These days the term AI is used so often, I wonder if it has any meaning anymore.
I completed my PhD at Stanford University in AI applications for construction, became an Adjunct Professor in 2019 and currently still teach AI Construction at the engineering school. A decade ago I founded ALICE Technologies which has developed the world’s first AI Generative Construction Simulator.
I love AI – I think it’s the “bees knees” and believe it has changed the world for the better and will continue to do so. Having worked at the cutting edge of AI in Silicon Valley over the last 16 years, and interacted with many others that have done so, I want to share some of what I know about AI, what excites me about it, and where I think it's headed.
This seven part series is targeted towards people who don’t know much about this technology but want to be dangerous at a cocktail party. This is not targeted towards people with advanced degrees in artificial intelligence and computer science, though comments are welcome. (FYI, this series was not written by AI.)
We’ll cover:
-
What is AI?
-
Symbolic AI
-
Neural AI (or LLM AI)
-
Large Language Models
-
Where to use LLMs?
-
What are LLMs good for if they hallucinate?
-
The Future of AI (Neuro-symbolic AI & World Models)
Let’s start with “What is AI?”
Currently, the term AI is broad and undefined.
John McCarthy, one of the founding fathers of AI coined the term “artificial intelligence” back in 1956 and he defined it as:
“The ability of machines to solve human-like problems.”
Other definitions are technique-based: “is it an LLM or Neural Network,” or self learning-based “does it improve itself?”
The definition that several of my computer science friends use is “Artificial Intelligence are complex algorithms and techniques that solve human-like problems." The public sometimes refers to it as “whatever technology couldn’t do 3 years ago.” What was referred to as “vision algorithms”(algorithms that could recognize objects) a few years ago, started to be called “machine learning” and then “artificial intelligence.” So terminology morphs.
All of these definitions work, but at the end they’re all just algorithms, each having its own advantages and disadvantages.
The field is about 70 years old and currently, has roughly 250,000 people building and researching cutting edge AI, including some of our species' smartest people so as you can imagine it is vast and contains multiple disciplines.
Here are some subsets of complex algorithms and techniques that solve human -like problems” “complex systems and disciplines used to solve human problems” or what was at various points called AI (and we will also touch upon what might be called AI in the near future):
All industrial grade AI solutions are “composite AI” - systems that combine several of these techniques to create systems that work in complex reality.
Historically, AI has been used to solve very deterministic problems. Problems that are mathematical in nature, require a specific methodology to solve and produce the same answer every time you run the algorithm. Problems like calculating the trajectory of a rocket in orbit as its fuel was being burned. Or the structural loading in a complex structure.
Until about half a decade or so ago, AI was terrible at solving pattern based problems. Problems like correctly identifying what was in an image. Or telling you what the recurring theme was across a set of texts or paragraphs.
These types of problems require different types of algorithms and for the purpose of this series we will separate AI into two broad main categories that solve those two families of problems: Symbolic AI (rules-based) and Neural AI (patterns-based).
We will be explaining both and then talking about what is beyond these systems. This is by no means the only or correct way to split the broad field of #AI, but given recent advances in #LLMs it will provide an exciting way to analyze what’s happening.
