AI and Economics: Industry Dynamics

• The content provides an introduction to Artificial Intelligence (AI) and its economic dynamics.
• The course is called: Artificial Intelligence and Economics Dynamics (AIED).
• The course offers an economic perspective on AI.
• The main goal is to grasp the dynamics of breakthrough AI tech and evaluate if existing economics capture how AI is adopted and impacts the economy.
• The course emphasizes the dynamics of AI as a system-with many parts evolving in an interdependent manner.

Learning Outcomes

• Explain key aspects of AI as a technology
• Demonstrate critical awareness of current issues in AI development
• Understand how economic models explain AI industry dynamics

AI Course Programme:

• Lecture 1 to 2: What is AI? And how new is it really?
• Lecture 3: The emergence of the Al industry
• Lecture 4 to 5: Basic economics of digital technology and Al
• Lecture 6 to 7: Model #1: replicator dynamics & Al
• Lecture 8 to 9: Model #2: industry life cycles & Al
• Lecture 10: The course recaps AI competition and industrial policy issues.

General Introduction to AI

• The course addresses whether AI is really a new phenomenon.
• AI is defined as a class/collection of software technologies that execute tasks linked to human intelligence like vision, speech pattern recognition, and classification.
• AI's functionality is grounded in data processing and enabled by software/algorithms like artificial neural networks.
• AI is a diverse field, covering maths, computer science, and philosophy, its subfields include: AI (narrow/weak) vs AGI (Artificial General Intelligence) vs ASI (Artificial Super Intelligence, or strong).
• AI can be defined as a 'suitcase word'.
• "in silico brain" describes embedding natural intelligence mechanisms into computers.
• Another definition, emulation, focuses on AI performing tasks as well or better like a human, without needing to focus on how this works.

Perspectives on approaches to AI:

• Symbolic systems: AI that uses rule based instruction
• Subsymbolic AI: AI that "learns" through algorithms, where today's generative AI is an instance of this
• Machine learning: collection of AI tech that allows AI systems to learn and operate
• Robots: AI isn't strictly needed for robots

Other Intelligence

• Humans naturally give consciousness to anything from each other to inanimate objects
• Cognition needs both empathy and social needs, which AI is missing

Defining AI as a System Technology:

• AI systems are specialized and emulate human intelligence.
• AI is a system technology
• The key parts of an AI system combine to make "talent": AI = LA + HDD + CP + DS + talent.
• LA (Learning Algorithm): AI Software, Deep Learning / LLMs
• HDD (High-Dimensional Data): big data
• CP (Computing Power): hardware
• DS (Domain Structure): multi mode

How AI Works

• AI works with statistical AI methods (see Mitchell 2019 on Perceptrons)
• Artificial Neural Networks (ANNs) have instructions organized as a network
• AI learning occurs in training, with memory encoded in connections
• Once shaped, can be used to extrapolate insights from new inputs
• A Deep Learning technique automates structure updating

GenAI and Large Language Models

• The latest in AI are GenAI (producing content) called Large Language Models (LLMs) known as Foundation models.
• LLMs have emerged as an industry standard
• LLMs are models of language as a probability distribution over a natural language.
• Criteria for LLMs are that the models perform three tasks:
• Can model text to generate content.
• Use large-scale pretraining.
• Make inferences with transfer learning.
• LLMs can be thought of as a technology (engine) and products (engine and interface) creating commoditization of AI.