Day 106 in MIT Sloan Fellows Class 2023, AI for business2 "Birth of AI, definition of AI, and emerge of AI"
This course is to learn potential AI roles from various perspectives without any technical deep-dive.
The goal of this course is to help you become an architect to draw a blue print. Not a builder.
The failure of AI comes from typically architecture failures, not engineering failures. Also, we can partially automate engineering, but difficult to automate architecture.
In actual buildings, we have a lot of crucial mistakes in architecture.
In AI field, the most notable example of architecture failure is UK exam prediciton by biased AI in COVID situation because many teatures hesitate to organize exams under covid situation and threw all the responsibilities to algorithm.
The definition of AI
https://www-formal.stanford.edu/jmc/whatisai.pdf
“It is the science and engineering of making
intelligent machines, especially intelligent
computer programs.It is related to the similar task of using computers to understand
human intelligence, but AI does not have to confine itself
to methods that are biologically observable.” John McCarthy(2007)
The beggining of AI
Artificial Intelligence (AI) Coined at Dartmouth | Dartmouth
In 1956, 10 researchers get together and study AI.
- Language
- Abstraction
- Problem-solving
- Self-improvement
And then, the focus of research moved to "gaming". Especially chess because it was specifically good at doing the following tasks.
- Think about the rule you apply
- Program those rules
- Iterate and improve
However, 70-80s, optimism over AI was fading and all the funding was drying up.
There was one quote.
Computers could beat us at chess, but they couldn’t recognize a digit
So how researchers overcome this rule-based AI development?
What makes the recent phenomenon?
Emerge of AI
The true turning point of AI is the moment it replace "rule" with "inference from data".
Then, what is the difference between AI and statistics?
Leo Breiman decribed two cultures in statistical modeling.
- Data modeling culture: Insight from models. Interprettable for humans.
- Algorithmic modeling culture: Hard to interpret and blackbox.
And then, thanks to data explosion and improvement of computing power, algorithmic modeling culture developed modern AI and achieved massive commercial success.
Neurosymbolic AI
This concept is outside this course, but in a nutshell, some people think we need to combine data with rules.
