Grandpa careful this machine may be learning…
Over Christmas break, I had the pleasure of visiting my grandparents and explaining to them not only what I am currently up to at Holberton but what I am planning in my career as well as recent topics unbeknownst to them but yet everyone hears the same words from time to time especially over the news when it comes to Artificial intelligence so a lot of questions were raised on that topic (other than the facts that the last 30 years of movie and TV have covered this topic as well). To be quite honest I don’t know their background and experiences so it was a challenge, as usual, to jump into topics such as Machine Learning with broad concepts of Statistics. Machine Learning has helped to bring such advancements in technology to us such as full self-driving cars on their way, medical research now discovering new ways to gather data more efficiently, facial recognition, speech to text, auto-correct and much more! Going into this blindly as I had no expectations of what they understood of modern technology I had to explain some concepts which were, what does AI mean exactly? What is Machine Learning? How is it possible for computers to replicate the way we think? How do these concepts make sense logically and arithmetically to the average mind? Since that visit, I have had my own questions and upon research, I still ask myself where all of this is heading for our future as a species. This point of this article is to tackle not only what these concepts are for our generation but also to properly introduce them to the previous generation so we can all go forth together in a new age of Technology.
What is Artificial intelligence?
So what exactly is AI? Well, people tend to kind of confuse the terms and mix them around when AI really is really just the contrast of artificial versus natural intelligence. This quote I think puts it even more simply. “First coined in 1956 by John McCarthy, AI involves machines that can perform tasks that are characteristic of human intelligence. While this is rather general, it includes things like planning, understanding language, recognizing objects and sounds, learning, and problem-solving.” — (McClelland, https://medium.com/iotforall/the-difference-between-artificial-intelligence-machine-learning-and-deep-learning-3aa67bff5991). Its really about whether if a machine can mimic cognitive functions such as learning and problem solving just like you and me so there is can be an AI known as a string AI which should possess the ability to apply intelligence to any problem while Weak AI has to learn to solve the problem. This concept was very easy to understand for my grandparents and it made them think about their mobile phones particularly a feature they use and is common throughout all devices nowadays, text to speech. If AI is trying to get better at mimicking us then when it comes to text to speech how is it exactly going about learning from us which brings us to Machine Learning.
Machine Learning
Machine learning is exactly how it sounds, it is a machine that learns from input such as data but how the computer gets feedback, this quote summarizes it pretty much. “Machine Learning is a subset of artificial intelligence which focuses mainly on machine learning from their experience and making predictions based on its experience.” — (Atul, https://www.edureka.co/blog/what-is-machine-learning/). So how exactly does a machine exactly learn from experience? Well, you cannot simply just program to do this and that as it has to learn to make these choices on its own to achieve the correct result and that is how machine learning is more a way to achieve AI. So there are 3 main different techniques when it comes to machine learning, Supervised learning, unsupervised learning, and Reinforcement Learning. Supervised Learning is simply is giving the machine learning model a set of data and to achieve a result and tell it what the result should be and over time it will learn from these results and data such as giving it an image of a dog and telling it is a dog and overtime as it sees a dog we will expect it to give an image without the result and for it to tell us if it isn’t a dog or not and well this method takes a while as a model has to handle all the little exceptions, for instance, what if the model saw a Wolf which has a lot of fur just as a dog it would classify it as a dog due to the similarities between the two while we both know they aren’t one and the same. So handling these test cases are very important when it comes to working with a supervised model to teach it what it needs to know to correctly identify only the right thing and not things of similarity, this way of identifying involves a lot of statistical analysis and involves creating a decision tree, which is basically a way of drumming up conclusions from observations on the data given which could be either come out as either continuous or discrete. The main goal of supervised learning is to determine the function based on the data and predict the output over time. Unsupervised Learning tries to find structure in the data given and creates cluster which is the assignment of an observation. The model will right away find relationships in the data using clusters, so if we gave the model pictures of animals such as a cow and a chicken, it would create a cluster for each animal to build up an idea of what each animal generally looks like and separates them based on that, the model doesn’t know the end result like supervised learning. Unsupervised learning’s main goal is to find underlying patterns in the data given to it to which is why it doesn’t know the end result of the data given, this model is generally used to pre-train supervised learning models and pre-process data. Reinforcement learning is basically a way in which a model interacts with the data given to it and has to determine the best outcome off of its observations and the model is rewarded for correctly getting the right answer or penalized for getting the wrong answer and the observations of the model aren't supervised this it is free to make its own observations and is told after of which if it is right or wrong.
What is deep learning?
Deep learning is a subset of Machine Learning and can involve any of the 3 types of machine learning. the word deep comes the fact that there are hidden layers in the neural network, There are about 100 billion neurons in the human brain and while it is hard to replicate this on a machine as hardware becomes more powerful we will able to replicate the hardware built into our brains. “Deep learning is an artificial intelligence function that imitates the workings of the human brain in processing data and creating patterns for use in decision making. Deep learning is a subset of machine learning in artificial intelligence (AI) that has networks capable of learning unsupervised from data that is unstructured or unlabeled.” — (Hargrave, https://www.investopedia.com/terms/d/deep-learning.asp). Deep learning uses large sets of labeled data that it can learn from and can solve complex problems even when using a data set that is very diverse, unstructured and inter-connected, by using neurons which are just algorithms inspired by the human brain, its similar to how humans learn from experiences, a neural net will perform a task but each time will change the scenario a bit to see the different outcomes and learn from it. The model will perform tasks repeatedly tweaking it a bit each time to improve the result. In recent years thanks to the massive amount of data creation neural nets have been able to grow unlike before with all this training material. Deep learning neurons work the same way as they do in our brain, one neuron is useless as it gets data but with billions they can work big data sets, working together, constantly making the next input for the next neuron. This is how deep learning has so many layers to it. Deep learning requires a lot more data to teach the model as the model picks and chooses what separates one piece of data from another. After the input layer, everything is fed into a hidden layer consisting of a channel of neurons called bias to which goes into a thing called the activation function, the result of the function determines if the neuron gets activated after all the neutrons are done it gets passed to the output making things like converting text from language to another a simple task for deep learning.
In conclusion, Machines are rapidly learning and thus improving. Computers are slowly becoming more and more like us with the way they now can organize data, understand us in a variety of ways and with the industry putting billions into machine learning we will only see improved services and machines start to interact with us in new ways we never expected it may just be soon enough we start to see our own Data android in the world helping us to explore the unknown.
references:
https://medium.com/iotforall/the-difference-between-artificial-intelligence-machine-learning-and-deep-learning-3aa67bff5991
https://www.edureka.co/blog/what-is-machine-learning/
https://www.investopedia.com/terms/d/deep-learning.asp
