For hundreds of years humanity has been both enamored and fearful of intelligent machines. Now, we are in an era of leading-edge discoveries with new information published about machine learning, deep learning and artificial intelligence every day. These terms, however, are not as interchangeable as tech companies and marketers would have us believe. They more realistically form a pyramidal structure where one concept advances the previous one.
Machine Learning: Basic Training
Machine learning occurs when a computer utilizes complex data processing algorithms to make predictions. Devices equipped with this capability employ an Artificial Neural Network (ANN) which allows a device to make sense of what it sees. It can then craft assumptions based on past behavior.
There are three types of machine learning: supervised, unsupervised and reinforced. The first version is like learning a new language; the instructor shows you an apple and says “this is an apple.” You put together the visual (sample) and word (labeled data) to come to a conclusion.
Unsupervised learning occurs when the same sample is given, but the data isn’t labeled. You are shown the apple yet not given the word for it. You have to pull on previous life experience to try to find the answer. A computer in this case has been trained to make these identifications by being provided with hundreds of thousands of data points to come to the correct conclusion.
Finally, the reinforced method involves rewards. When the computer makes the correct association on its own, a type of code triggers saying it is right. The machine will then continue to find the ideal pattern that results in the greatest reward. You are more likely to retain the translation for “apple” if you receive a sort of gold star when you are successful.
A common way we see machine learning is the rise in chatbots. These virtual beings are taught through a reinforced model how to correctly respond to humans. When undergoing machine learning, developers input a variety of data from prior customer service transcripts to real-world (i.e. human) knowledge. Companies can then offload standard call/chat centers to the machine learning devices, saving time and money in the process.
Deep Learning: Adding Layers
Deep learning builds on machine learning by utilizing an ANN with many layers, effectively becoming a “deep” neural network. It takes a vast amount of unlabeled data and translates it into an output that can be utilized. This reduces the amount of time human engineers need to spend training a device on-site as it has the necessary tools to learn from its own environment.
There are essentially three parts to deep learning. The first is data input where the computer sensor receives as much information as is available. It then undergoes a sort of “if/then” algorithm. Based on hundreds of hours of pre-training human developers put into a device, it seeks to determine an accurate conclusion or hypothesis from the data, the final stage of deep learning. Once deployed out of the lab, a deep learning computer or software should be able to read its environment, whether virtual or physical, on its own without ongoing human interaction.
The medical field is already working on implementing deep learning into their diagnostics systems. This is not to say that machines will take on the role of health professionals; rather, deep learning enabled devices will assist physicians in identifying illnesses and increasing overall accuracy. When a computer has been properly trained with millions of images that show healthy vs. diseased organs and tissue, it can be applied to current patient health records as an effective support tool.
Artificial Intelligence: Autonomous Learning
Where deep learning is a step above machine learning, artificial intelligence goes even further. You can train a computer to identify objects, animals and people, but it will not inherently understand the variances among the three. Artificial intelligence is the difference between a device knowing how to get from one point to another and understanding why it needs to make the journey.
The sheer amount of effort that goes in to training a supposed “artificial intelligence” is a giveaway that it is not quite there. Machine and deep learning devices require quintillions of data and months to years of human training just to get them to 90% accuracy. True artificial intelligence would be able to accomplish the same, if not better, with minimum teaching.
Most technologies currently dubbed “artificially intelligent” are actually high-powered deep learning models. They are excellent users of data and can greatly assist humans in making sense of the immense amount of digital information we are bombarded with daily. Machine learning and deep learning are providing researchers with a rich foundation for what will eventually be artificial intelligence.
The future of autonomous learning is closer than ever. Are we ready?