Lecture 5 + Lecture 5 (IDK)

AI helps explain cognition and the brain, while Cognitive Psychology and Neuroscience help simulate cognition and the brain

An abstract representation of a system or reality that helps to understand, predict, or explain something

A model of cognitive processes that can describe, simulate, predict, or explain mental functions

To describe, simulate, predict, and explain cognitive processes

They may oversimplify processes

They may still remain abstract and simplified

They may overfit or depend heavily on the quality of training data, and raise ethical issues such as bias and fairness

They may be too abstract, disconnected from evidence, or risk circular reasoning

Models that only tell what is going on without simulating mechanisms

Models with labeled boxes representing cognitive functions and arrows representing information flow

What happens inside the box and how does the arrow work?

Models that simulate or predict information processing at some level of abstraction, formulated mathematically or as computer simulations

No, because models are constantly evolving, none generalize across all functions, and often multiple models exist for one function

Frameworks that integrate multiple models, generalizing across cognitive functions and inspired by brain structures

Cognitive models describe specific functions, while architectures describe how these functions integrate and collaborate

Defines what the system does, e.g., detecting people, traffic signs, and vehicles

Defines how the system processes information, e.g., using algorithms or CNNs

Defines where and how the process is implemented, e.g., hardware and software in a car

The “where and how” pathway toward the parietal lobe, specialized for spatial processing and action-related representations

The pathway toward the temporal lobe, specialized for object recognition and memory

They break down a function into goals (what), processes (how), and implementation (where)

Time must be considered when modeling cognitive phenomena

Biological, Cognitive, Rational, and Social, each corresponding to different timescales of processes

1–10 ms, concerning neuron firing and synaptic delays, basic cognition and motor movements

70–700 ms, perceptual and cognitive operations like recognition and rapid decisions

10 seconds–10 minutes, involving reasoning, problem-solving, and planning

10 minutes–several hours, involving social interactions and higher-level tasks

They help understand processes at different levels and timescales, providing frameworks for model design and practical applications

They adapt to user behavior, simulate cognition, improve usability, support tutoring, games, and adaptive interfaces

User’s behavior, user computation, and user process

Adapting to user responses without representing their mental state

A mathematical representation of user abilities and performance linked to psychological concepts

Incorporating detailed cognitive processes, constructs, and strategies, e.g., ACT-R in tutoring systems

They enable human-like experiences, adaptive decisions, and better design of software and interfaces

Describe (Insight in (temporal) aspects of a process; Foundation to develop a model)

Can be vague

To provide a dynamic, replicable view of cognitive processes, which helps understand underlying mechanisms.

Risk of oversimplification and lack of validation

Forecast future behavior

May overfit, or dependent on data quality; Ethical considerations

How cognitive processes occur

Explanation can be too abstract; Risk of circular reasoning

Apply to new context

Misinterpretations

An overarching framework that describe the structure in which individual cognitive models can operate and interact.

A foundation for understanding, simulating, and creating intelligent systems that can perform multiple cognitive function.

Towards Parietal Lobe

Spatial processing (where)

Towards temporal lobe

Object recognition and memory

Computational, Algorithmic, and Implementation Level

"What is the goal?" or "What information is processed and why?"

"How can this be done?" or "How do we process it?"

"With what hardware is this done?" or "Where / how is this process implemented?"

Computational Level

Algorithmic Level

Implementational Level

The level below (e.g., Algorithmic)

The level above (e.g., Algorithmic)

1-10 ms (neuron firing, synaptic delay)

70-700 ms (perceptual/cognitive operations, rapid decisions)

10 sec - 10 min (reasoning, problem-solving, planning)

10 min - several hours (social interactions)

Break down at different levels (what, how, where)

Understand their timing and duration (bands)

Help to design better algorithms, their goals and how to implement them

Help to match systems more closely to time-scales of human thinking and acting

A mathematical representation of user abilities and performance (e.g., adaptive learning platforms using IRT).

Adapt to user responses, without a representation of the ‘mental state’ (e.g., a recommender system).

Incorporating detailed cognitive processes, constructs and strategies (e.g., a learning tutor using ACT-R theory).