types of medical images by 2026”
Evaluation | | Prospective | |
| | No | Yes |
Multisite | No | | |
Yes | | | |
| | | |
Low risk Bias
High risk Bias
Wu et al. Nat Med. 2021;:1–3.
Hinton 2016
“We should stop training radiologists now, it’s just completely obvious within five years (2021) deep learning is going to do better than radiologists.”
Hinton 2022
“The transition is slightly slower than I hoped but well on track for AI to be better than most radiologists at interpreting many different types of medical images by 2026”
127 clinical findings, 821 681 chest x-rays from 520 014 cases
Seah et al. Lancet Digit Health. 2021;3:e496–506.
2 Key problems in Dentistry
Uribe, Innes, Maldupa. Int J Paediatr Dent. 2021;31:817–30.
The Implementation Gap
Schwendicke et al. Clin Oral Investig. 2019. 23, 3691–3703
Lee et al. Sci Rep. 2021;11:16807.
Better
diagnosis
Better
treatment
plus
Usual AI performance metrics
Therapeutic change
Patient outcomes
Societal outcomes
Clinically relevant performance metrics
Schwendicke et al. J Dent. 2022;119:104080.
Seah et al. Lancet Digit Health. 2021;3:e496–506.
Kiani et al. NPJ Digit Med. 2020;3:23.
Current Model Personalized AI
Personalized Medicine Precision AI
Low Sensitivity
Low Specificity
Best treatment?
Current Model Personalized AI
Personalized Medicine Precision AI
Low Sensitivity
Low Specificity
Best treatment?
2 Key problems in Dentistry
2 key metrics to keep in mind when developing an AI model in Dentistry
WHO Global Oral Health Action Plan goals
My new AI model/app/algorithm
increases dental UHC by (insert here)%
and
decreases (insert pathology here)� prevalence by (insert here)%
The role of AI in Solving Translational and Implementation Research Challenges in Dentistry
sergio.uribe@gmail.com
@sergiouribe
Cabitza et al. Exploratory Usability Evaluation of Activation Maps in Radiological Machine Learning. In: Machine Learning and Knowledge Extraction. 2022. p. 31–50.
Meskó, 2020. Nature Digital Medicine 3, 1–8.
Pathology
Radiology
2 855
3 398
Personalized AI
Paris Set al. How to Intervene in the Caries Process in Older Adults: A Joint ORCA and EFCD Expert Delphi Consensus Statement. Caries Res. 2020;54:1–7.
Schwendicke et al. How to intervene in the caries process in adults: proximal and secondary caries? An EFCD-ORCA-DGZ expert Delphi consensus statement. Clin Oral Investig. 2020. https://doi.org/10.1007/s00784-020-03431-0.
Splieth et al. How to Intervene in the Caries Process in Children: A Joint ORCA and EFCD Expert Delphi Consensus Statement. Caries Res. 2020;:1–9.
How to close the AI translational gap?
Image from Model development to clinical impact
How to close the AI translational gap?
How to close the AI translational gap?
The problem of Dentistry
How to transfer research findings into clinical practice efficiently.
Image Research and prevalence of oral diseases
Image Costs in Dentistry
AI as an opportunity
Three components of AI
Data: the fuel that drives AI
Quantity and quality
What we know about the data?
What to do?
Levels of maturity for dataset annotation
What to do?
Better reporting
What to do?
Better benchmarking: patient and societal outcomes
AI in Healthcare
CPU power + AI Algorithms + Data = AI in Healthcare
AI in Healthcare
CPU power + AI Algorithms + Data = AI in Healthcare
3 Key areas
Diagnosis
Treatment
Prognosis
Use of non-human evaluators in diagnostic healthcare
Sars CoV-2
Sensitivy 98% (95% CI 95-100)
Specificity from 99% (95% CI 97-100)
Breast Cancer Sensitivity 99% and specificity 98%
The Key: Nudge the appropriate treatment exploiting the possibilities of the user interface
Turing trap
Modified from Brynjolfsson. The Turing Trap: The Promise & Peril of Human-Like Artificial Intelligence. American Academy of Arts & Sciences.
Turing trap
Modified from Brynjolfsson. (2022) The Turing Trap: The Promise & Peril of Human-Like Artificial Intelligence. American Academy of Arts & Sciences.
New tasks than human can do
with the help of machines
e.g. � - how to get more value out of the �diagnostic tests we use
Things humans like to do or do well
like to do or do well�e.g. agree on a diagnosis
Things that humans
DON'T like to do or
DON'T do well
e.g. fill forms and �Administrative tasks
AI
Effect of prevalence (or pre-test probabilities)
Effect of prevalence (or pre-test probabilities)
Deep learning classification getting the "Right" answer for the wrong reason