Generally speaking, artificial intelligence (AI) is the human-like intelligence that is exhibited by machines or computers. AI can weigh the pros and cons of complex decisions and tell you which one is best, without you having to do all the thinking.

Similarly, it can learn new things without much intervention, and understand spoken language as well as the text of a book, just like an intelligent human can. Basically, AI can holistically assess any situation it finds itself in, and make mental calculations/decisions like a human being — sometimes even better, as we will see later.

The chess playing computer software is making use of AI, and so is the automatic vacuum cleaner. As you can already tell, AI has vast applications in nearly every field. Using AI, we have already automated much of the stuff that only humans could ever do. Now a computer can play chess better than humans, for instance. This implies that we have instilled some of our human intelligence into the AI!

How is AI transforming the world around us?

Thanks to recent advances in AI, we now have semi-autonomous self-driving cars that can manoeuvre roads and turns, as well as being able to drive in traffic!

Human beings don’t drive for fun (unless they happen to be Formula 1 or NASCAR drivers!), and it makes sense that such a tedious thing should be automated. Moreover, as older people who have vision problems are not fit for driving, they have to rely on others to take them places. With fully autonomous self-driving cars, this problem would be taken care of in the future.

In the field of education, AI has been employed to successfully create adaptive assessments for students, leading to individualised solutions tailor-made to address the weak-spots in each student’s knowledge (which would not have been possible for a human being to do single-handedly).

This is important because no two people are entirely identical (even identical-twins are bound to have some differences), and a single strategy cannot be used for all students. Personalised assessments in this case would help level the playing-field.

Although large-scale application of such research is rare, in the near future, we can expect these to be incorporated into the mainstream system of education.

There have also been experiments with AI that aim to have it analyse and understand the text from a book, and then to create new questions based on it. It’s arduous work for a human, but unlike human beings, AI doesn’t get bored or tired. Some of the experiments have yielded promising results, but still need future research before being made into commercially available solutions.

Once this technology becomes viable, AI can scan a chapter in your textbook and create questions from it without you having to do anything! It can create entirely new questions without having to repeat old ones. This can give students the ability to drill themselves in areas where they are not confident, which is actually the key to high performance in all areas of life: identifying and practicing the areas in which you are weak.

In addition, AI can label photos and recognise the objects/people in them. Its capability to distinguish people has been used by security and border control agencies to identify and arrest known criminals. At concerts, AI based facial identification systems have been employed to stop and/or track known stalkers of famous people. You wouldn’t have ever thought of AI taking the place of human detectives, would you?

There are also various apps that aim to identify and label any pictures you take with a smartphone. Gone are those days when you’d have to consult books and encyclopaedias to figure out the species of that exotic bird with that peculiar beak that you saw the other day. Just point your phone at it, and the AI-based app would identify it for you!

Perhaps the most familiar example that many of you will recognise is Facebook. It uses AI to tag you and your friends in the photos you upload. Although you can opt out of it, the feature is there if you want to use it.

But how does AI achieve such amazing feats? How does a self-driving car avoid obstacles and accidents? How does face-recognition technology work? How can a computer create questions directly from the textbook? How on earth can a machine make better diagnosis than a radiologist looking at the same x-ray? Yes, these are all true AI ‘miracles’.

Although the exact science (more like mathematics!) of AI is rather complex and cannot be easily explained, below we will see how some of the AI marvels actually work. We will do this by exploring different types of learning used in the creation and ‘training’ of AI.

How does AI learn?

AI is a very broad term, and there are many overlapping sub-disciplines and techniques (e.g., machine learning, deep learning, etc) that fall under its umbrella, but to avoid sounding too abstract and getting bogged by too many intricate details, we will simply list the main types of machine learning methods here:

1. Supervised learning

This is often used in image classification tasks: We present pre-labelled images to the computer software and ‘make it associate’ each image with a word. How the association happens is beyond the realm of this article. However, for the curious souls out there, there are some additional learning resources near the end of this piece.

Let’s say we are trying to create an AI-based software that can recognise images of cats and dogs. There are: Image 1 – Cat; Image 2 – Dog; Image 3 – Cat; Image 4 – Cat; Image 5 – Dog … Image 1001 – Dog.

After presenting the software with this labelled data of images, we can expect the software to identify and classify a new image into categories of cat, dog and neither. How the software ‘learns’ depends on the algorithm that was used in its training. It will also give us a “confidence percentage”, which basically means how sure the computer is that a given image (one it has never seen before) is a cat, dog, or neither.

2. Unsupervised learning

This is often used to see patterns that might be present in a data or in collections of things. The aim is to group similar things together. The things can be anything/any combination of the following: height, weight, colour, etc.

If there’s a pattern in the data or images, this method would display it without us having to tell the software that there’s supposed to be a pattern.

Using the similar examples of images of dogs and cats, we’ll give the images to the computer, but this time we would not tell it that the images are distinct (i.e., there are cats and dogs in the images and that these are two different species). By employing unsupervised learning algorithms, we can expect the computer to ‘learn’ that there are two categories of images. The computer itself would not label the categories as such (i.e., as cats and dogs), but it will distinctly tell us that the images contain two categories: Category 1 and Category 2.

There’s a third form of AI called Reinforcement Learning (RL) which is based on the basics of the psychologist B.F. Skinner’s ‘operant conditioning’: using rewards and punishments as ways of promoting and demoting behaviours. In RL, rewards and punishments are, of course not real-world things or money but numerical in nature, as we cannot motivate computer software using food or money, can we?

Will humans become subservient to AI?

Some might tell you that there’s a danger of AI overtaking us humans (i.e., the robots/computers becoming more intelligent than humans), in which case we’d become slaves to the tyrant ‘AI-master’. In other words, surpassing human intelligence, the AI would become the most intelligent species.

Will AI spell the end of humans?

Many prominent people including the late well-known physicist Stephen Hawkings, as well as Elon Musk of Tesla, have voiced these concerns. Hawkings even famously said that AI “could spell the end of the human race.”

However, AI would very much be like fire: a good servant but a bad master. Bill Gates compared it to nuclear energy, calling AI “both promising and dangerous.”

Generally speaking, as long as we remain vigilant and cautious, we should not have to face any apocalyptic scenarios similar to those seen in the movies TheTerminator or Matrix.

Researchers are positive that in the near future of AI, we can expect fully self-driving cars, nearly human-like virtual or robotic assistants, as well as more autonomous and individualised learning solutions for classrooms (but do not worry: the teacher would still be there. It’d be very long before a teacher can be replaced!). In the medical field, there will be more automated diagnosis of diseases, and less reliance on diagnosticians.

If your interest has been piqued, try looking up Scott Mayer McKinney of Google Health. He, along with some colleagues, published a paper which concludes that the AI algorithm developed at Google Health can diagnose and classify a mammogram/X-ray (whether the mammogram shows cancer or not) better than a radiologist! Would we not want an accurate diagnosis of such diseases as early as possible? You bet we would!

Now that you have this new knowledge about AI and have learned ways of further exploring it, what are you waiting for? Go learn, experiment and make your mark! Who knows you might end up being the next Einstein in the field of AI!

Want to learn more about AI?

Khan Academy:

For a topic as ‘hot’ as AI, there are many resources. One of them is the website of Khan Academy (KA) where you can find articles about machine learning, algorithms and other computing related topics. KA is a good resource to start your AI learning journey, but the articles might use some technical terms without explanation as they are part of an advanced course. Nevertheless, the website offers an excellent introduction to computing and algorithms; and what is AI if not advanced mathematics and complex algorithms?

Machine Learning for Kids

This is a website for kids where you can test various classification tasks, e.g., putting images into categories. Without knowing how the website works behind the scenes, you can play with the classification tasks and familiarise yourself with the basics of machine learning. The instructions given are easy to understand.

The website, at times, is not very responsive, neither is its interface very eye-catching, but you can still have an enjoyable learning experience.

https://machinelearningforkids.co.uk/

Looking for a more advanced resource?

AI is an exciting field to be in, and according to the authors Russel & Stuart, who have written an excellent and famous textbook (Artificial Intelligence: A Modern Approach) on AI: “A student in physics might reasonably feel that all the good ideas have already been taken by Galileo, Newton, Einstein, and the rest. Al, on the other hand, still has openings for several full-time Einsteins and Edisons.”

Although the textbook is most often used by university students, the authors have made it accessible to laypeople by explaining all the technical terms as much as they can. They use everyday examples to make sure that abstract concepts are substantiated and explained such that everyone can understand them.

Published in Dawn, Young World, July 24th, 2021