Entering the artificial intelligence maze


Original Source Here

Entering the artificial intelligence maze

Photo by Benjamin Elliott on Unsplash

The topic of artificial intelligence is vast and complex, encumbered by pretentious marketing, misunderstanding and false hopes. Like everything in science, everything gets easier if you have commercial backing so artificial intelligence is riding the same horse. Everything these days is stamped with the artificial intelligence or machine learning sticker to receive a certain aura of actuality and modernism. Weather the application really uses those technologies, nobody seems to care much. It has the stamp so it has to be better.

If artificial intelligence is explained by the science that invented it, computer scientists will fail to impress: artificial intelligence consists of algorithms that implement aspects of natural intelligence to give their results. The sentence is dry but there is a lot packed in there. In this article we will try to explain what is and what isn’t artificial intelligence, what makes it work and what computer scientists believe natural intelligence to be. All of this will become a small set of articles detailing the most important artificial intelligence algorithms used today. The articles will also help you understand what the AI or ML sticker on software products mean and how to identify fraudulent marketing.

Biology and computer science

Photo by National Cancer Institute on Unsplash

If we want to understand artificial intelligence, we need to understand what scientists think about natural intelligence. This is an impossible task to cover in an article, but there are intuitive shortcuts we may take. For example you can’t make a fish climb a tree, but you can see it selecting its food and taking certain paths and adapting to new situations and life conditions. All of this is a sort of biologic intelligence. Yes, the fish cannot talk and cannot use tools, but these advanced features are not where intelligence begins. Intelligence begins with natural selection, a force that drives adaptation and survival among biological systems.

We can add then on top of that multiple features, starting from very basic to very advanced: variability, fitness, memory, attention span, focus, language processing, classification and so on. But all of these are intelligence features, ways in which intelligence manifests itself. What is intelligence then at a more fundamental level? Scientists may boast and launch in explanations at neural level, maybe atomic level using fancy quantum mechanics, but in the end the ultimate explanation is still elusive. We know as much about intelligence as we know about what causes the space-time quantum fluctuations. They just “are” because that’s how quarks work. Why they work like that is a question not meant for science, it’s meant for architects that want to design a new Universe: we don’t want that though, do we?

Even though computer scientists don’t want a new Universe, they sure would like to have a new brain in their applications. That’s because a human brain is able to classify systems naturally, with extreme efficiency and very low power cost. It’s not just about heavily trained scientists either. Children learn about their whole environment at a frightening pace. Language, faces, intuitive physics: everything is learned in two to three years after a child is born. Yes, most of it is imitation from their parents but since we cannot fundamentally define natural intelligence we must conclude that imitation is a feature of it too.

Attempts at translating intelligence

Photo by Brett Jordan on Unsplash

So what did computer scientists do? They saw how biological systems work, they saw that they are able to optimize their behavior to match the problem they face, their environment and their need to survive, and wondered if they can implement all that in an algorithm. But how does a biological system adapt and solve problems? How does an ant find food? If an ant is able to find food without having a brain as complex as a human, surely an algorithm would be able to mimic that search for food and maybe translate it into something useful for science, like searching for a route, searching for an active web page mirror in an unreliable Internet.

And this is in the end, the definition of artificial intelligence. It’s not about your artificial foe that tries to evade your gun shots in a shooter game. That foe is better off using direct algorithms with ifs and elses to hide from your shot. It’s not about path finding in a strategy game. Again, path finding has better mathematical algorithms that work just fine, maybe better. It’s not about Google learning where you have been on the map and suggesting ads. Those features are easier if you just use traditional statistical formulas. Data mining is a statistical science at its core, it’s not about artificial intelligence.

Artificial intelligence is simply translating biological solutions into code. And in this article series I will talk about several translation options that have been explored so far. There are many more and of course, there is no bound where we can go, but what I cover here are the basic ones, the ones easier to grasp and use in actual industrial settings. So without further ado, here is the plan for the series, with bullet points updated into links as the articles are ready:

  1. Evolutionary algorithms
  2. Genetic algorithms
  3. Particle swarm optimization
  4. Neural networks

Each article will cover a huge class of artificial intelligence algorithms and I will try to explain it as easy to understand as possible. I will also try to cover fake marketing and what is advertised as artificial intelligence but isn’t and what would a true artificial intelligence solution look like.


Trending AI/ML Article Identified & Digested via Granola by Ramsey Elbasheer; a Machine-Driven RSS Bot

%d bloggers like this: