By Gael Hyppolite
Artificial intelligence and machine learning have both become something of a buzzword within modern tech. Many modern products boast the use of neural networks or advertise AI as a part of their functionality. You’ve likely experienced targeted ads or filtered content on some media platform; that’s AI. Perhaps you have a voice assistant on your phone or computer that can answer questions and perform simple tasks; that’s AI. Maybe you’ve played against the CPU in a video game; that’s AI. AI shows up in a much wider range of situations than most people might think, from chatbots to email spam filters.
However, despite the prevalence of artificial intelligence, many people don’t actually have a great understanding of what “AI” truly means, and what its connection is to machine learning. For example, of the technologies I listed, only a couple of them make use of ML despite all of them being AI. This of course raises the obvious question: what, if any, is the difference between AI and ML?
What’s the Difference?
The simplest, or most direct answer to this question, is that AI refers to what the technology does, while ML refers to how a technology works. To understand what this means, let’s take an example: chatbots. In its simplest form, a chatbot provides pre programmed responses to specific statements. You give a greeting and it responds with “Hello”, or you ask a question and it looks up an answer. This is an example of a “Type I” or “Reactive AI”. Being the most basic type of AI, it simply takes in and responds to information. These types of AI have no memory or internal understanding of the tasks they perform, they simply provide a particular response when presented with a particular situation. In other words, these types of AI are incapable of “learning”.
The chatbot qualifies as AI because it attempts to mimic humans through its limited ability to converse. However, ML refers to a machine that is capable of learning from the situations it encounters and adapting how it approaches and solves on its own. Therefore, the chatbot doesn’t make use of ML, as it fails to “learn” anything from its interactions. An example of AI which does make use of machine learning would be the youtube recommendation system. Everytime you click on or like a video, youtube updates an internal metric that it uses to decide which videos that you’ll most likely want to watch next, and then recommends those videos to you on its front page. It is therefore a “Type II” or “”Limited Memory AI”, as it remembers and learns from the situations it has previously encountered. Most modern AI fall into the “Type II” category, which effectively requires that they make use of machine learning.
Therefore, the fundamental difference between machine learning and artificial intelligence, is that ML refers to how a subset of AI is implemented. As for AI itself, the definition of artificial intelligence is much looser than many people realize, simply referring to machines that mimic human intelligence in some way.
The Future of AI
There are in fact 4 types of AI that we use to classify the development of artificial intelligence. I’ve already explained the first 2 of these, but I will list all 4
Type I AI: Also known as Reactive AI, they are only capable of giving predefined responses when presented with a specific situation. These AI have no concept or understanding of the tasks they perform.
Type II AI: Also known as Limited Memory AI, they can keep track of and learn from past events in order to adjust their actions in the future. Most AI today fall into this category.
Type III AI: Also known as Theory of Mind AI, these are currently the realm of imagination and science fiction. While none have been created, they would theoretically be capable of understanding humans and other entities as individual agents. They can understand the deeper meaning behind human emotion, learning and responding to stimuli more quickly by understanding intent.
Type IV AI: Also known as Self Aware AI effectively represents the singularity. This is the point at which AI not only understand the thoughts and emotions of others as agents, but recognize their own individuality and possess their own thoughts and feelings.
The jump from Type I to Type II AI required the development of one of the greatest innovations of the 20th century: machine learning. Furthermore, it required us to refine ML over the course of nearly a century in order to make it a feasible technology. The ability for machines to learn was so revolutionary in the field of AI, that many people began mistaking it for AI itself (though you’re no longer one of those people). Since then, we’ve made attempts to develop Type III, but we’ve made no significant progress. Such a massive jump in AI technology will likely require the development of a strategy similar in power to ML, not to mention what would be required to develop Type IV AI.
Such developments are probably decades if not centuries off, as we’re still only now beginning to understand the power and applications of Type II AI.