Traces of wanting to replicate human-like mental ability can be found in the history of almost every society. However, it wasn’t until the 1950s that artificial intelligence was officially recognized to be a field of research. With the field being more than half a century old, it’s a wonder we haven’t seen more progress. Or at the very least, why haven’t we progressed further? Yet, it seems that more and more these days we’re hearing about the next intelligence agent coming to your cell phone. Why do you think that’s happening now? Why is artificial intelligence the hot field?
Other forms of media like to use the idea of AI because it seems so simple to just say that this machine is acting like a human. You just write the character like any other character, but you preface their history with the fact that they are a robot. The truth is, Artificial intelligence on its own is a deceivingly wide field of study. Not only is it considered a popular field of study in Computer Science, but its roots propagate deeply into the fields of mathematics, physics, philosophy, and even psychology. What does it actually mean to be artificially intelligent?
All living creatures are born with some accumulation of what is called natural intelligence. In other words, there are certain things that we’re just born knowing and nobody has to teach us these things. For example, humans are naturally inclined to know not to repeat behaviors that cause us any form of pain. Nobody has to tell us that we don’t like being hurt, we just know it to be true and we take action to prevent those situations from happening again. Not only do we have the natural intelligence that we’re born with, but we also have an unquestionable ability for learning and adapting to unseen situations. But, what about computers? What do they have?
Artificial intelligence itself has since expanded into multiple sub-fields. Many of these fields have yielded incredible results, but these results are typically only relevant in that same field. We’ve had AI systems built into video games for years, systems that can detect cancer earlier and earlier, or the ability to know when your credit card has been charged for a transaction that wasn’t initiated by you.
However, up until now we haven’t found a way to handle what many call “Strong AI”. Strong AI is a synonym for human level intelligence, which we know requires multiple fields of AI to be able to mimic. Among these fields are computer vision, machine learning, and natural language processing. Computer vision is the field of understanding what might be captured from a video stream, machine learning is understanding how to adapt to new data, and natural language processing is being able to understand the words being presented to you.
When looking at how artificial intelligence is handled by members of the community, you’ll quickly find that there’s ultimately 3 factors that directly affect the success of artificial intelligence solutions. These three factors are:
1. The math involved with a given problem
2. The sheer computational power needed to process data
3. The amount of data available to help study and solve the problem
For a very long time, many researchers have announced that we’re just around the corner from solving the AI problem and we’ll have human level intelligence any day now. We’ve been saying this ever since the initial dive into AI, and the result always ends up the same. We underestimate all of the resources we need to achieve a good AI system.
Interestingly enough, if you look at the history of AI, you’ll notice that there’s a massive increase in interest each time a breakthrough is made in one of the 3 factors listed above. Every time one breakthrough is made, we make the assumption that we’ll be able to solve the other 2 almost immediately. If you remember some simple algebra, only having 1 variable in an equation does NOT let you solve the rest of the family of equations. We need more than that.
First let’s look at the math involved with AI. AI very heavily revolves around calculus and even more so it requires statistics. Artificial intelligence seen from a math perspective is a web of weighted averages. In other words, mathematicians need to find a way to take multiple sources of data or input, and compile them into usually what is a single value or a very small set of values. This value needs to very efficiently represent many different features of the various inputs. Pioneering different strategies for this problem is very difficult, but with how old math is, it didn’t take very long for AI researchers to find multiple different strategies that could transform data into formats that became much more meaningful to computers. However, these math equations haven’t changed in decades, and though minor tweaks have surfaced recently, the main concept of the math involved hasn’t really shifted much.
This brings us to AI factor 2. Humans process more information in any given moment than most people seem to realize. For example, if we were to assume that human vision is a constant stream of 4k resolution video, that means that we’re processing 3840 by 2160 pixels roughly 30 times per second. This doesn’t even account for our field of vision. That becomes 248,832,000 pixels of data that we are accounting for at any given moment. Pixels, or color in general, is often represented using 4 different values. Alpha, red, green, and blue values. Assuming a standard computer allocates 1 byte for each of those values, that’s 4 bytes per pixel. This comes out to 995,328,000 bytes worth of data. This means that every single second, the human brain is visually processing roughly 1 gigabyte of data. If you start adding in the data collected by our other senses, the numbers make you really appreciate how powerful our brains are. Before even thinking about how to handle processing all of that data, we had to find a good way to store it all. The first 1 gigabyte hard drive was released by IBM in 1980, meaning we had to hold off on AI before we had a reasonable way to keep all of this information. It wasn’t until the last decade or so that we’ve invented equipment that can process that much data in real time. Even today, we’re using a previously unheard of amount of computers linked together to be able to process all of that data.
However, those 2 factors bring us into today. Now that we have 2 variables, and our target solution, we actually have what we need to grab the 3rd and final factor for what makes AI work. It takes humans a countless number of hours to learn many things in our lives. Even just before we start school, we’re already controlling the muscles in our legs and arms to help us balance and walk. We’ve already learned to extract semantic meaning from conversations we have with our parents, and we can even reply in a meaningful way. With advances in computers, we’re not only able to store all of this information and process it, but we’ve actually stockpiled enough information in the last few years that we can use it to experiment.
Whenever you ran a science experiment in middle school, the scientific method always called that you had a well-defined plan before you could collect data. Then you had to go and collect the actual data. After you had the data, you could make a conclusion as to what it all means. This is the same pattern that AI has evolved over time, and now that we are able to make conclusions over the information we’ve amassed, for the first time in history, a real artificially intelligent system can actually be built.
This is why AI is the hot new buzzword in tech hubs around the world. You hear about it constantly, because it is actually happening. We have cars that are able to assess driving conditions on their own without any human interaction, and have them drive you to their destination without ever needing to tell the car what to do. We have neural networks that are able to derive what it thinks is a cat, again without ever needing to be explicitly told what to pay attention to. It’s an incredible time to work on AI, because we have all of the parts we need to build a true AI. Now it’s just a race to see who gets there first.
This post was written by Aaron at Josh.ai. Previously, Aaron worked at Northrop Grumman before joining the Josh team where he works on natural language programming (NLP) and artificial intelligence (AI). Aaron is a skilled YoYo expert, loves video games and music, has been programming since middle school and just turned 21.