Machines/software perform tasks that typically require human intelligence

Focus on perception, reasoning, learning, language understanding, decision-making

Map inputs (data) to outputs (actions/answers) in complex environments

Improve through experience (learning) vs fixed rules only

Traditional programming

AI (especially Machine Learning)

Learns from data to make predictions/decisions

Common tasks

Subset of ML using multi-layer neural networks

Strengths

Typical models

Understanding and generating human language

Capabilities

Interpreting images and video

Tasks

Connect perception to physical action

Components

Represent facts, rules, relationships

Approaches

Uses

Find sequences of actions to reach goals

Methods

Learn via trial-and-error with rewards

Applications

Specialized in a specific task/domain

Examples

Human-level capability across many tasks (not achieved today)

Broad transfer of learning and reasoning across domains

Beyond human intelligence (hypothetical)

Major governance and safety considerations

Sources

Quality requirements

Function approximator mapping inputs to outputs

Choice depends on

Learn parameters to minimize errors

Paradigms

Metrics

Validation practices

Serve predictions in real time or batch

Requirements

Monitor drift and performance decay

Retrain with updated data and new labels

Vision and audio/speech

Comprehension, dialogue, summarization

Handle uncertainty, trade-offs

Combine evidence from multiple sources

Improve from data and experience

Generalize to new examples

Create text, images, audio, code

Use cases

Explicit expert-crafted rules

Pros

Cons

Learn patterns from data

Pros

Cons

Combine symbolic rules with ML

Benefits

Search ranking and retrieval

Recommendations

Spam/fraud detection

Assistants/chatbots

Content moderation/personalization

Imaging diagnostics support

Clinical decision support

Drug discovery/protein modeling

Triage/virtual nursing assistants

Operational optimization

Credit scoring/underwriting

Fraud detection and AML

Trading and risk modeling

Claims automation/document processing

Predictive maintenance

Vision quality inspection

Demand forecasting/inventory optimization

Warehouse automation/routing

Segmentation/churn prediction

Dynamic pricing/promo optimization

Personalized campaigns/A-B automation

Visual search/virtual try-on

Route optimization/fleet management

ADAS/autonomy research

Last-mile delivery optimization

Traffic prediction/smart infrastructure

Personalized learning pathways

Automated grading/feedback (with safeguards)

Tutoring/content generation

Accessibility tools

Drone/satellite crop monitoring

Precision farming

Pest/disease detection

Yield prediction

Load forecasting/grid optimization

Fault detection

Renewable prediction

Building energy management

Network anomaly detection

Malware/phishing detection

Incident triage/response support

Document processing/citizen services

Resource allocation/fraud detection

Disaster response modeling

Smart city analytics

Generation/editing for text-image-video

Dubbing/voice synthesis/subtitling

Game AI/procedural content

Design assistance

Code completion/refactoring

Bug detection/test generation

Log analysis/incident summarization

DevOps automation/observability insights

Bias and poor representativeness

Privacy/consent/security risks

Labeling cost and quality

Fail outside training distribution

Hallucinations in generative models

Weak causal reasoning/common-sense constraints

Black-box behavior, limited explainability

Regulatory and trust concerns in high-stakes decisions

Sensitivity to adversarial inputs

Drift over time from changing data

Compute, energy use, deployment complexity

Overreliance/automation bias

Skills gap, organizational change management

Audit disparate impact

Balanced datasets, fairness-aware modeling

Data minimization/anonymization/secure storage

Federated learning and differential privacy (where applicable)

Model documentation (datasheets, model cards)

Traceability, audit logs

Protect models/data from theft/tampering

Defend against prompt injection/adversarial attacks

Human-in-the-loop for high-risk tasks

Clear acceptable-use policies for generative outputs

Sector compliance (health, finance)

Emerging AI regulations/standards

Define objective, constraints, success metrics

Decide if AI is appropriate vs simpler methods

Collection, governance, labeling, privacy

Baselines, feature engineering (if needed), model selection

Training, tuning, validation

Integrate into products/workflows

Latency, scalability, monitoring, fallback behavior

CI/CD for models, monitoring, drift detection, retraining

Version control for data and models

A/B testing, user feedback loops

Continuous risk assessment and improvement