How to distinguish types of AI?
Before we start working with Artificial Intelligence we should know what it is, at least a bit. There are many definitions of AI and there are multiple types of AI. Prepare a mug of coffee and let’s start.
What is AI?
There are many definitions of AI and many rules and to be honest any one can say his own definition, but there is one generic sentence we can apply for it to say what the Artificial Intelligence is.
Artificial Intelligence is a computer program that can solve problems without strict instructions how to do it.
It is really good definition, because it can be adapted to every of AI definitions, rules or properties. What follows from this?
If a system can solve a problem without instructions it needs to have an ability to learn. By learning:
- From data, when you push data to the system and it is able to consume them and give us correct answers. You can push the data in the beginning or during the system usage.
- From experience, when you give the system a feedback after its response. Then the system can adjust its behavior for the next times.
To meet the definition it needs to change its behavior during time. So there should be a feedback loop to allow the system to do it.
Weak AI vs Strong AI
The first one is to distinguish between weak and strong AI. It is not very practical division, but it is more about ideas and concepts.
Weak AI
- It is a system that can solve only one problem like:
- Play chess, checker or go
- Generate images
- Drive a car
- Recognize speech
- It is the group that actually exists and are more and more popular.
- We can imagine them as islands that can solve one problem and are not connected with each other.
- It is a system that can solve only one problem like:
Strong AI
- It is a system that can solve multiple problems and can combine knowledge from different areas and it can find not obvious interactions between them.
- It is rather theoretical concept of system that can do anything or can work as a real brain.
Super AI
- It is a next level of AI that is more intelligent than humans.
- When we are thinking about terminator or matrix we are thinking about strong AI.
- It is a interesting topic to discuss, because it is thr type when we hear about dangers of AI.
Deep Learning vs Reinforcement Learning
The second one is more practical and it is about how the system is learning. We can compare it to learning with teacher or without.
Deep Learning
- It is a way of learning AI by giving it a lot of data as in input and expected results as in output.
- It like learning with teacher. We teach it what it the correct answer for the given input.
- It is the main way of learning AI and it is the most popular one.
- It is more predictable with result and time of training, because we have limited input data.
Reinforcement Learning
It is a way of learning AI by giving it some simple instructions and when we have a result we give it a feedback if it is correct or not.
It is like learning without teacher.
- It is like child is learning to walk. We are not giving him instructions how to do it, but it take the feedback. When it falls, then it hurts, then it knows that it is not correct.
- We do the same with AI. When the answer is correct we give it a reward, when it is not correct we give it a punishment.
To learn it needs to have a huge amount of tries, because it cannot think. It just try and try and try with some random changes.
When you see an AI that tries to solve a maze or FlappyBird it is a good example of reinforcement learning.
Machine Learning vs Deep Learning
The third one is not so obvious, because Deep Learning is a subset of Machine Learning.
Machine learning
- It rather a group of algorithms that can learn from data.
- It is a simpler way of learning then Deep Learning.
- It is about finding a function that can map input data to output data.
Deep Learning
- It is more advanced way of learning then Machine Learning.
- It uses neural networks to learn.
- It is about finding more complex relations between input and output data.
- It requires more data to learn.
Summary
As you can see there are many ways to divide AI. You can be sure that that post will be updated in the future when I have more knowledge about AI.
Would you like to add any other division? Let me know in the comments.