3. Biology & Technology

• Brain = biological basis of intelligence (learning, memory, problem-solving)

• Neuroscience studies brain & nervous system

• Brain vs. computers:

  • Brain → neurons + chemical signals

  • Computers → circuits + algorithms

• Both aim to:

  • perceive environment

  • process information

  • act intelligently

• Insight into brain → helps design better AI systems

• Brain:

  • ~1.4 kg

  • ~86 billion neurons

• Controls:

  • movement, reasoning, memory, emotions, language

• 4 lobes:

  • Frontal lobe → planning, decision-making, personality, motor control

  • Parietal lobe → sensory processing (touch, pain, spatial awareness)

  • Temporal lobe → hearing, language, long-term memory

  • Occipital lobe → vision

• Hippocampus → forms long-term memories, spatial navigation

• Amygdala → emotions (fear, reward), emotional memory tagging

• Cerebellum → coordination, balance, error correction, learning

• Frontal → plan movements

• Parietal → sense finger position

• Temporal → process sound/music

• Occipital → visual input

• Hippocampus → store memory

• Cerebellum → coordination

• Amygdala → emotional experience

• Neurons = basic processing units

• Functions:

  • receive, process, transmit signals

• Dendrites → receive signals

• Soma (cell body) → processes input

• Axon → sends signals

• Action potential = neuron “fires”

• Inputs:

  • Excitatory → increase firing

  • Inhibitory → decrease firing

• Synapses:

  • gaps between neurons

  • neurotransmitters carry signals

• Synapses change strength over time → learning mechanism

• Frequent use → stronger connections

• Rare use → weaker connections

• Brain = complex adaptive network

• Default Mode Network → introspection

• Salience Network → detects important stimuli

• Central Executive Network → goal-directed behavior

• Neuroplasticity:

  • brain reorganizes itself

  • adapts based on experience

• Memory = storing & using information

• Learning = forming & strengthening memories

• Short-term (working memory):

  • temporary (seconds–minutes)

  • limited capacity

• Long-term memory:

  • long duration (days–lifetime)

  • large capacity

• Declarative (explicit):

  • Semantic → facts

  • Episodic → personal experiences

• Procedural (implicit):

  • skills (e.g., riding a bike)

• Consolidation:

  • short-term → long-term memory

  • sleep is important

• Emotional events → remembered better

• Memory types:

  • Iconic (visual) → ~0.5 sec

  • Echoic (auditory) → few seconds

• Memories change over time (reconsolidation)

• “Neurons that fire together, wire together”

• Simultaneous activation → stronger connections

• Basis for associative learning

Key properties

• Local learning (only involved neurons)

• Unsupervised (no external control)

• Builds internal representations

Example

• Smell + positive experience → strong association

• Brain learning principles inspire AI:

  • neural networks

  • adaptive learning

• Key concepts:

  • synaptic plasticity

  • Hebbian learning

  • network-based processing

• Studies how the mind:

  • learns, reasons, decides, acts

• Focus: information processing, not just brain biology

• Interdisciplinary field:

  • psychology, neuroscience, linguistics, philosophy, computer science

• Goal:

  • understand & model intelligent behavior (human + AI)

• Psychology:

  • behavior, memory, attention, learning, decision-making

  • uses experiments & observations

• Neuroscience:

  • brain structures & neural activity

• Linguistics:

  • language structure, meaning, communication

  • important for NLP in AI

• Philosophy:

  • nature of mind, knowledge, reasoning

• Computer Science / AI:

  • builds computational models of cognition

  • simulates perception, memory, decisions

➡️ AI ↔ Cognitive science = mutual relationship

• Core functions of any intelligent system:

  • Perception → interpret environment

  • Attention → select relevant info

  • Memory → store & use information

• Not just sensing → interpreting information

• Uses:

  • context

  • prior experience

  • expectations

• Handles:

  • incomplete, noisy, ambiguous data

Key concepts

• Gestalt principles:

  • group patterns, simplify perception

• Brain uses internal models + probabilistic reasoning

• Filters information → selects what to process

• Limited cognitive resources → prioritization needed

• Flexible & goal-driven

• Perceptual → focus on features (color, motion)

• Cognitive → mental tasks (memory, problem-solving)

• Executive → task management & priorities

• Social → interaction with others

1. Spotlight model:

   • attention highlights one area

2. Biased competition:

   • stimuli compete → most relevant wins

Temporary mental workspace

• Supports:

  • reasoning, planning, language

Components

• Phonological loop:

  • verbal/auditory info

• Visuospatial sketchpad:

  • images, spatial info

• Central executive:

  • controls attention & resources

• Sensory input → sensory register (very brief)

• → attention filters information

• → working memory processes it

• → long-term memory stores it

• Controlled:

  • slow, conscious, effortful

  • e.g. learning new task

• Automatic:

  • fast, unconscious

  • e.g. routine skills

• Cognitive models → inspire AI systems:

  • perception (computer vision)

  • attention (transformers)

  • memory & reasoning models

• Combine:

  • symbolic reasoning + data-driven learning

• Key cognitive ability:

  • communication, knowledge sharing

Components

• Syntax → structure

• Semantics → meaning

• Pragmatics → context

Characteristics

• Incremental processing (word by word)

• Context-dependent interpretation

• Well-defined → clear rules (e.g. puzzle)

• Ill-defined → unclear goals (e.g. life decisions)

• Trial & error → learn from failure

• Algorithms → step-by-step rules

• Heuristics → mental shortcuts

• Insight → sudden solution

• Functional fixedness:

  • difficulty seeing new uses for objects

• Decisions affected by:

  • emotions

  • time pressure

  • limited information

• System 1 (intuitive):

  • fast, automatic

• System 2 (analytical):

  • slow, deliberate

➡️ Humans are not fully rational

• Intelligence = interaction of:

  • perception, learning, memory, reasoning, action

• Neuroscience → biological mechanisms

• Cognitive science → mental processes

• AI → implementation in machines

➡️ Strong interdisciplinary connection ➡️ Human intelligence → blueprint for AI

• Brain research → inspired artificial neural networks

• AI systems use:

  • attention

  • memory

  • learning from experience

• Goal:

  • build adaptive, flexible systems

• Aim:

  • not replicate brain exactly

  • extract key principles

• Approach:

  • abstraction → simplify biology into models

• First mathematical neuron model (1940s)

• Key idea:

  • neuron = binary unit (on/off)

Function

• Inputs → weighted sum → threshold → output (0 or 1)

• Can implement:

  • AND, OR, NOT logic

➡️ Foundation of neural networks

• Many simple neurons → complex networks

• Enable:

  • logical reasoning

  • computation

• Improved artificial neuron

Key features

• Adjustable weights → learning from data

• Bias term → shifts threshold

• Still binary output

➡️ Can:

• recognize patterns

• adapt based on experience

➡️ Inspired by synaptic plasticity

Knowledge = symbols + rules

• Strengths:

  • logical reasoning

  • interpretability

• Weaknesses:

  • rigid

  • poor with uncertainty

  • no learning

• Knowledge = distributed in weights

• Learns from data

Strengths:

• pattern recognition

• handles noise & ambiguity

• adaptable

Weaknesses:

• low interpretability (“black box”)

• Combines both approaches:

  • symbolic reasoning + neural learning

• Example:

  • perception (NN) + reasoning (rules)

• Intelligence requires a body + interaction

• Learning through:

  • sensory input

  • physical action

In AI:

• agents with:

  • sensors (perception)

  • actuators (actions)

➡️ Example:

• robot learning via trial & error

• Intelligence depends on:

  • environment

  • context

• Not static → dynamic & unpredictable

Key idea

• behavior adapts to:

  • current situation

  • goals

  • constraints

➡️ Example:

• delivery robot navigating obstacles

• Intelligent systems operate in cycle:

  • perceive → interpret → decide → act → learn → adapt

• Full intelligence requires:

  • physical interaction (embodied)

  • structured reasoning (symbolic)

➡️ Modern AI = hybrid systems

• AI is shaped by:

  • neuroscience (biology)

  • cognitive science (thinking)

• Intelligence emerges from:

  • interaction with environment

  • continuous learning & adaptation

• Goal: build systems that:

  • learn from data

  • adapt to new situations

  • improve over time

• Inspired by:

  • neuroscience

  • cognitive science

• Used for:

  • language understanding

  • image recognition

  • forecasting

➡️ Key strength: generalization to unseen data

• Learn from examples, not rules

• Create internal representations of data

• Learning = adjusting weights between neurons

Basic unit: Artificial neuron

• Inputs → weighted sum → activation → output

➡️ Single neuron = limited ➡️ Networks = powerful

1. Input Layer

2. Hidden Layers

3. Output Layer

• Receives raw data

• One neuron per feature (e.g. pixel, value)

• No computation

• Perform computations:

  • weighted sum + bias + activation

• Extract features:

  • early layers → simple patterns

  • deeper layers → complex patterns

• Produces final prediction

• Depends on task:

  • Binary → 1 neuron (e.g. spam detection)

  • Classification → multiple neurons

  • Regression → continuous value

• Data flows:

  • input → hidden → output

• No feedback loops

• Fully connected structure

Learning = iterative improvement using feedback

4 Key Components

1. Forward propagation

2. Loss function

3. Backpropagation

4. Gradient descent

• Input passes through network → prediction

• Uses current weights (initially random)

• Also called inference phase

Measures prediction error

Types

• MSE (Mean Squared Error):

  • for continuous values

• Cross-Entropy:

  • for classification

➡️ Output = single error value

• Computes how each parameter contributed to error

• Uses:

  • derivatives

  • chain rule

• Propagates error backwards through network

➡️ Determines how to adjust weights

• Updates weights to reduce error

Key ideas

• Move downhill in error landscape

• Use gradient (direction of steepest increase)

• Update rule:

  • adjust weights in opposite direction

Learning rate (λ)

• Small → slow but stable

• Large → fast but risky

• Repeat:

  • predict → compute error → adjust weights

• Gradual improvement over time

• Many hidden layers

• Enable:

  • complex pattern recognition

  • hierarchical feature learning

Examples of tasks

• face recognition

• speech translation

• text understanding

• autonomous driving

Examples:

• AlexNet → improved performance with depth

• VGG → deeper structure

• GoogLeNet → efficient architecture

• ResNet → very deep networks (100+ layers)x

• Virtual assistants

• Translation & NLP

• Robotics & autonomous vehicles

• Healthcare (diagnosis support)

• Finance (fraud detection)

• Recommendation systems

• Require:

  • large datasets

  • high computational power

• Long training time

• Low interpretability (“black box”)

• Hard to explain decisions

• Adversarial vulnerability:

  • small input changes → wrong outputs

• Lack of true understanding

• Weak in:

  • symbolic reasoning

Strengths

• pattern recognition

• scalability

• adaptability

Limitations

• opaque decisions

• data & resource intensive