AI in Policing and Big Data Insights

AI in Policing

• AI application in policing: Early intervention system (EIS)
• AUC (Area Under the Curve): Represents the model's ability to distinguish between classes, higher AUC indicates better accuracy
• Binary classification problem: Identifying individuals at risk of future criminal activity
• Goal of EIS: To prevent future crime through early intervention and support
• User-centric approach: Focus on the needs and concerns of communities and individuals to ensure safety for all
• Data de-identification: Irrevocably deleting personal information from datasets
• Regulatory framework: General Data Protection Regulation (GDPR)
• Technology for legal information: Internet and online legal databases

Big Data

• United Nations' definition: Big data as a new kind of challenge for human rights
• Concern about personalized algorithms: Potential for manipulation and reinforcement of existing biases
• Surveillance capitalism: The commodification of personal data for profit and control
• Web server identification: Through cookies, IP addresses, and browsing history
• Constraint in old world shopping: Limited information availability and product options
• Soft fraud: Examples include manipulating insurance claims
• Privileged learning in AI: Ability to use sensitive data for model training without directly exposing data to the model
• Explainable AI in fraud detection: Provides transparency and human oversight for algorithm decisions
• Challenge for insurance industry: Adapting to the digital environment and managing risk

Artificial Intelligence (AI)

• Overfitting: Model performs well on training data but poorly on unseen data
• Three V's of big data: Volume, Velocity, and Variety
• Fourth V of big data: Veracity
• Advantage of big data in AI: Increased data availability for training more accurate and sophisticated models
• Turing test: Evaluates a machine's ability to exhibit intelligent behavior indistinguishable from a human
• Generalizability: An AI model's ability to perform well on tasks it hasn't encountered during training
• General AI: AI capable of performing any intellectual task that a human can
• Goal of AI: To build machines that can perform tasks that typically require human intelligence
• Value of big data: Insights and predictions, improved decision making
• Non-standard data: Data collected haphazardly and inconsistently
• Key difference between big data and classical data science: Volume and complexity of data
• Distinguishing line between big data and alternative data: Traditional data sources vs. non-traditional sources (e.g., social media)

Big Data and AI: Applying Technologies

• Key characteristics of big data: Volume, velocity, and variety
• Essential skill set for AI in big data: Data science, machine learning, and AI expertise
• ShotSpotter technology: Acoustic sensors to detect gunshots
• Ralph Clark's change: Shifting focus from selling the technology to providing a service with data insights
• Zettabyte: One followed by 21 zeros
• Technology for accessing large data: Cloud computing
• AI models generating deep fakes: Generative Adversarial Networks (GANs)
• Technology creating hyper-realistic images and videos: Generative AI or Generative Adversarial Networks
• Role of the "test" in AI training: Evaluating model performance on unseen data
• Main challenge of big data: Managing and processing large amounts of data
• Contributing factors to energy consumption in large models: Computation, model size, data storage
• Reason for smaller model architecture: Efficiency and reduced computational burden
• Core feature of transformer models: Attention mechanism to understand relationships within text
• Word2Vec breakthrough: Creating word embeddings to represent words as vectors, capturing semantic meaning
• Limitation of early NLP models: Struggled with complex language understanding
• Governance for AI in public services: Human-centered, ethical, and transparent