Very Basic Artificial Intelligence and Machine Learning
Reading Time: 4 minutes
This article will hopefully eradicate some of the mysteries behind Artificial Intelligence (AI). Note that AI is an extremely advanced topic so this article is aimed at beginners.
Basic AI:
AI is such a fascinating and exciting topic that makes it one of the most mysterious fields. We've all asked the same questions to ourselves: Will AI transcend humans? Will AI replace all our jobs? Will AI destroy humans?
Before diving into the details, one thing must be noted: Computer intelligence and human intelligence act in very different ways. Computers are much better than humans in identifying and matching different patterns. But Humans are more creative and have more common sense (or that is what I hope at least).
A game is a perfect environment for AI. It has a set of rules and a set of possibilities. When Deep Blue by IBM played chess, it had set rules and possibilities and used pattern matching to check its database for a possible countermove. A computer would play out many thousands of scenarios before its opponent even reached across the board which is why it managed to beat the chess world champion.
If you are wondering about the role of AI in your industry, does your work have a lot of pattern matching, does it have set rules and possibilities? That is why AI is extremely useful in areas such as Trading, Health, Transportation, Education, etc.
Symbols were key in creation of AI. When you see the stop sign symbol, you know that this means stop. When you see the letter 'A' symbol, you know that letter will make a certain sound. When you see a sandwich, you might think of eating. Scientists argued that if a machine was trained to understand these symbols, it could behave more like humans. Scientists thought a key part of human reasoning was simply connecting these different symbols.
The computer does not understand the meaning or the content. It is just matching symbols from a long list of instructions.
When you ask Siri how is she doing, she might say fine. But that doesn't mean that she is really fine. In reality, she doesn't even know what you're asking, she is just matching your question to a programmed response.
In Google maps, you'll put your end location and the system will step through all possibilities in its Database (all roads, street names, etc.) to find you the best route. There is also what we call Planning AI system, which means that the AI has some common sense by trying to limit what data to look for: For Google Maps, it doesn't make sense for Google to look through every possible road and street name in the world if you are wandering around London, so it limits what patterns the program has to match at any one time.
Machine Learning:
Symbolic matching worked well for machines who had limited matching patterns. The challenge is when the list is very long. Let's say you want to create a program that identifies animals from a database of images. You need to create a sizes list, eye shapes list, number of legs list, and on and on. "If it has 4 legs, check if it has fur, if it has fur, check the shape of its ears". A list like that would be too long. There are way too many opportunities where the system would get stuck.
AI researchers then started to wonder if instead we planned out the matching patterns, a computer would be programmed to learn the new patterns by itself. This is machine learning. A machine capable of learning by itself.
These machines still match symbols and identifies patterns like in Siri for example. But instead of having an expert list every phase, you have the machine learning the patterns by looking at the data. So after many conversations with Siri, it may learn that conversations always start with similar greetings and responses, so it would create its own list. Again, it is about pattern recognition.
Artificial Neural Networks: Machine learning got a big boost from Artificial Neural Networks. This is a computer program that tries to mimic the structure of the human brain. Your brain is filled with neurons, they connect to each other and send signals as a way to learn and react to the outside world. The neurons will increase the strength of their connections based on your experiences. So if you want to learn how to play football, your brain will create new connections and strengthen existing signals.
The neurons in an ANN are organized into layers. You have a dot that hops down from layer to layer until it reaches the output layer. All the layers in between are known as hidden layers. The ANN could train itself to understand the input (music, images, etc) and then recognize that input when looking through massive amounts of data. It literally learns through trial and error and scientists try to make this neural network more efficient by making the process of learning quicker.
Deep learning: This is when it gets complex i.e. when there are too many hidden layers. Which creates a large gap between input and output, giving the ANN a lot more room to match patterns. One method is clustering, which is pattern matching for patterns. For instance, let's say you want to use deep learning to identify pictures of cats where you could give it a picture of anything and it would tell whether there's a cat in that picture. To train that computer, you need to feed a few million pictures. The computer on its own would start to think about the pictures and categories, it finds certain categories of patterns that help the network identify the image. It might notice that animals have hair, then each time it looked at a new image, the computer could discard anything without hair as it didn't recognize it as animal. That way, the computer could completely focus on identifying the remaining photographs until it is trained well enough to identify a cat.
One final piece of information. There are two ways a computer can essentially learn. Let's say you have an AI program that you feed a song to it and it returns whether that song is Hip Hop, R&B, Jazz, etc. There are two ways you can teach it:
- Supervised learning: You give it a small set of data called training set (several Hip Hop songs for instance), then you tell the program that this type of music is called Hip Hop. Essentially, you are training the machine on how to classify Hip Hop music (same is true for other types of music).
- Unsupervised learning: With that, you feed all the music you can to the program, then you ask the software to create different categories based on what they hear, ending up with a bunch of AI created categories. In unsupervised learning, an AI system may group unsorted information according to similarities and differences even though there are no categories provided.
Again, this is very brief and basic. But I hope this was a good introduction to these dangerous little geniuses.
Hope you enjoyed the read.
