Artificial Intelligence

Artificial Intelligence

Current Trends 3T25-26

Artificial intelligence coined

The term “artificial intelligence” (AI) was coined in the 1950s

Artificial intelligence definition #1

Idea of building machines capable of performing tasks normally performed by humans

Artificial intelligence definition #2

Machines capable of performing human tasks

Machine learning definition

A subdomain of AI that learns intrinsic statistical patterns in data to cast predictions on unseen data

Deep learning definition

Machine learning technique using multi-layer mathematical operations for learning and inferring on complex data like imagery

Popular ML model

Neural networks (NNs)

Neural networks advantage

Outperform classical ML algorithms on complex data structures such as imagery or language

Main constituent of NN

Artificial neuron

Artificial neuron definition

Mathematical non-linear model inspired by the human neuron

How neural networks are engineered

By stacking and concatenating artificial neurons and connecting layers using mathematical operations

Purpose of engineered neural networks

Solve specific tasks like image classification

Example of image classification

Radiographic image showing a decayed tooth: yes or no

Natural intelligence workflow

Perception → Interpretation → Response

Software 1.0 workflow

Data and rules → Explicit programming → Interpretation and actions of human agents

Software 2.0 machine learning workflow

Data and outcomes → Engineered features → Feature mapping → Interpretation and actions of human agents

Software 2.0 deep learning workflow

Data and outcomes → Feature learning and mapping → Interpretation and actions of human agents

Goals of AI-based applications in dentistry

Streamline care; Relieve workforce from laborious routine tasks; Increase health at lower costs; Facilitate personalized, predictive, preventive, and participatory dentistry

Value of AI applications in dentistry

Improving access to and quality of care

Another value of AI in dentistry

Increasing efficiency and safety of services

Another value of AI in dentistry #2

Empowering and enabling patients

Another value of AI in dentistry #3

Supporting medical research or increasing sustainability

Important ethical considerations in dental AI

Individual privacy, rights, and autonomy

Possible solution to privacy concerns

Shift from centralized to distributed/federated learning

Benefit of federated learning

Improves scalability and robustness

Requirement for dental AI solutions

Trustworthiness and generalizability need to be guaranteed

How trustworthiness is maintained

Continuous human oversight and standards grounded in evidence-based dentistry

Explainable AI definition

Methods to visualize, interpret, and explain the logic behind AI solutions

Main reason AI not routine in dentistry

Limited data availability, accessibility, structure, and comprehensiveness

Another limitation of dental AI

Lacking methodological rigor and standards in development

Another limitation of dental AI #2

Practical questions around value and usefulness

Another limitation of dental AI #3

Ethics and responsibility

AI history milestone #1

1943 – Artificial neuron

AI history milestone #2

1957 – Neural networks

AI history milestone #3

1960s – First AI winter

AI history milestone #4

1980s – Rule-based expert systems

AI history milestone #5

1990s – Second AI winter

AI history milestone #6

2006 – Deep learning

AI history milestone #7

2012 – Computer vision leverages neural networks

AI history milestone #8

2015 – AI-based systems outperform human experts in image classification

AI history milestone #9

2015/16 – AlphaGo defeats world-best Go players

AI history milestone #10

2018 – Sundar Pichai compares impact of AI with disruptiveness of electricity and fire

Machine learning definition in AI history

Learning intrinsic statistical patterns and structures in data allowing predictions for unseen data

Deep learning definition in AI history

Form of machine learning using multi-layered deep neural networks trained to learn features of complex data structures

AI history characterization

Characterized by ups and downs

Current optimism in AI

Optimism is greater today than ever before

Applications of AI in dentistry

Image analysis; Prediction making; Record keeping; Dental research and discovery

Dental education role in AI

Foster digital literacy in future dental workforce

AI in dentistry now sample size

Largely under 2,000 instances/images

AI in dentistry future sample size

Millions of multi-level connected instances

Current AI data sources

Single hospitals; Insurance claims data

Future AI data sources

Federated learning; Data from multiple institutions

Current AI focus

Detection of structures on imagery; Association modelling

Future AI focus

Multi-class detection of pathologies; Predictive modelling; Decision support

Current AI training mode

Supervised learning

Future AI training mode

Unsupervised or semi-supervised learning

Current AI testing mode

Cross-validation

Future AI testing mode

Hold-out test set; Independent datasets

Current AI metrics

Measures of accuracy such as accuracy, area-under-the-curve, F1-score, segment overlap

Future AI metrics

Measures of value including impact on treatment decision, clinical and patient-reported outcomes, cost-effectiveness, and trustworthiness

Current AI study types

Diagnostic accuracy studies on retrospectively collected data

Future AI study types

Randomized controlled trials or large cohort studies collecting data prospectively

Black box AI model example

Deep neural network classifies image as “rooster”

Explain prediction method

Backward redistributing output to input

Correct image classification example #1

Image classified as rooster because of rooster’s comb and wattles

Correct image classification example #2

Image classified as cat because of cat’s ears and nose

Incorrect image classification example

Image classified as horse because of a copyright tag

Reference for AI in dentistry

https://journals.sagepub.com/doi/full/10.1177/0022034520915714

``` oai_citation:0‡6CurrTrendArtificialIntelligence3T25-26 (1).pdf