Please enable JavaScript.
Coggle requires JavaScript to display documents.
DHA 6040: HC Artificial Intelligence - Coggle Diagram
DHA 6040: HC Artificial Intelligence
Agent Based Modeling
Equation Based Model: Y = ALαKβ
Y = output, L = labor input, K = capital input, A, α, β are constants determined by technology
ROAMEF Cycle: rationale, objectives, appraisal, monitoring, evaluating, and feedback
enables a researcher to create, analyze, and experiment with models composed of agents that interact within the environment
Scale models – smaller versions of target
Ideal-type – some characteristics of target are exaggerated to simplify the model
Analogical model – based on drawing analogy between some better understood phenomenon & the target
Agent characteristics: perception, performance (motion, communication, action), memory, policy
Macro level features & patterns: precise definitions of collectives is open to negotiations, no accepted consensual definition that can be used, many members share characteristics in common, membership of collectivity entrails process related knowledge, features thought to be relevant can change
Micro level behavior: position of actor at a moment in time in an abstract knowledge space represents knowledge that he knows at that time, some actors are of higher status and other actors are motivated to gain that status, actors with highest status want to preserve their status
AI Learning
AGI—Artificial general intelligence
Machine Learning- the ability for an algorithm to learn from prior data to produce a behavior, teaching machines to make decisions in situations they have never seen. machine interprets, produce actionable insights/predictions
Deep Learning- ML where artificial neural networks –algorithms inspired by the way neurons work in the brain-find patterns in raw data by combining multiple layers of artificial neurons, as the layers increase, so does the neural network’s ability to learn increasingly abstract concepts, map input/transform data/precise output
Natural language Processing-level of understanding when AI communicates with humans and humans communicate with each other, key part of how computers understand and manipulate language, unstructured data, uses deep learning algorithms
Turing test - game with 3 players / 2 human and 1 computer. Evaluator asks open ended questions and tries to determine who is the computer. If evaluator cannot tell difference then computer is considered intelligent
Kurzweil-kapor test - computer carries on conversation for 2+ hours that 2/3 judges believe is human talking
Coffee test - robot must be able to go into strangers home, locate kitchen, and brew a cup of coffee
Strong AI - machine truly understands what is happening, Artificial General Intelligence (AGI)
Weak AI - machine is pattern matching and focused on narrow tasks, ie. Siri
Current AI only as smart as we can program it to be
Trainers: AI systems can automate tasks and find patterns in data, but still require humans to provide meaning, purpose, and direction
Explainers: Advancing AI algorithms often have a “black box” nature, making suggestions without clear explanations, requiring humans versed in both the technical and application domains to explain how such algorithms can be trusted to drive practical decisions
Sustainers: The intelligence needs of human endeavors will continually evolve, preventing the advent of “completed” AI systems. Humans must continue to maintain, interpret, and monitor the behavior and unintended consequences of AI systems
kNN: k-nearest neighbor - algorithm based on the notion that values that are close together are good predictors for a model
Neural networks: CNN (patterns in images), RNN (sequential data), GNN (predict relationships between graph data)
Theories/Principles
Complexity theory is a scientific theory that explains how systems can display behavioral phenomena that are not present in any individual component of that system. It emphasizes interactions and feedback loops that constantly change systems
Network Theory: Understandings complex systems within a complete system
CRISP-DM Process-method to manage data for a project with steps that include business understanding, data understanding, data preparation, modeling, evaluation, and deployment
Key considerations
Data Quality Matters
Transparency and Documentation
Bias in Data
Choosing Metrics for Model Performance
Stakeholder Education and Managing Expectations
Utility Assessment
Validation: Clinical Validation Axis, Alignment with Clinical State
Data dependency
Algorithmic bias
Laws/Policies
FDCA: Enforced by the FDA, the FDCA regulates the safety and effectiveness of drugs, medical devices, and some medical software, especially clinical AI systems
HIPAA: establishes privacy and security requirements for certain health information. The HIPAA Breach Notification Rule mandates notifications of health information breaches for applicable entities
Common Rule: This rule sets requirements for research involving human subjects that are federally funded or conducted at institutions receiving federal research funding. It includes the need for institutional review board review and is enforced by the Office for Human Research Protections
FTCA: prohibits deceptive and unfair trade practices affecting interstate commerce. This can relate to false health claims, misleading software performance representations, or consumer privacy and data security, all of which may pertain to healthcare AI products
FTC Health Breach Notification Rule: this FTC rule mandates that certain businesses notify consumers of breaches of personal health record information, which includes data collected for training, validation, or use in healthcare AI systems
State tort law: In the context of healthcare AI, this can apply if the behavior of developers, providers, hospitals, or other healthcare entities falls below the standard of care, with the specifics determined by state law
Application
Disease surveillance
Environmental & occupational health
Use AI systems to engage, rather than stifle, uniquely human abilities
Use automated systems to reach patients where existing health systems do not
supervised learning algorithms: neural networks, linear regression, logistic regression, support vector machines, kNN, random forest
unsupervised learning: clustering, association rules, dimensionality reduction
Big data: 5 V's (volume, velocity, variety, veracity, value)
Examples of deep learning AI: ChatGCP, DALL-E, Midjourney, Stable Diggusion
Apha-fold: makes protein structures instantly available without need for costly lab work/ now know how structure of 98.5% of proteins
Adoption process
Get familiar with AI
Identify the project you want AI to Solve
Prioritize concrete value
Acknowledge the internal Capability Gap
Bring in Excerpts and Set up a Piolet Project
Form a taskforce in integrate data
Start Small
Include Storage as part of your AI Plan
Incorporate AI as Part of your daily tasks
Build with balance
RPA (Robotic Process Automation) - automate workflows
Physical robots
Use cases: security, floor-scrubbing robots, online pharmacy, robot scientists
Cognitive AI
Cognitive computing helps us make smarter decisions. AI makes the decisions for us.
Neuromorphic computing – brain inspired computing as it relates to hardware and software systems modeled on neural systems
Decentralized autonomous organization – complex smart contracts that define bylaws of organization into the smart contracts
Ex. (bitcoin – money without banks)
Apple watch: predict health events, call 911 for emergency recordings
MRI reading assistance
Surgical robots to improve precision
Early predictions of disease based on health patterns