Laymen's Understanding of Basic AI

Have you ever thought about what was involved in the making of Sophia? An extremely famous human-like robot, who claims herself, in part, as a ‘personification of our dreams for the future of AI, and is proud to be designed to help’.

Isn’t it interesting that we are all trying to fix some kind of a problem and need help in many ways? It’s even crazier to think about why there are problems in the first place. That’s a topic for another day, even Sofia, with all her wisdom and her own emotions would have a hard time empathizing with us! AI is one of those tools that can help solve some of these problems and aid humans in doing life better.

Artificial Intelligence

Alright, before we look at some complex concepts, let’s try to understand some of the popular and fundamental concepts of AI in this article.

 From a high-level perspective, there are three kinds of AI

  • Narrow AI
  • Artificial General Intelligence (AGI)
  • Super Artificial Intelligence

AGI is defined as something that can understand or learn any intellectual task that a human being can, and we know that we are not there yet. Super AI, is the next level to AGI, which can be defined as something that possesses intelligence far surpassing that of the brightest and most gifted human minds. What are the implications of creating one is a whole another topic for another day!

Narrow AI is about learning patterns from the data. In this article, let’s go through some popular kinds and methods of narrow AI to solve these problems. One of the popular methods is using Machine Learning. 

Machine Learning

In the late 1950s, after WWII, there was a man named Samuel Arthur who did groundbreaking research in the field of AI while working for Big Blue (nickname for IBM). He designed a computer program to play checkers. The more the program played, the more it learned from the experience, he then popularized the term ‘Machine Learning’.

As the name suggests, the machine is learning from the data and finding the patterns without human intervention, and it is the goal! To find patterns and learn, so the machine (technically speaking, model) can predict the new data point based on the learnings.

These problems can be solved using different kinds of ‘learnings’, let’s look at the popular ones.

  • Supervised Learning
  • Unsupervised Learning
  • Semi-supervised Learning
  • Reinforcement Learning

Let’s say you want to predict the price of a newly built red house in your neighborhood. You give the data of all the surrounding houses in the neighborhood with its data like the number of beds, baths, garage, porch etc, you get the idea, and you also provide the prices which is dependent on the other things we just mentioned. You might be thinking why would I do that when I can take an easy route and would just ask a realtor, and you are right, I would do the same, but for the sake of learning, let’s assume it. So the idea is that the model will learn from this data to predict the price of the red house. This is called Supervised Learning, where the data is tagged (or technically, labeled) and the model learns from it

What if you own a business and want to know the types of customers you have? You can make the model learn similarities and make groups (technically, clusters) based on the data and be more strategic in your business by sending different offers to different types of customers! Do you remember the story published by Forbes in 2012 called ‘How Target Figured Out A Teen Girl Was Pregnant Before Her Father Did’? This is called Unsupervised Learning, where we are grouping the ‘types’ using the untagged data.

Have you ever used the Photos app on your phone (iOS/Android), where you tag the face of the person in one picture and it brings all the pictures taken of that person in one place? This is Semi-Supervised Learning, where the model learns from a small set of tagged data. 

What if you want to make your own car that can drive itself? You want to reward the model when it makes the right decision and punish it when it mistakes. Remember the Boston Dynamics Atlas, which makes a backflip? Here’s an animated video of a car learning to drive using this method, if you’re interested. This is Reinforcement Learning, where the model learns to maximize the reward and avoid punishment.

If you want to get into technicals and learn about more methods, refer to this article that talks about many more that are not covered in this article.



Check our other posts