AI and Expertise: Judgement, Learning, and Work in the Age of Large Language Models
Dominik Lukeš (dominiklukes.net)
About me
About me
My work
Dominik Lukeš dominiklukes.net
AI Competency Centre University of Oxford
AI Trends Tracking
What I used to do
Reading and Writing Innovation Lab
What I really do
Applied epistemology
What I teach people
How to become better at learning
What happens to knowledge and expertise with Large Language Models in the world?
What happens to knowledge and expertise with Large Language Models in the world?
ChatGPT Timeline
30 Nov 2022
- ChatGPT is released
7 Dec 2022
- 1 million people have used ChatGPT
2023-2024
- 100s of millions people use ChatGPT to code, write, translate, learn and ... cheat
Sept 2025
- 1 billion people use ChatGPT every week
- 2026AI agents happen
What happens to knowledge and expertise with Large Language Models in the world?
What is ChatGPT?
- outputs tokens
- provides rich knowledge and judgment
- limited in precision and consistency
- parses tokens to run them
- provides absolute logic
What happens to knowledge and expertise with Large Language Models in the world?
Large Language Model is a Universal translator
- language-to-language
- style-to-style
- unstructured-to-structured / structured-to-unstructured
- text-to-code / code-to-text
- image-to-text / text-to-image
- question-to-answer / answer-to-question
- problem-to-plan
What happens to knowledge and expertise with Large Language Models in the world?
Sometimes LLMs are a literal translator
What happens to knowledge and expertise with Large Language Models in the world?
Orchestration
- parses formatting
- interprets commands
- passes commands to computer
- sends outputs from computer to model
What happens to knowledge and expertise with Large Language Models in the world?
Two parts of intelligence
- recognise objects
- recall ideas
- mental manipulation
- thinking about problems
- keeping mental lists
- look up facts
- creating budgets
- asking questions
What happens to knowledge and expertise with Large Language Models in the world?
Strengths of cognition and tools
Cognition
Judgement
Knowledge
Comparison
Estimation
Tools
Precision
Exact retrieval
Complex conditional
Precise calculation
What happens to knowledge and expertise with Large Language Models in the world?
Intelligence as Grep and Grok
What happens to knowledge and expertise with Large Language Models in the world?
What do humans share with Large Language Models
What happens to knowledge and expertise with Large Language Models in the world?
What tools do we use
What happens to knowledge and expertise with Large Language Models in the world?
What tools do we use
Humans
- Mental checklists
- Focus
- Body parts
- Physical objects
- Computers
- Relationships
Large Language Model
- Shell commands
- CLIs
- APIs
- Scripts
- Search
What happens to knowledge and expertise with Large Language Models in the world?
What can artificial "intelligence" do with tools today
AI agents are on everyone's lips, what does it mean for us?
- Vivian Balakrishnan, Foreign Minister of Singapore
Al agents have crossed a threshold I didn't expect so soon. Not just impressive demos - but practical tools for daily use.
What happens to knowledge and expertise with Large Language Models in the world?
What does it mean for humans?
- Work is amplified
- More people get access to that work
- No job for someone to do the work
What happens to knowledge and expertise with Large Language Models in the world?
Is this true?
Access to an LLM does not replace:
- Domain expertise - Architectural experience - Understanding of trade-offs - Awareness of what is not written in the documents
AI democratizes capability. It does not democratize judgment.
- Alex LeBlanc (AI Amplifies Expertise, Not Replaces It \| LinkedIn)
What happens to knowledge and expertise with Large Language Models in the world?
Why are you studying at university?
- Change your brain
- Change the world
- Prove to the authorities you can pass an exam
What happens to knowledge and expertise with Large Language Models in the world?
There's a hidden conflict
Changing your brain to change the world is in conflict with changing your brain to pass exams.
What happens to knowledge and expertise with Large Language Models in the world?
Can LLMs help us resolve this conflict?
What happens to knowledge and expertise with Large Language Models in the world?
True fact about learning
The amount of time and effort required to change your brain so that you can do something is constant.
What happens to knowledge and expertise with Large Language Models in the world?
Nobody can change your brain for you
What happens to knowledge and expertise with Large Language Models in the world?
Effort spent on building pyramids
What happens to knowledge and expertise with Large Language Models in the world?
Effort spent on learning
What happens to knowledge and expertise with Large Language Models in the world?
Role of LLMs in learning
What LLMs can do for you
Use AI as a ramp to higher learning.
What happens to knowledge and expertise with Large Language Models in the world?
What LLMs can't do for you
Learning
No muscle this way
What happens to knowledge and expertise with Large Language Models in the world?
What LLMs cannot do at all
Learn from experience
What a new expertise can look like
What a new expertise can look like
What was the last time you used ChatGPT?
What a new expertise can look like
I haven't ChatGPT regularly in months and not at all last week.
What a new expertise can look like · I have been using Codex every day
I have been using Codex every day
What a new expertise can look like
I have been using Codex every day
What a new expertise can look like
What this means
- I am still using AI
- I am using the same model power in a different context
- {icon=trending-up} I can get much more done
What a new expertise can look like
With Codex and Claude Code I have
- Scheduled my calendar to come here
- Built my own slide presentation tools
- Organised old file archives
- Rescued old websites and built new ones
- Build my own apps for
- Work out tracking
- Photo management
- Searching my highlights
What did I need to know to do this?
What did I need to know to do this? · The knowledge needed today
The knowledge needed today
What did I need to know to do this?
The knowledge needed today
Knowledge you don't need
- Complex algorithms
- Syntax of computer languages
- Variables, functions, etc.
- How to use developer tools
Knowledge you do need
- How your computer works
- Principles of software architecture
- Developer tools landscape
- How AI agents work
- What is possible with software
What did I need to know to do this?
Some useful things to know
- File system details
- Plain text / Markdown vs Rich Text
- CLI tools
- Pipes and deamons
- Libraries
- Git and GitHub
- What an API can do for you
- Hosting (Cloudflare)
- When you need a database
- Refactoring
- Content vs display
- Security concerns
- Model v harness
- Model tool use
- Model capabilities
What did I need to know to do this?
Big lesson from the agent revolution
It is now worth learning more about computers at every level because you can now do something about it.
What did I need to know to do this?
Yes, you do need judgment
What did I need to know to do this?
What kind of judgment do you need?
- What something is
- Something may be a problem
- How to backtrack through a problem
- How to evaluate a plan step by step
What did I need to know to do this?
Problem with "critical thinking"
Good slow judgment needs good fast judgment.
What did I need to know to do this?
What is fast judgment like?
- Fluency in language
- Participation in a game
- Playing a sport
What did I need to know to do this?
You cannot act in a world and look up everything. You must know things in a special way.
What did I need to know to do this?
You cannot just have a list of facts and rules for
- speaking a language
- participating in games
- being active in sports
- playing music
What did I need to know to do this?
You cannot
- look up words and speak
- look up rules and play
- follow manual and play
What did I need to know to do this?
You must have knowledge that can only be developed through
What did I need to know to do this?
You cannot evaluate claims and look up every concept or fact. You must know these in a special way.
What did I need to know to do this?
How do you develop good fast judgment?
Future
Future
Why are you at university
- Knowledge and skill
- Relationships
- Opening doors
Future
What is the impact of AI on the system?
- Knowledge is less valuable
- What is behind the door is different
- Some doors disappear
Future
What is the impact of AI on you?
- You can learn more
- You can do more once you're through the door
- You can make new doors
Future
Will AI make learning easier?
No. You now must learn more, work harder, think harder. There will be less pseudo learning. There are no excuses for not learning what you need to know.
Future
Will AI replace your need to learn?
Future
Lessons from chess
- Machines are better than best humans
- More people play chess than ever
- Best people practice with chess machines
Future
Why are people getting better at
- Musical instruments
- Sport
- Coding
Future
New world
- More variety of learning sources
- Easier access to instruction
- Easier access to expert thinking
Future
Finally
- You live in the golden age of learning.
- It is now worth learning more not less.
- Learn because you can
- Learn with and about AI
- Don't give up
Future
You must
- Use all the tools to change world.
- You have some of the most powerful tools available.
- Do not waste them!
Thank you
Dominik Lukeš (dominiklukes.net)