🏗 AI Basic Concepts
There are a few basic concepts you need to get started in AI. Weak vs. Strong AI, and why are developing strong AI. The importance of Data, and high quality data. And the learning and inference phases of AI.
Weak vs. Strong AI
There are two ends of the AI spectrum; weak and strong. Weak AI fits a very limited use case, it generally solves one task. Strong AI is meant to be a mind that cannot be distinguished from a human. We think about strong AI as general Artificial Intelligence.
Strong AI creates generalized problem solving. Imagine you ask a chat bot any question and they respond with a correct answer. This would solve many problems we face on a day to day basis.
Data
AI finds patterns in data and applies those patterns to a problem. The big shift here is that we can now generalize our problem solving. We used to program a tool to solve one single problem, now we can teach a computer to solve a class of problems. This makes high quality data the most important part of an AI system.
The biggest bottleneck in AI is a lack of large scale, high quality data sets. AI doesn't work without good data The classic motto is Garbage in, Garbage out. Bad data creates bad predictions.
Garbage in, Garbage out
Computers cannot reason for themselves. They learn exactly what they are told by the data. Bad inputs lead to bad outputs.
How AI solves problems
AI solves problems in 2 main steps; training and inference.
Training
Training takes a huge data set and packages it into neat mathematical objects called models.
Our world is so rich with data, and we cannot capture all of it. This is like trying to paint a picture without mixing colors. You have to be very precise with how you teach the computer to problem solve
Training simplifies the real world into a mathematical model the computer can digest
Inference
Inference answers questions about the model created in the training step. It is ultimately the application of the model we created.
As a user of a model, we must be aware that AI does what it's told... not what makes sense. So often times it will take shortcuts we didn't see coming.
Inference answers questions based on the model