Machine Learning: Both Sides of the Coin — TechVirtuosity

Original article was published by Brandon Santangelo on Artificial Intelligence on Medium


Machine Learning: Both Sides of the Coin — TechVirtuosity

Copyright : Kittipong Jirasukhanont

An Artificial Intelligence That Learns

Artificial intelligence has long been something that needs constant improving. Whether it’s the games we played as a kid or our smartphone voice assistants. AI is basically used almost everywhere and machine learning is the next big thing to already happening!

Years ago we referred to artificial intelligence or AI for short, as something that was smart. Truthfully, AI was initially just a set of instructions or code that would be used to guide game opponents or other somewhat aware objects. They were anything but smart and they were plagued with flaws. They failed tasks, ran into walls constantly or just simply didn’t work.

The future of AI is to incorporate different methods for learning. When they fail they’ll learn and improve and this is known as machine learning. It’s a program with goals and parameters and then the ability to measure success. The better it does the more likely it will succeed. But it isn’t that simple!

The Long Journey of Learning

When using machine learning you have to first figure out what your goal is. Is it to reach a destination? Maybe it’s to solve another problem? It needs a goal but more importantly it needs to understand success.

AI can be very smart but if it doesn’t know how to measure success then it doesn’t know how to learn. We learn by doing and discovering better ways to accomplish something. Maybe we are driving down a road and then later discover a shortcut to get home faster. The parameter would be the measure of time it takes to get home, from point A to point B.

AIs can solve problems in a similar way, but they take longer. Often we see machine learning being used for hundreds of hours just to solve a specific problem. Unlike humans who recognize the options available, if an AI were to start driving it might drive off road to get there the fastest. Clearly that’s not a solution.

It takes a lot of time and lot of rules before any kind of learning can begin.

What Machine Learning can Accomplish

While it’s definitely not perfect it can be used to help improve our understanding. By taking AI a step further we start to see a solution begin to develop. Most commonly our version of AI is set in stone and it does what we tell it to do. But this way of thinking can limit how an AI can improve when it’s already hitting it’s max capability from the beginning.

By using machine learning principles we can actually improve an AI beyond what we can tell it to do ourselves. Maybe the solutions are better, or perhaps new methods are used we didn’t think about. The aspect of learning is what makes the AI capable of solving problems that would otherwise take us years.

It’s also important to understand that machine learning can run thousands of simulations at once, whereas a human is limited in what they can focus on at one time. This allows an AI to solve a problem in a fraction of the time, if all goes to plan.

Following the Path of Nature

In nature it’s the weakest that tend to die and AI is no exception. Machine learning when built properly, will collect data and compare it to the success parameters. As it improves things like maybe time completion, it removes older data sets that had poorer results. This is how it learns.

It takes the data from multiple simulations and picks out the best performing ones. This can work depending on the rules of success, but generally the AI will gradually improve.

Sadly, unlike nature, AI needs a helping hand and doesn’t have some of the more basic intelligence that it needs at first. This is where machine learning starts to cause problems.

Why we Struggle to Build Successful Learning

With machine learning becoming so popular, it’s become a buzzword in itself! But a lot of demonstrations don’t prove much at all. An AI that takes thousands of hours to improve isn’t much of an improvement from what we currently have.

A lot of companies will build a prototype that doesn’t properly show what a learning AI can accomplish. This can lead to wasted time and resources, as well as another example why AI can’t be trusted on basic tasks. But that’s not entirely true, because when it’s implemented the right way we have gotten improvements to medicine, facial recognition and a variety of other fields and technologies!

But in general, companies fail who view it as a buzzword only. You need to have a clear objective with a clear approach to something that AI can do. In the end, AI does repetition the best and that allows it to learn the best. If it’s a task that would require repetition from us then maybe AI should be considered instead!

Knowing What Machine Learning Does Best

If we build on the idea that AI can replace humans in everything, we are bound to fail early on. While AI is amazing it isn’t a solve all kind of solution at this point in time. Machine learning does best with repetitive tasks.

This isn’t to say that a learning AI can’t accomplish more than basic tasks, especially as technology improves. It does great with comparing patterns to improve itself, but it struggles on things that don’t necessarily have predictable outcomes.

If you are trying to create solutions to pollution, better battery technology or anything that requires creative thinking, then AI will likely not succeed. But machine learning does allow us to gather data into one place and create results on what we know already.

What it lacks in creative thinking it can make up for in speed when solving complex problems.

[Copyright : Sebastien Decoret] © 123RF.com

Being Patient for the Future

Like with most technologies, it takes time to improve them. AI has been around for many years but machine learning is a newer way of thinking about it. It isn’t perfect, sometimes it isn’t pretty, but it’s a necessary step forward!

Machine learning could one day go well beyond how we see it now. It could be improved to understand the parameters beyond what we tell them (some companies have started this already). Maybe it’ll develop solutions that we could never think about on our own. After all, AI has the ability to operate and solve at speeds that far exceed our own thought process!

But until then, we’ll just have to be patient with how machine learning works. Letting it use data sets and attempt to improve itself over several hours of trial and error.

What do you think about machine learning? Do you or your work place use it? Share your thoughts and comments below!