Most of you by now have heard of Artificial Intelligence. It has been hyped up as the technology that will change the world in the coming decades. It promises self driving cars, customer service chat bots, programs that can think and learn almost anything by themselves, and much more. But what is the term Artificial Intelligence really referring to? In this post we will attempt to cut through the hype and explore some of the sub categories falling under the umbrella of AI.
WHAT IS ARTIFICIAL INTELLIGENCE
There is an easy way to define AI and then there is the long winded more abstract way to define AI. It can be defined as a sub-field of Computer Science. The goal of this field is to figure out ways to program computers to learn on their own which can be accomplished by simulating “intelligence” since this is how humans learn. This form of intelligence would be created artificially by a programmer hence the term “Artificial Intelligence”.
One could also define it as any kind of man made entity that has the ability to use working memory to utilize cognitive functions like abstract reasoning and logical deduction and able to learn new things autonomously. This entity would also have the ability to form long term plans into the future using these cognitive abilities. Of course, this definition won’t accurately describe AI until we actually reach the point where the programs we create possess real intelligence. Current AI is far behind this benchmark, most programs are only able to perform autonomously in a very narrow domain which limits their functionality.
The overall goal it seems is to create an agent that possesses what is referred to as “general intelligence” which would allow the agent to learn a variety of novel things across multiple domains. Humans accomplish this with what is called “fluid intelligence” and it is also what IQ tests attempt to measure. The methods by which an agent of such sort could be created are what define the various sub categories of AI. There’s more than one way to make an intelligent machine.
Machine learning is an application of artificial intelligence that provides systems the ability to automatically learn and improve from experience without being directly programmed. The focus here is development of computer programs that can learn on their own given a set of data. The process begins with a data set related to a particular area of study. There is an observation phase in which the program has to parse over the data and organize it into a form that it can use. It then has to spot all the patterns in the data and as a result make better decisions than a human could. There are a few different kinds of machine learning algorithms
Machine Intelligence is disrupting multiple industries as shown by the landscape map above.
SUPERVISED MACHINE LEARNING
Supervised machine learning algorithms can apply what has been learned in the past to new data using labeled examples to predict future events. A solid and preferably large data set is needed in order for it to work properly. These algorithms can compare their own output with the input to determine if any errors were made and if so the model can be modified as needed.
UNSUPERVISED MACHINE LEARNING
Unsupervised machine learning algorithms are typically used when the information needed to train is neither labeled or classified. Unsupervised learning studies how systems can infer a function to describe a hidden structure from unlabeled data. These systems draw inferences from data sets and attempt to describe hidden structures in the unlabeled data.
REINFORCEMENT MACHINE LEARNING
Reinforcement learning is a learning method that interacts with its environment by producing actions and discovering errors or rewards. This is modeled after the reward learning and fear learning processes that take place in the human brain. It’s a trial and error search for the correct way forward as the agent receives feedback on its performance and uses the information to improve. Simple reward feedback when the agent performs correctly is all that’s needed to ensure success using this method. The process is strikingly similar to how humans learn to navigate novel environments and how to behave in them.
SEMI SUPERVISED MACHINE LEARNING
These algorithms fall somewhere in between Supervised and Unsupervised learning. They typically use some labeled and some unlabeled data for training. A small amount of each data type is combined to make for a more unique training set. Usually, semi-supervised learning is chosen when the acquired labeled information requires skilled and relevant resources in order to learn from it. This model is actually quite a bit more accurate than either the Supervised and Unsupervised methods.
MACHINE LEARNING APPLICATIONS
With the typical machine learning algorithms ability to parse over data, find patterns, and make more accurate predictions than humans there are a lot of real world applications that are now being dominated by machine learning. Speech recognition programming is a common area where appropriate algorithms are able to recognize what constitutes speech and learn the associated patterns to effectively learn languages. This opens the door for all kinds of language translation programs and also chat bots.
Medical diagnosis is another area in which machine learning outperforms humans. In fact, many experts believe that doctors will be replaced completely by AI programs equipped with the appropriate machine learning algorithms. We aren’t quite at that point yet but it’s coming. These programs can effectively recommend treatment options for patients given the correct data, they can also diagnose a wide variety of diseases and conditions given sets of symptoms, and even predict how a disease might progress. Naturally the main focus here is that in finding the best treatment for crippling diseases we might also find cures.
Image recognition and facial recognition is one more area of machine learning that is already in widespread use. The facial recognition has some potential uses for law enforcement and is already being used to identify suspects based on an image alone. This program accomplishes this by measuring each pixel in terms of brightness and color, white and black pixels are assigned a certain value and color pixels are assigned a value according to their brightness which is usually measured in the 3 color components that make up the RGB scale. The result is the program can figure out if there is a face present in the image or not, if there is then it can cross reference the face against the many faces in a database and make an accurate match.
CONTINUING WITH PART 2
In this post we defined Artificial Intelligence and explored one category and many subsequent sub categories of the field. In part 2 we will look at Deep Learning and Neural Networks and much more.