AI for Beginners

The basics of how AI works, and how it can be used

Akshaj Darbar
The Startup
Published in
8 min readSep 18, 2019


When I say Artificial Intelligence (AI), what comes to mind? Chances are that it’s something about robots taking over the world.

That’s what I thought of the first time I started learning about it just a few years ago. But as I dove deeper into how AI really works, on both a programming and even somewhat of a mathematical level, I realized that AI is nothing like that, and yet so much more.

Artificial Intelligence: the art of making computers that behave like the ones in movies.

First, let’s clear something up. What most people think of is Artificial General Intelligence (AGI). It can do everything that humans can do, but even better. But the truth is that we’re not even close to developing one. AI right now comes in the form of Artificial Narrow Intelligence (ANI), which is highly focused. You train it to do a few specific things, and it will train itself to do them really (and I mean really) well. But give it another task, and it will fail miserably. (Believe me. I’ve tried).

To make a program that can work on multiple tasks, you would need to write multiple different AI algorithms into the same program. For an AI that can hypothetically take over the world, this would mean lots of AI algorithms to be coded, which would not only take a long time to write, but also take MASSIVE amounts of processing power to train and run!

But does that make AI any less cool? Not at all! Even if we can’t make a program that can do everything well, we can still make programs that do a few things really well.

How? Let me explain.

Machine Learning

So what makes AI so smart? The magical process of machine learning.

In normal programs, the user directly inputs the pattern that the program should use to make decisions. In AI, the machine learning process allows the program to recognize the pattern by itself.

Here’s an example.

Imagine a dataset that provides the average temperature on specific days, and asks for the user to classify each day as freezing, cold, warm, or hot. If someone was to code the program themselves, they would hardcode it so that if the temperature is 0 °C or lower, it would classify it as freezing, and so on and so forth.

On the other hand, if the same thing was done in AI, the user would provide a dataset with the temperatures and their correct classifications, which would train the AI to determine the ranges of temperature in each class, by itself.


By updating values of different constants, known as weights and biases, continuously, and then testing its prediction capabilities with the current constants.

You can imagine it like a ball rolling down into a ditch. When it’s at the very top, it will start rolling down the hill slowly. However, as it goes down, it will increase in speed, until it reaches the very bottom. Once here, it will continue going and might start going up the hill on the other side, because of its speed. But as it gets pulled by gravity once again, it will slow and start coming back down to the bottom. It will continue doing this, slowly losing energy, until it finally settles at the bottom of the ditch.

In AI, the ball is the value of the constants, and the bottom of the ditch is their ideal value. As the program trains more and more, it will bring the constants closer to their ideal value. But as it continues going, it will likely overshoot the ideal value. The program would then start changing the constants in the opposite direction. It would keep doing this until the ball has finally settled on the bottom, at which point the constants have the ideal value.

During the decision-making phase, the weights and biases throughout the algorithm are put into an equation (which varies from model to model), which leads to a final output that the program uses as it’s decision.

So, let’s do a recap.

  1. AI can’t take over the world
  2. It is highly focused at performing specific tasks
  3. It uses a process known as machine learning to recognize and learn patterns in its data

Now that we know how AI works on a beginner level, let’s talk about how it can be used.


Artificial intelligence can be applied to essentially every field in our society today. This extends from medicine to autonomous cars, to even military use. Here are two applications that professionals are working on using AI for right now.

  1. Medical Diagnosis

Today, when you go to the doctor with a problem that can’t be easily diagnosed, there is a chance the doctor might send you to a blood-testing lab or a radiologist.

There, they would take your blood sample, or take MRIs, X-rays, and other forms of images of your body, after which you would be sent home.

In about one or two weeks, your doctor would receive the image or the test reports, after which they would begin analyzing the reports. In about one more week, the doctor would call you in, with a possible diagnosis in mind.

Overall, this is way too much time. For someone who has a broken bone, or is slightly sick, it might be fine. But in hospitals, where patients are in a serious condition, even a minute could mean the difference between life and death.

Enter AI. What if, instead of taking a few days to come up with a certain diagnosis, you could come up with the same diagnosis in a matter of minutes? By determining the cause of the problem quickly, the program would save essential time required to save the patient, allowing professionals to begin treatment earlier. This could involve detection and treatment of cancer at an early stage, or allowing doctors to determine the exact location of a tumour or injury in a patient’s body, which they can then repair through surgery.

Another problem AI solves is that of misdiagnosis. Today, 10% of all cases are found to be misdiagnosed, many of which lead to the death of the patient due to lost time, or complications from the incorrect treatment. In the US, medical error is currently the third leading cause of death.

AI is capable of drastically reducing the number of incorrect diagnoses in patients, potentially saving millions of lives around the world. Even today, it has been found to outperform clinical professionals in many areas, such as many forms of cancer diagnosis. As AI gets stronger, it will soon outperform professionals in diagnosing all diseases.

If [AI’s] not already the world’s best diagnostician, it will be soon — Andrew McAfee, MIT Scientist

However, the use of AI in diagnosing patients also comes with many ethical problems. First, what if the program was wrong? Who’s to blame then? Similarly, what if it encounters something it’s never seen before? A surgeon would try to find other sources to help them decide on a diagnosis. AI would just classify it as the most similar disease, which could have completely different treatments that won’t work on the patient.

Finally, there is the problem of AI taking jobs. If it can perform much better than doctors, what’s to say it won’t soon replace doctors completely?

Until we have ways around problems such as these, AI can’t be introduced into the medical environment and must be made stronger and smarter at performing its function, and laws must also be made to ensure it does not take over our jobs.

2. Autonomous Driving

We’ve all seen movies where the character gets into a car, tells it where to go, and the car drives them there by itself. Well, the day where a normal person can do the same isn’t far.

Many companies, including Uber, Google, and Tesla, are already developing and testing self-driving cars. This year, Tesla actually announced that it had successfully developed full-self driving cars and that all Tesla cars released following this year would have the self-driving system.

But how does self-driving work? The main task when developing a self-driving car is to create an AI that can perform multiple tasks at once:

  • Connect to a GPS to determine the fastest path to its destination
  • Pay attention to different traffic laws (stopping at red lights and stop sign; slowing down in neighbourhoods, etc.)
  • Be aware of its surroundings. The AI must know when pedestrians are crossing the street, where other cars are, how far the stop sign is, etc.
  • Make quick decisions in dangerous conditions

Being able to develop an AI that performs all those tasks effectively is difficult, which is why fully-independent systems have still not been introduced.

Along with the difficulty in making such an AI comes the ethical issues. First, at this stage, AI isn’t powerful enough at performing these tasks. There have been numerous cases of self-driving cars making incorrect decisions and putting their owners in dangerous situations. In May 2016, 40-year old Joshua Brown lost his life in a fatal crash after his Tesla self-driving system failed to stop as it ran into a trailer crossing the road.

Similarly, many people have raised concerns about the lack of any form of emotion in the systems. Imagine this. You’re driving in your self-driving car, and a truck in front of you suddenly stops. You are going too fast to be able to stop, and there is a family of pedestrians on the sidewalk beside you. In that case, what does the AI do? Does it swerve into the family to save your life? Or does it sacrifice you, its owner, to save the life of the family?

Until we have developed systems that can deal with situations like these, it isn’t possible to make completely reliable self-driving cars, and drivers must still be ready to take manual control of the car at any moment.

There are hundreds of other applications for AI, from space exploration to quantum computing, to business management.


Here’s one last recap:

  1. AI cannot take over the world
  2. AI has capabilities only in specific tasks
  3. It uses machine learning to learn how to make decisions
  4. It can be applied in medical diagnosis to reduce the number of misdiagnoses and the amount of time
  5. Self-driving cars can use AI to make them completely independent and reliable
  6. There are still many ethical issues surrounding AI, such as their lack of emotion, and their relatively low experience when dealing with problems

Thus, AI is set to change the world in many ways, with numerous effects on society (except for taking over it). As the different models of machine learning are slowly developed to be more powerful, AI will finally be able to accomplish more complex tasks with full independence.

AI is a “core, transformative way by which we’re rethinking how we’re doing everything.”

— Sundar Pichai, CEO of Google

Till then, we will just have to wait and hope that no one makes a program that can take over the world.



Akshaj Darbar
The Startup

Incoming MD Student at McMaster University. Researching blood cancer detection.