Machine Learning Techniques - 0

Introduction to Artificial Intelligence and Machine Learning

Julien Fouret
julien.fouret@cpe.fr

Table of Contents

What is Artificial Intelligence ?

Can Machines Think ?1

  • Defining a machine
  • Understanding “thinking”

The Imitation Game

Context:

  • A → Man
  • B → Woman
  • C → Interrogator

Rules:

  • Genders of A and B are hidden to C
  • C interrogates them
  • Responses are typewritten

Objectives:

  • For C :

Determine the Gender of A and B

  • For A : Deceive C

  • For B : Assist C

What if A is a machine ?

Would C be as often wrong ?

A Modern Experiment: The Imitation Game Revisited

Human or Not? A Gamified Approach to the Turing Test2

Online 2-player game:

  • One player asks questions and must guess “Human” or “Bot”
  • The other player responds and might be substituted by a bot
  • Bots are advanced LLM models (e.g., GPT-3)

Turing test: Current outcomes

73% Chance to recognize a real person

60% chance to recognize a bot

Ineffective strategies:

  • Checking grammar, spelling, recent events

Effective strategies:

  • Recognizing known issues and biases
  • Identifying hallucinations

Beyond The Turing test

Alternative text

Pionneered Definitions of AI3

“Artificial Intelligence is the science and engineering of making intelligent machines, especially intelligent computer programs.”

From John McCarty (2007). Stanford University.

Artificial Intelligence is “the science of making machines do things that would require intelligence if done by men.”

from Marvin Minsky (1968). MIT Press.

Alternative definitions

Human-oriented

“We call programs intelligent if they exhibit behaviors that would be regarded intelligent if they were exhibited by human beings.” - Herbert Simon

“AI is the attempt to make computers do what people think computers cannot do.” – Douglas Baker

“AI is the study of how to make computers do things at which, at the moment, people are better.” – Elaine Rich and Kevin Knight

Mathematics-oriented

“AI is the study of techniques for solving exponentially hard problems in polynomial time by exploiting knowledge about the problem domain.” – Elaine Rich

“There is no intelligence in AI. It’s just pure mathematical optimization.” – Julia Luc

Intelligence-oriented

“Physicists ask what kind of place this universe is and seek to characterize its behavior systematically. Biologists ask what it means for a physical system to be living. We in AI wonder what kind of information-processing system can ask such questions.” – Avron Barr and Edward Feigenbaum

Levels of AI

  1. Narrow AI: Specialized in one task. Translation
  2. Broad AI: Multimodel/Multitask. ChatBot
  3. General AI (AGI): Can perform any intellectual task that a human can.
  4. Superintelligent AI: Surpasses human abilities. It’s speculative and doesn’t exist yet.

What is intelligence ?

Theories of intelligence

  • Two factor Theory (Spearmann)
  • Multiple Intelligence (Gardner)
  • Emotional Intelligence (Goleman)
  • Fluid vs. Crystallized Intelligence (Cattell)

What Makes Human Intelligence Unique ?

  • Instinct: Innate behaviour Respiration
  • Memory: The capacity to recall knowledge and events.
  • Learn: The capability to acquire knowledge and skills.
  • Logic and Reason: Ability to apply rules of logic to reach conclusions.
  • Emotional Intelligence: Recognizing and understanding emotions in oneself and others.
  • Abstraction/Concepts: Conceptualizing ideas and making connections between unrelated domains.
  • Creativity: Generating novel ideas and solutions.
  • And so on …

→ Human intelligence is multifaceted.

Machine Intelligence

  • Operates based on algorithms and data.
  • Doesn’t “understand” or “feel” in the way humans do.
  • Can process information faster and more accurately than humans.

Measuring human intelligence

  • IQ Tests
  • Complex and Controversial
  • Educational exams
  • No single definitive method

Measuring Narrow AI

  • Precision / Specificity for classification
  • Regression coefficient for regression
  • Will be extensively discussed later.

Measuring Artificial General intelligence

  • General Tests:
    • AI2 Reasoning Challenge4
    • HellaSwag5
    • MMLU6
    • TruthfulQA7
  • Field-specific Exams:
    • University exams (e.g., Bar exam, Math exams)
    • Math challenges

→ AGI measurement is closer to human intelligence assessment.

What is Machine Learning?

Definition

Machine Learning is a subset of AI where machines can learn from data.

What is a machine ?

  • Logical model
  • Mathematical function
  • Statistical model
  • Rule-based Process

Machine Learning

  • Machine learn through optimization.
    • Numerical optimizations with objective function
    • Reinforcement learning
    • Genetic algorithms

The term optimization is used loosely, i.e. PCA is an optimization to maximize variance.

Major Dates and Periods

  • 1950: Turing’s “Computing Machinery and Intelligence”
  • 1956: Dartmouth Workshop - Birth of AI
  • 1980s: Rise of Expert Systems
  • 1990s: Machine Learning Gains Popularity
  • 2010s: Deep Learning Era Begins
  • 2020s: Emergence of Large Language Models

Conclusion

  • Artificial intelligence is a wide concept with multiple definitions
    • human-oriented
    • mathematic-oriented
    • intelligence-centered
  • Intelligence is a a wide notion when applied to humans.
  • Machine intelligence

Objectives

  • Definition of Machine Learning
  • Mathematics and Statistics
  • Implement and train models
  • Knowledge and best practices
    • Feature engineering
    • Model engineering

Planning

25/10 PM

  • Part 0 Introduction to Artificial Intelligence and Machine Learning : – 1h30
  • Part 1 Mathematical foundations to Modelling and ML – 2h30
    • Theory of probability – 45min
    • Statistical modelling – 30min
    • Model inference – 45min
    • Important definitions – 30min

26/10 AM

  • TP 1 Predict medical insurance costs – 4h
    • Import dataset – 30 min
    • Simple Linear Model – 30 min
    • New feature and interactions – 20 min
    • statmodels package – 20 min
    • In-sample and Out-sample errors – 30 min
    • Scikit-learn pipeline – 1h
    • Anomaly detection – 30 min
    • Conclusion/Discussion – 20 min

08/11 PM

  • Part 2 Base of Machine Learning 1/2 – 2h
    • A: Typology and nomenclatures – 30min
    • B: Evaluation metrics – 1h
    • C: Bottlenecks and Issues – 30min
  • Part 3 Advanced Machine Learning –2h
    • A: Feature engineering – 1h
    • B: Model engineering 1 – 1h

09/11 AM

  • TP 3 – 4h

Acknowledgement

For fruitful discussions and corrections.

  • Felix Geoffroy
  • Thomas Chaverondier
  • Grégory Morel
  • John Samuel

References

1.
Turing, A. M. I.—COMPUTING MACHINERY AND INTELLIGENCE. Mind LIX, 433–460 (1950).
2.
Jannai, D., Meron, A., Lenz, B., Levine, Y. & Shoham, Y. Human or not? A gamified approach to the turing test. ArXiv vol. abs/2305.20010 (2023).
3.
McCarthy, J., Minsky, M., Rochester, N. & Shannon, C. E. A proposal for the dartmouth summer research project on artificial intelligence. (1955).
4.
Clark, P. et al. Think you have solved question answering? Try ARC, the AI2 reasoning challenge. ArXiv abs/1803.05457, (2018).
5.
Zellers, R., Holtzman, A., Bisk, Y., Farhadi, A. & Choi, Y. HellaSwag: Can a machine really finish your sentence? (2019).
6.
Hendrycks, D. et al. Measuring massive multitask language understanding. ArXiv vol. abs/2009.03300 (2020).
7.
Lin, S. C., Hilton, J. & Evans, O. TruthfulQA: Measuring how models mimic human falsehoods. (2021).