AI in Education

What is AI (Artificial Intelligence)?

• Turing’s test
• Strong AI
• Weak AI
• Machine Learning

The term “AI”

• Is vaguely defined. It can have many meanings.
• Is a popular buzzword.
• Is often overused to describe technologies that are not exactly AI.
• Is (IMO) overhyped.

Turing’s test

• Can a computer pass as a human via remote (terminal) interaction?
• Being human \(\ne\) being intelligent!

Strong AI

• General intelligence.
• Not mission specific.
• Do what we (humans) can do.

Weak AI

• Mission-specific intelligence.
• Chess-bot, Roomba, route finding, etc.
  • Our WayFinder was once considered AI!
    • It uses Dijkstra’s algorithm, it is now a 3-hour lecture topic.
• Algorithm-based.

Machine learning

• A general form of modeling
• Artificial neural network
• Random forest
• etc.

Does AI depend on machine learning?

• Most weak AI technologies do not depend on machine learning.
• Generally, AI does not depend on machine learning.
• In a sense, machine learning is a way to imitate.
• Neural network is a machine learning mechanism.
• It is difficult to “extract” what a neural network has learned.

Machine learning and modeling

• A “model” in this context is an emulator.
• A model (in this context) does not need to know the “why”, just the “how”.
• Machine learning is an “automatic” way to create a model.
• Not exactly automatic because neural network machine learning is a black art!
• The general approach of using neural network is to throw resources at the problem.

What is GPT?

GPT

• Generative Pre-trained Transformer.
• Generative: not to recognize/classify, but to “output content.”
• Pre-trained: not learning on-the-fly.
• Transformer: a technology/technique to output content based on a prompt.

ChatGPT

• Text-based GPT
• Trained on all text material freely accessible on the Internet
• Complex neural-net approach
• Learns patterns, patterns of patterns, etc. (multilayer neural network)
• Has an “attention” mechanism to maintain context and focus of a conversation
• Has the ToM (theory of mind) of a 9-year-old human

More about Large Language Model (LLM)

• Learn by training.
• Patterns of what follows what.
• Ability to “abstract” and find patterns of patterns.
• Learns grammar on-the-fly based on presented examples.
• No intrinsic reasoning/logic capabilities.
• Good example of garbage-in-garbage-out if training samples are not selected or filtered.

Education

• What is education?
  • Objectives
  • How it is done
  • The role of educators

The objectives of education

• From a student’s perspective:
  • To acquire knowledge and skills.
  • Skills include:
    • Reading/writing,
    • Critical thinking, and
    • Problem-solving.

How education is done

• Classes:
  • Content,
  • Assessment,
  • Instruction, and
  • Interaction with an instructor.

The role of an educator

• This is Tak’s subjective opinion:
• To acquire, curate, and structure learning material.
• To present learning material.
• To interact and address questions from learners.
• To assess learning objectives.

Student Learning Outcomes

• What do students need to learn?
  • Knowledge?
  • Skills?

Slide rule vs. spreadsheets

• Is this comparison applicable?
• The effective use of ChatGPT relies on the prompt.
• The prompt seeds the “transformer” to generate output.
• AI tools are powerful, but there are skills required to use AI tools effectively.

Knowledge

• Even with search engines, knowledge by itself becomes less important as a learning objective.
• Few employers expect employees to be able to recall specific knowledge.
• Instead, employers value the application of knowledge to solve problems.

Skills for the post ChatGPT era

• A continually changing list of skills.
• “What AI cannot do, yet.”
  • Recalling knowledge.
  • Analyze a scenario.
  • Apply reasoning.
  • Critical thinking.
  • Troubleshooting.
  • (Ironically) pure logical thinking.
  • (Also ironically) out-of-the-box thinking.

Critical thinking

• What is critical thinking?
• ChatGPT limitations
  • Can only compare-and-contrast based on existing literature that compared-and-contrasted.
• Critical thinking is a form of out-of-the-box thinking: what if the current theory is wrong?
  • Is there evidence that the current theory may be wrong?
  • What is a better theory to propose?

Troubleshooting

• What is troubleshooting?
• ChatGPT limitations
  • Can only troubleshoot symptoms with known troubleshooting process.
• Human troubleshooting process is not entirely at a conscious level.
  • ChatGPT is trained only based on written text, there is no data to model the subconscious processes.

Pure logical thinking

• ChatGPT is probabilistic, there is no intrinsic knowledge of logic.

Out-of-the-box thinking 1

• The ability to “borrow” concepts from an entirely different domain.
• Examples:
  • The use of analogies, personification, etc.

Out-of-the-box thinking 2

• The ability to “question” assumptions. (Beyond critical thinking.)
• Examples:
  • Einstein’s questioning of Newtonian physics, and the new assumption of the speed of light being constant, warp time and space, instead.
  • Planck’s questioning of Newtonian physics, and the new assumption of the quantized nature of physics.

Open discussion

• How do we update our curricula?
• Courses not intended for transfer.
• Courses for transfer.

Using AI and cheating

What are we really testing?

• Does what we test match the learning outcomes?
• Do the learning outcomes match what employers expect?

What do employers expect (post ChatGPT) from human employees?

• The use of AI as a tool to:
  • Improve efficiency,
  • Improve accuracy (?),
  • Solve problems (?).

What do we teach our students?

• Prohibit the use of AI tools:
  • The students will not have the skills to properly leverage AI tools.
• Allow the use of AI tools:
  • Then what are we teaching and assessing?

Employers expect efficiency

• Students should know how to use AI tools when they are applicable.

Employers also expect correctness

• Students should be able to evaluate AI-generated results.
• Students should still know the fundamental concepts and solve problems “by-hand”.