Source: Deep Learning on Medium
Your future depends on AI — we’ll help you understand it.
A special thank you to Dr. Richard B. Freeman, Harvard Professor of Economics, for his guidance on this production.
You’ve probably heard of AI stealing your jobs, machines beating chess grandmasters, and everything short of robots taking over the world…
But what does Artificial Intelligence really mean, I mean why is it suddenly so trendy? The answer is surprisingly simple. Though artificial intelligence and deep learning have been long hypothesized, the computing power and Big data, which ranges from everything like google searches to Spotify playlists, has only just become realistic in the past decade.
So what exactly is AI?
Artificial intelligence is the umbrella term that houses words like Machine Learning, Deep Learning, and Neural Networks. It simply refers to when machines are programmed to make decisions as humans do. For example, instead of simply giving a machine an algorithm to follow that’s full of “if this, then that” statements, we program machines to genuinely learn like humans do and make predictions based on their past experiences to improve the accuracy of their outcomes.
Machine Learning, a subcategory of AI, describes the process machines use to learn from their past experiences, in this case, data. Depending on the type of data we provide the machine, it will follow through with different types of learning. The two main categories of Machine Learning are Supervised and Unsupervised Learning.
With Supervised Learning, we feed the machine labeled data. Labeled data is simply data that comes with a title and corresponding characteristics. For example, let’s create a machine, and let’s call her Bot. We can teach Bot to identify a particular dog amongst a group of cats by feeding Bot different pictures of this dog, as well as pictures of cats. Along with these pictures, we will give Bot the differentiating characteristics of the dog in question, like big drooping ears and a large snout. Bot will be able to find the dog in the group of cats by filtering through these characteristics and, the more data Bot processes correctly, the more she will be able to improve her identification algorithm.
If, however, Bot makes a mistake and accidentally identifies a cat as the dog, human intervention is required to correct the algorithm. With more data and more positive results, Bot will slowly learn from her past experiences and become more accurate at identifying correct results — that’s machine learning in action.
With Unsupervised Learning, on the other hand, we would simply hand Bot lots of data that has no label. One of the reasons why we might do this would be because the differentiating characteristics between the data types are now hard for humans to identify. For example, let’s say we wanted to challenge Bot to identify any dog instead of one particular dog. Unfortunately, now we can’t simply tell Bot that “all dogs will have large snouts and larger ears” because that just isn’t true. Instead, we leave Bot up to its own resources to identify patterns and classify data by feeding her large amounts of data, in this case, different pictures of dogs.
How exactly does this Machine Learning happen? Here’s an example of a potential learning technique that Bot might use to tell the difference between dogs and cats. For our purposes, let’s say that Bot has discovered that all dogs have large ears and long snouts and no other animal shares this characteristic combination. Bot will spread out the data that it is fed into a chart, where the x-axis represents how large the snout of the dog in the picture is and the y-axis indicates how large the ears of the dog in question are. The data will fall into clusters and the upper right-hand corner will represent the cluster of animals with large drooping ears and long snouts — these are the animals Bot will identify as dogs. It is obvious now that if Bot receives a picture that falls into this description, Bot will say that the picture is that of a dog. However, what if a data point falls in between different clusters? What will Bot do then? Well, Bot can use the K Nearest-Neighbor learning technique and draw a circle around the datapoint in question. In this case, this circle will encapsulate 4 other data points, 3 of which are confirmed dog photos and 1 that is a confirmed non-dog picture. This learning technique, coupled with other analytical strategies, helps Bot make up her mind that the picture she is being presented is indeed that of a dog. If Bot gets validation for this prediction from a human, the datapoint will be preserved and she will have successfully completed machine learning because Bot now has one more data point and will be able to better identify dogs in the future.
What happens, though, when the characteristics of the data are even harder to define? For example, how would a machine recognize a handwritten number 9? If a math problem asks for the answer of three times three, humans can easily recognize that a nine in cursive, block letters, or bubble letters are all equally correct. This, however, seems like an impossible task for machines to understand because the variation means we cannot program the computer with a template of the pixels that constitute a figure 9. This is where Deep learning comes in.
Deep learning is a subset of machine learning and is used when differentiating characteristics between data points are even tougher to identify. Deep learning can only work with the help of large volumes of labeled data and Neural Networks.
A Neural Network is an algorithm that mimics human neurons. What’s really interesting about Neural Networks is that, like the human brain, the algorithm itself changes as new information is processed and interpreted to make the machine’s outcomes more accurate. The main difference between Deep Learning and simple machine learning is that deep learning does not require human intervention. Instead, deep learning is self-mediating and will adjust and ameliorate its own algorithm. In addition to being self-sustaining, certain deep learning techniques, like Convoluted Neural Networks, also do not require manual input of characteristics because the neural network alone, through large volumes of labeled data, can deduce the traits necessary to classify the data.
Let’s zoom out! In just ten minutes, you learned that AI is only possible thanks to the computing power and big data of the modern age. You also learned that Machine learning is simply the process that machines use to learn from their past experiences, and that there are different types of learning, like supervised and unsupervised learning. Finally, you learned that Deep Learning is just a fancier type of learning that uses neural networks, algorithms that are self mediating and modeled after the human brain.
Now that you now know all of the basics of AI, is all the gossip about the AI Robots takeover really a threat? Will AI really steal your jobs? Tune in to our next video and article to discover its repercussions on your job opportunities. A special thank you to Dr. Richard B. Freeman, Harvard Professor of Economics, for his guidance on this production. For all of our sources, facts, and figures, and a copy of this note, head over to our website at Letstimetravel.com or head over to our youtube channel to watch our live animations. Stay curious!