We are slowly starting to live in the world filled with AI. Or so you were told. But what does this term mean and what it changes for us? Should we be terrified? How Close are we to the so called True Artificial Intelligence? We will try to briefly answer these question.
Artificial intelligence (AI) refers to the simulation of human intelligence in machines that are programmed to think like humans and mimic their actions. Even though there are many ways to make our desired target behave intelligently, every Machine Learning Algorithm is internally based on mathematical models. These models are very often inspired by nature and biology. For example, evolutionary algorithms are based on Darwin’s Theory of Evolution by Natural Selection. Neural networks – a very popular word recently – which models are greatly inspired by the structure of the human brain. With the usage of probability, statistics and calculus these algorithms can look at data and infer results based on it. With the right choice of data and method for problems that we face we can obtain results unobtainable otherwise.
However, what is important, that artificial intelligence is not some magic that we can just thrust data onto and obtain results we want. It is nothing more than a tool and as with any other tool we have to use it properly.
How Close are we to artificial intelligence?
2019 was called the year of AI. It was named that way because of the fact that in this particular year we really have witnessed great progress in this field of research. The question of how close we are to artificial intelligence is not a correct question because we live in a world where AI powers things on our daily basis. For example, thanks to it you can not only easily find information you need – it can be directly served in front of you knowing how you behave. That is how Youtube, Netflix and others are recommending you cat videos and series you like so much. However, if you ask how far are we from so called “strong” artificial intelligence, which means AI that has a capability of human brain and beyond, well – we are quite far from that. But we are making progress. Anyway, even though currently AI is not as smart as humans, it can surpass us in many fields. The cause for that is the fact that computers that run our AI algorithms are never tired, never distracted, and can process much more data and do it faster than we do. We are not far from artificial intelligence. We have already lived in the age of A.I. But there is still much more to be done and discovered.
How Artificial intelligence works?
Depending on the type of Artificial Intelligence it can work differently. There can be an AI that works by looking at some dataset and memorizing it. Then it uses its memory to make some predictions and decisions. For example, we are trying to learn a car how to drive without any help from human. One approach to this problem is to give the AI tons of hours of videos about driving a car. Based on that, it can learn how to make decisions on the road. On the other hand, we have the evolutionary approach. It was presented by DeepMind company which created an AI for the game called StarCraft. The AI was supposed to play with itself but on different settings. With each subsequent game it improved itself and learned how to make some decisions and predictions. Thanks to that by the hundreds of gameplay we have almost perfect StarCraft AI that is able to defeat 99.8% of players. There are other types of AI that can work pretty differently from one side. But it does not matter if you are working with some text classification or image generation. The basis of this whole area of science is making predictions. But by any means we have to realize one thing – the AI is not capable of being self-aware. At least yet.
How to make artificial intelligence?
Actually this question is not easy to answer. It depends on the complexity of a problem, availability and quality of data, hardware that we have and many other things. However some general approach can be shown and it looks as follows. To make artificial intelligence, firstly we have to understand what we want to achieve. Depending on the problem characteristic we will be taking This different strategies. Then we need the right data. If our problem concerns recognition objects on photos we have to collect data in this form. If we want to check the sentiment of people commenting under our youtube videos then we have to collect many comment messages and also label it as positive or negative. way algorithms can look for patterns and check what is characteristic for both types of comments. And the list goes on and on. We have to remember that our data has to include the information from which we can actually infer results. For example, you cannot predict the price of a house based on the number of flowers in it. No matter how much data you would supply. Then, we have to choose the right tool. There is a theorem known in machine learning called there is no free lunch theorem. It basically says that there is no universal tool for each problem. When we have a general idea we have to prepare our data, clean it from a mess, delete unnecessary information, visualize it, split to parts for training, testing and validation. After all these steps we can run our algorithm and check results. Very often they will be not satisfactory and there can be many causes of that. It can be a fault of the data, wrongly chosen hyperparameters of our machine learning algorithm, wrongly chosen algorithm itself, or others. If that happens we have to visualize our results again, think what went wrong and tune our model to work better. This way, we can obtain a working algorithm that not only works for data we gave it but also the data that will just happen. And that is our goal.
How artificial intelligence will change the future?
Right now many people in the world are scared of AI taking control over the humanity. We have seen many examples where artificial intelligence is simply terrifying. There are examples of chatbots that are so human-like that there is no telling that it is a machine. Also, there is a matter of AI being not 100% trustworthy. Because of that society is angered by AI taking some important positions. But as Sun Tzu said in Art of war “In the midst of chaos, there is also opportunity”. Thanks to large feedback from our society AI developers can create better and better solutions. Also, there is a new, younger generation coming. For them AI is something normal, something that we use on a daily basis. We need more and more people to be good with AI taking care of most of the jobs in our society. Probably in a couple of years we will have completely automated fabrics, shops, cars. Possibilities are countless. You can say that the future is now as many of this stuff is happening right now. It is almost now, we just need to wait for the social approval.
Which programming languages are used for AI?
There are many programming languages in the world. Typical programmer could say that machine learning could be implemented in ‘W-machine’, it is just an algorithm. But there are some languages that have certain benefits while making artificial intelligence.
Current No.1 in the market of AI. Has many libraries (TensorFlow, Numpy, PyTorch) and community support. Easy to learn. Can be later easily implemented in some other project.
One of the most popular languages in the world. Computationally faster than python. Even easier to implement in other project
Great for statistics. Lots of libraries for mathematics and machine learning. Syntax friendly for mathematicians and statisticians.
About the authors:
Krzysztof Kramarz – Big fan of USA and Asia culture, recently created an AI that mimic his behaviour on messenger. One day would like to write an AI that runs a bar where you could listen to some AI-driven jazz music and drink some good whisky. Machine Learning and Data Science developer in WASKO S.A. and a student of Silesian University of Technology.
Damian Kucharski – Working on data engineering at WASKO S.A. and studying Computer Engineering at Silesian University of Technology. Privately NLP enthusiast. Currently working on intelligent assistant for programmers helping to find answers for questions on for example stackoverflow.