Task 1-Cognitive science, artificial intelligence, prediction Flashcards
Article: Clinical versus Mechanical Prediction (Grove)
• Mechanical-prediction techniques were about 10% more accurate than clinical predictions
• Depending on the specific analysis, mechanical prediction substantially outperformed clinical prediction in 33% to 47% of studies examined
• Superiority for mechanical-prediction techniques was consistent, regardless of the judgement task, type of judges’ amounts of experience, or the types of data being combined
• Humans are susceptible to many errors in clinical judgment
Include ignoring base rates, assigning nonoptimal weights to cues, failure to take into account regression toward the mean, and failure to properly assess covariation
Heuristics can similarly reduce clinicians’ accuracy
Clinicians often do not receive adequate feedback which gives them scant opportunity to change
machine learning
- Machine learning – the ability of computers to learn without being explicitly programmed involves programming computers to learn from example data or past experience
- Environments with high degree of complexity are where machine learning is most useful
- AI developers now understand that it is easier to train a system by showing it examples of desired input-output behaviour than to program it manually by anticipating the desired response for all possible inputs
prediction
Prediction – anticipating what will happen in the future -> huge decrease in costs of prediction
Machine learning can be used to predict whether a bank customer will default on a loan
Also be used to develop medical diagnosis using symptoms -> predicting presence of disease
• Prediction becomes more valuable when data is widely available and more accessible
• As data availability expands, prediction becomes increasingly possible in wide variety of tasks
• A vehicle could drive autonomously by predicting what a human driver would do in response to set of inputs (inputs from camera images for example)
The Managerial challenge
- Prediction is not the same as automation
- The most valuable workforce skills involve judgement
• The future’s most valuable skills will be those complementary to prediction (= judgement)
• Judgement skills will be in greater demand if price of prediction falls due to advances in AI - Managing may require new set of talents and expertise
Deep learning
learning method based on several layers between input and output, allows to learn overall relation in successive steps
Intelligence analysis
• Intelligence analysis involves searching for, selecting, processing and interpreting data in order to gain an awareness of current situations and forecast potentially important future developments in areas of interest to decision-making stakeholders
3 pillars of counselling
• Definition can be broken down into 3 pillars of counselling
a. Forming a professional relationship
b. Empowering
c. Accomplishing goals
• Act of counselling requires fulfilment of all 3 pillars
fuzzy expert systems
• Fuzzy expert systems – expert systems that use fuzzy logic instead of Boolean logic
Fuzzy logic = a method of reasoning that deals with approximate values rather than fixed and exact values and is useful for working with uncertainties during decision making
Used to assist with the optimization of rules and membership classification
• Ex: in an AI-based mental health diagnostic expert system
problems counselling and AI
• Therapeutic bonds with the AI?
• AI lacks warmth, empathy and compassion
• Cultural differences and expectations?
• Legal and ethical considerations (errors, legal responsibility…)
advanced artificial intelligent agents may be capable of developing their own personal values and beliefs that inform decisions—which raises the question of whether those decisions will be consistent with those of their creators or the cultural context of use
concerns about who should be legally responsible for the decisions and any mistakes made by AI systems
moral and ethical considerations
should not harm humanity
• Job loss in mental health care
only improve and advance psychological practice and research, but have the potential to supplant mental health care professionals in core activities that require human intelligence and social interaction
Economic impact
Can be applied to any knowledge-based profession
Current and Future Applications and Implications (Luxton) of ai in psychological practice
Virtual reality (VR) stimulation \+Augmented reality \+Therapeutic computer games \+EMR (Electronic medical record \+The Super Clinician \+Expert systems
• The criteria for evaluating theories of mental representation (MR
- Representational power – how much information a particular kind of representation can express
- Computational power – mental representations are important for what they express but also for what you can do with them; computational power can be evaluated in terms of how it accounts for three important kinds of high-level thinking
Problem-solving – a theory of MR should be able to explain how people can reason to accomplish their goals; should be able to explain (1) planning, (2) decision making, (3) explanation
Planning requires a reasoner to figure out how to get from initial state to goal state
Decision-making involves selecting best choice from different means of accomplishing goals
Explanation requires figuring out why something has happened - Psychological plausibility – cognitive science has the goal of understanding human cognition, so theory must be concerned with how people think;
Must explain the particular ways that humans carry out a task, not only how the task is possible computationally
Must account for the quantitative results of psychological experiments concerning qualitative capacities - Neurological plausibility – must be consistent with neuroscientific experiments
- Practical applicability – as there are many desirable practical results to which understanding the human mind can lead, theory must tell us about (1) education, (2) design, (3) intelligent systems; (4) mental illness
machine learning algorithm
• Machine learning algorithms can be seen as searching through a large space of candidate programs (which have been guided by training experience) to find a program that optimizes the performance metric
There are 3 major paradigms in machine learning
• There are 3 major paradigms in machine learning:
1. Supervised learning – learning a function that maps an input to an output based on example input-output pairs; most widely used machine learning method
Parallels concept learning in human and animal psychology
Exemplify function approximation – infers function from labeled training data consisting of a set of training examples; training data take the form of a collection of (x, y) pairs
Inferred function can be used for mapping new elements – the goal is to produce a prediction y* in response to a query x*
Requires learning algorithm to generalize from training data to unseen situations in a “reasonable” way
Form predictions via learned mapping f(x) which produces an output y for each input x (probability distribution over y when given x); different forms of mapping f exist including (1) decision trees/forests, (2) logistic regression, (4) support vector machines, (5) neural networks, etc.
one high-impact area of progress in supervised learning involves deep networks
- Unsupervised learning – a type of self-organized Hebbian learning that helps find previously unknown patterns in data set without pre-existing labels; known as self-organization and allows the modeling of probability densities of given inputs
- Reinforcement learning – concerned with how software agents ought to take actions in an environment in order to maximize some notion of cumulative reward; intermediate between supervised and unsupervised learning
Does not need labelled input and output pairs to be presented
Does not need sub-optimal actions to be explicitly corrected
Focus is on finding a balance between exploration (of uncharted territory) and exploitation (of current knowledge)
cognitive science
• Cognitive science is the scientific study of the human mind, a highly interdisciplinary field, combining ideas and methods from psychology, computer science, linguistics, philosophy, and neuroscience
Marr
Marr describes 3 levels of analysis
1.Computational level – specifies the goals of a process, the suitability of the process, and the logic behind the manner in which it is executed; what does the system do?
2.Representational and algorithmic level – address how the process can be executed, present a representation of the inputs and outputs, the algorithms which transform input into output; what steps does the system go through?
Cognitive psychologists frame their hypotheses at this level
3.Hardware implementation level (physical level) – addresses how algorithm and representation may be physically realized; in what ways are the steps the system goes through implemented?
Neurobiologists describe the world at this level
Singularity Principle
Singularity Principle: Denotes a point in which a function becomes infinite and Kurzweil says that this is the point beyond which technological evolution is so fast that humans cannot predict or understand what will happen.
•
Turing Point
Turing Point: The point where machine intelligence exceeds human intelligence
expert systems
Systems designed to incorporate knowledge and ability of a human expert in a domain these Systems can identify patterns, trends and meaning from complex data
Moores law
says that the complexity in computer circuits doubles every two years
Strong AI
creative conscious social, AI has intellectual abilities that are undistingusihabke from human
weak AI
generell intelligence
symbolic
traditional AI , rule based (1;=)
sub symbolic
aspects of the body (movement, visualisation)are needed for higher intelligence
neural networks and connectionism
Cognitive modelling
development of psychological concepts – translated to computer programs
fuzzy logic
Probabilities (truth values 0-1) instead of clear cut categories (0 or 1)
Boolean logic
values reduced to true or false (0 or 1
Marovec‘s paradox:
Higher level thinking requires less computation than lower level (e.g. perception) – White collar jobs are more replaceable than blue collar jobs
Singularity
Unforseeable point in future (when AI will be more intelligent then humans)
Moore‘s Law
processor speeds and processing power for computers will double every two years, while prices go down
MABA/MABA list
‘Men are better at, Machines are better at’
Searle‘s Chinese Room:
A computer communicates with a woman by using only a book, without the woman realizing it.
BUT – the system does not understand what it is doing.
Strong AI: the system understands what it is doing (conscious, social)
Weak AI: System simulates ability to understand (general intelligence)
a computer performing an input output performance is not a strong AI
Turing test
Turing test: PC must have a real time conversation
Undistinguishable whether it is a machine or a human
Thagard: Computational Representational Understanding of Mind (CRUM)
Three way analogy of mind, brain and computer
Traditional computers are serial processors
The mind is a parallel processor
Mind (cognitive theory)& machine (computational model)
Mind (Cognitive Theory) Machine (Computational Model) mental representational structures (neuronal structures) = Data structures
Computational procedures operate on those structures
(neuronal firing)
=
algorithms
CRUM Advantage
representations are computable & quantitative (wich helps making predictions)
CRUM limits
ignores emotion, consciousness, the body, physical environment and social thought
Crum analogy
representations= nodes /data structures
mental procedures,computations
=algorithms
thinking is like running a program
prediction types
Timelines and outcome predictions with specific dates
Scenarios: conditional prediction: if A, then B
Plans: conditional: if someone follows a specific plan, a specific goal will be achieved
Issues and metastatements: includes all approaches (also impossibilities
prediction methods
Causal models: physics and hard sciences
Non-causal models: without causality
Outside view: examples grouped together (trends)
Philosophical arguments: highlighting problems (even impossible ones)
Expert judgements: fill incomplete predictions
Non-expert judgement
Assessment of AI predictions
Extracting falsifiable predicitons
Clarifying and revealing assumptions
Empirical evidence and the scientific method
The reliability of expert judgement
Scientific Method = Best
mechanical judgement
Reproducible, statistical, objective
clinical judgement
Unspecific, subjective
ELIZA:
First psychotherapist Simulation
Expert systems:
use fuzzy logic and pattern identification
Able to learn
Virtual reality
Phobias, dementia, research
Artificial Companion:
Autism
Marr‘s Tri level hypothesis (T1
Computational level: what does the system do (e.g.: what problems does it solve or overcome) and similarly, why does it do these things
Algorithmic level (sometimes representational level): how does the system do what it does, specifically, what representations does it use and what processes does it employ to build and manipulate the representations
Implementational/physical level: how is the system physically realized (in the case of biological vision, what neural structures and neuronal activitiesimplement the visual system)
Ai
system designed by humans , they act in the world by perceiving environment , interpret data, reason on knowledge and decide on action
expert system
systems designed to incorporate knowledge and ability of humans expert - these systems can identify patterns , trends and meaning from complex data
super clinician
His could make a super clinician by providing her with capabilities beyond humans (infra rot)
Judgement
Judgment is the ability to make considered decisions and understand the impact different actions will have on outcomes in light of prediction.
New modes of machine learning might find ways to examine the relationships between actions and outcomes, and they use this info to improve predictions. It can help virtual assistants mimic human judgment, so that over time the feedback can turn some aspects of judgement into prediction problems.
Tasks where the outcome can be easily described and there’s limited need for human judgment are easier to automate.
In cases where decisions can be clearly defined with an algorithm, we can expect to see computers replace humans.
The managerial challenge
As AI technology improves, predictions by machines will increasingly take the place of predictions by humans. To assess what role humans play, consider 3 interrelated insights:
I. Prediction is not the same as automation:
A task exists of data, prediction, judgment and action. Prediction is an input in automation,
machine learning involves just one component→prediction.
II. The most valuable workforce skills involve judgment:
Human workers have remained involved in prediction tasks, employers will want workers to augment the value of prediction→valuable skills will be those that are complementary to prediction. We can only speculate what aspects of judgement are apt to be most vital.
III. Managing may require a new set of talents and expertise:
As AI becomes better at prediction, managers’ prediction skills will become less valuable while their judgment skills become more valuable→their role will involve determining how best to apply AI.
The biggest challenges for managers and other employers will be:
- Shifting from prediction-based training to judgment-based training for employees.
- Determining the development of AI and timing the shift in training well not too early, not too
late.
- Assemble the most effective teams of humans who judge and AI machines that predict.
a rule
if you want to pass you need to study
Marr
However, the mind or brain cannot be understood if it is only studied at one level. Marr (1982) describes 3 levels of analysis:
I. Computational level: specifies the goals of a process, the appropriateness of the process, and the logic behind how it is implemented; what does the system do?
II. Representational and Algorithmic level: deal with how the process can be executed, give a representation of the inputs and outputs, the algorithms that turn inputs into outputs; what steps does the system go through?
o Cognitive psychologists formulate their hypotheses at this level
III. The hardware implementation level (physical level): addresses how algorithm and representation can be physically realized; in what way are the steps the system goes through
implemented?
o Neurobiologists describe the world at this level
7 deadly sins
overestimation in short run/understeimatin in long run
Imagining magic
Performance vs. competence(generalisation of skills, computers are often very narrow skilled)
Suitcase words
Exponentials
Hollywood scenarios
Speed of deployment
Areas in wich AI can improve
Psychopathology
medicine
Decision science could help the IC to improve its forecast accuracy by statistical or behavioral interventions.
