Blog: Artificial Intelligence — History and Inner Workings
Artificial Intelligence, something everyone has heard of, yet very few know what it is, how it works or even when it was created. This article will dive into the history, inner workings, and potential future for AI.
The History of AI
The term “Artifical Intelligence” was coined by John McCarthy during an academic conference in 1956. However, this was not the first time where people looked into a computer’s ability to think. At that time everyone knew that a computer could easily follow a set of directions, but the real debate was whether computers could really “understand” something or not. Many people believed that a computer could understand information, while others believed that computers could only follow the rules.
Benchmark for AI
One benchmark for AI is the Turing Test. Developed by Alan Turing in 1950, the test was his interpretation of how a perfect AI machine would work. He believed that if a machine can answer questions like a human, it is an ideal specimen of a functioning and successful AI model. The Turing Test involves one human evaluator and two participants, one being a machine and one being a human. The evaluator would ask questions to both participants and based on the response that they received, the evaluator would decide which participant is a machine and which is a human. The point of the test was not the see whether computers can speak like humans or whether computers can produce correct answers, in reality, the test was designed to see whether computers can answer questions as a human would. The computer would pass the test if the evaluator could not reliably tell the difference between the human and machine participant.
With AI, one constant problem is the “10-year challenge”, this is where companies will promise massive breakthroughs within the next ten years, however, the problem with this is that in the field of AI, expectations often overtake reality in terms of speed. The “10-year challenge” is a problem as we don’t know when the next generation of AI will be available. We expect that we will be able to achieve something within the next ten years, but when it comes time to make it happen sometimes it can take more than twice the expected time. In our current state, we have made progress towards our goal of a self-thinking machine, but we aren’t entirely there yet.
Now many of you probably thinking, “cool there is a test to see whether a specific model of Artificial Intelligence works or not, but what is AI, and how does it work?” When reading anything about Artificial Intelligence, three rather abstruse terms come into mind, “Artifical Intelligence, Machine Learning, and Deep Learning,” but what is the difference between these three, and are they related to each other?
What is Artificial Intelligence?
AI is the most generic field and covers everything and anything that has to do with machines emulating human behavior, or machines being able to learn something.
Machine Learning is a subset within AI, and it deals with machines looking at patterns and algorithms to generate insights which allow computers to predict something based on data given to them. In the purest form, machine learning is merely a machine looking at data, finding a pattern, and when it is given some similar data, it applies the new model to that data to predict the outcome. Machine learning was a huge step forward in the newly emerging world of AI. It allowed programmers to avoid needing to program every single possible case and outcome of something.
Let’s say you are coding a simple Tic-Tac-Toe game. If you aren’t using machine learning, then you will have to tell the machine what to do for every single possible instance of the game. But if you are using machine learning, you simply show the computer different Tic-Tac-Toe games, and it will learn what to do when a certain instance arises.
Machine learning allows us to circumvent the procedure of hard-coding every case and enables us to do minimal work while creating something that has the potential to be wildly accurate and straightforward to use.
The next subset is Deep Learning. It is the most complicated version of artificial intelligence; it is the version of AI that is often portrayed in works of fiction. Deep Learning is a subset of Machine Learning which enables computers to legitimately learn how to do something
How does deep learning work?
The way these networks work is quite straight-forward. You have little nodes (value holders) called neurons. Each small neuron will hold a particular value. Let’s say we are talking about an image which is 100 pixels x 100 pixels. This will mean each pixel corresponds to a single neuron, so in this case, we would have 10,000 neurons. Now each neuron will be parsed by some code, which will find some pattern, create new neurons that exhibit that pattern, and send the neurons over to the second layer for parsing. The second layer will do the same thing, finds a new pattern, creates new neurons and sends it over to another layer of neurons. This keeps going until we get to the last layers of neurons which will parse the data one last time and find the output.
If we look at the image below, we can see that we have the input layer, which in our example would be all the pixels from the image. Then that input layer will send the data into the first hidden layers, which will send the data to the second hidden layer, and finally, it all refers to the output layer.
These are some of the key terms that will often be used in any discussion relating to AI especially deep learning.
- Input Layer — The layer with the first set of neurons
- Output Layer — The layer which holds the output after all the parsing is done
- Hidden Layer — Every little layer in between the input and output layers
What is the future of AI?
The future could hold many surprises for us, in terms of the advancement of artificial intelligence technology. We are currently seeing significant improvements in computers being able to do regular tasks by themselves. Very recently Tesla announced that they had created a computer which will predict the path of the road, even when the road is out of sight, hence, it will enable to the car to drive itself; something that was promised and anticipated for many years.
While Tesla is doing something that will benefit people in their daily life, other communities are trying to do something a lot more dangerous. The militaries or many countries are trying to implement some form of AI, which will be able to “predict” the enemy’s next move, and react accordingly.
Another example of AI that you see nearly everywhere and use daily would be your phone and virtual assistants. If you have an iPhone, and you swipe down on the home screen, you will see a search bar, and below that, you will see a couple of apps that are “Siri Suggestions.” This is a simple example of AI, Siri looks at the apps you spend the most time on and checks when you use those apps the most, using this information it will decide what apps to show on that screen during what times throughout the day. Furthermore, Siri, Alexa, Cortana, Bixby, and Google Assistant are all examples of Artificial Intelligence. They use patterns and models that they have at their disposal to mimic human speech and answer your queries as a human would.
All in all, we can see that we are making progress towards a perfect Turing Test AI machine. We have killer AI robots, robots that will answer your questions, and even robots that will drive you around. We are living in a world where AI is the next big thing, and it is imperative to know what it is, how it works, and how it applies in our life.
If you would like to learn more about AI, make sure you check out this video by 3Blue1Brown on youtube.