A Quick Introduction to AI

Original article was published by Ryan Shen on Artificial Intelligence on Medium


A Quick Introduction to AI

By now, you have probably heard the buzzwords: artificial intelligence (AI), machine learning and deep learning. But what do they mean? In this short guide, you will learn the basics of AI, the different subfields of AI, and the future outlook of the industry.

Credit: PCMag

What is Artificial Intelligence?

In general, artificial intelligence is getting machines to simulate human intelligence. This means machines can solve problems in the same way a human would. To understand the changes AI can bring to the future of humanity, we need to know the basics. There are three different types of AI. Each type corresponds to a different level of intelligence embedded in the robot.

  1. Narrow AI (A.K.A weak AI) — When a machine is designed to perform a single task and is extremely good at performing that task but not others. For example, driving a car or playing chess. This is also the only stage that technology has currently achieved.
  2. General AI (A.K.A strong AI) — When a machine can do any task that a human can.
  3. Artificial superintelligence — When a machine can do any task better than a human.

Why is AI so important?

Artificial intelligence can, in theory, do all the tasks that humans can do. AI has affected almost every part of our lives, from healthcare to human resources and from voice assistants to video games. It would be easier to just name what hasn’t been affected by AI. AI has been around for a very long time. The term “artificial intelligence” was coined in 1956 — over 60 years ago! At the time, people were interested in AI because machines never get tired, have no emotions or distractions and do not make silly human mistakes. This meant that machines could do the work of humans more effectively, efficiently and at a lower cost. However, it was held back by the weak computational power and the smaller amount of data. Now, with the exponential increase in computational power and the rise of big data, AI has begun to take off! Not only are the most successful companies in the world using AI, but AI can be applied to any sector to create new possibilities and efficiencies. One example is healthcare, AI has been learning to detect cancer and in many cases, it is better at spotting cancer than most specialist doctors are. AI is also currently being used to create autonomous vehicles that could bring car accidents down to virtually 0. The possibilities with AI are endless along with the benefits it could bring society.

Credit: IDTechEx

Different subsets of AI

Artificial intelligence is split into many different subtopics. Each subtopic is called a subset of AI as they are related to the main idea of AI but have different methods of implementing these ideas. This will be a quick overview of the most general AI subsets: machine learning, deep learning and reinforcement learning.

Machine learning

Machine learning is a component of AI, but machine learning and AI are not the same. Machine learning is based on the idea that machines can learn from experience while AI, more broadly, refers to machines being able to act smartly. Machine learning trains on past data to predict the outcomes of future data. What is special about machine learning is that it does this without being explicitly programmed — the machine finds the patterns itself without a human telling it what to look for.

Imagine you wanted to create an AI that could differentiate between cancerous tumours and benign tumours. Using machine learning, you would be able to show the machine pictures of tumours that are cancerous and then show it pictures of benign tumours. The AI would then begin to recognize subtle patterns that are related to cancerous tumours or benign tumours. The more data you can provide, the more patterns the AI can recognize and the better the AI gets at differentiating between the two.

Machine learning is especially useful due to its ability to process HUGE quantities of data, analyze it, and then also find correlations within seconds. To summarize, the machines learn through examples, not instructions. Currently, machine learning is being used in product recommendations, stock market trading, search engines and many more.

Reinforcement Learning

Reinforcement learning is a subset of machine learning. The method is the same as machine learning — letting the machine train itself. However, the key idea of reinforcement learning is to use rewards to train AI.

Imagine your training a dog. When he is a good dog, he gets a reward and when he is a bad dog, he gets punished. Over time, your dog learns to be better to get more “good dogs”. The same process applies to reinforcement learning, it will try to make choices to maximize its reward. Reinforcement learning has been used to beat chess champions in chess, win against professionals in video games and create breakthroughs in robotics.

Credit: IBM

Deep learning

Deep learning is also a subset of machine learning. This means the machine also trains itself, the key difference is that deep learning imitates the inner workings of a human brain in data processing and pattern making which involves neural networks. Neural networks are modelled after the brain. The name neural is from neuron which is a cell found in our brains. Neural networks are usually made up of many neural layers. Each layer is made up of several simple, connected processing units called nodes that are like the neurons in our brain. Deep learning contains multiple neural layers in each network. It finds patterns in data by breaking the data down into smaller pieces that are given to each neural layer.

For example, say you wanted to find an eye from a picture of a face. The first neural layer would look at the pixels. This information is passed onto the next neural layer that may look for edges formed by the pixels. This goes on and on until the final neural network can reproduce a picture of an eye.

By breaking the data down, deep learning can learn increasingly abstract concepts. Deep learning has many applications such as self-driving cars, voice assistants and image recognition.

Success in creating AI would be the biggest event in human history. Unfortunately, it might also be the last, unless we learn how to avoid the risks.” — Stephen Hawking

Problems with AI

AI is not a miracle that will save all of humanity. Currently, some of the biggest problems with AI are biases in data and privacy.

Humans are full of biases. These biases are then passed onto the data that is collected on us. When AI gets trained on the biased data it creates a biased AI. For example, Amazon had a hiring algorithm that favoured applicants based on words like “executed” or “captured” which were usually found on resumes of males, even though sensitive variables such as gender, race, and sexual orientation were removed.

Privacy is also a major concern. AI needs data to improve and that data is often private and personal. As AI gains more progress, more data will be required. But how much of your data do you want to give up? It is hard to find a balance between improving AI and keeping your privacy. How would you feel if your voice assistant was recording all of your conversations and uploading them to a third party just to train its AI?

Credit iphoneincanada

Future

The future of the AI industry is bright. Pre-pandemic, Gartner predicted that the business value created by AI would reach $3.9T in 2022. Recently, a poll conducted by Gartner shows that 47% of artificial intelligence investments were unchanged since the start of the pandemic and 30% of organizations plan to increase their AI investments. Also, IDC forecasts worldwide revenues surpassing $300 billion in 2024 with a five-year compound annual growth rate of 17.1%. As all the big tech companies begin investing heavily in AI there are sure to be great improvements in the road ahead. Researchers are also working towards creating artificial general intelligence, some scientists believe that it is only a few decades away! With AI inevitably intertwined with our future, it is important to be learning about it now.

If you enjoyed this article or found it helpful feel free to clap it!

Key takeaways:

  • AI means getting machines to simulate human intelligence.
  • Ai is important as it can do all the tasks that humans have to do more effectively, efficiently and at a lower cost.
  • Machine learning is the idea that machines can learn from experience without being told specific instructions.
  • Reinforcement learning is all about using rewards to train AI
  • Deep learning is modelled after the human brain.
  • Some of the major problems with AI are biases in data and privacy
  • The AI industry is expanding every year and now is a great time to start learning about it!

Resources for further learning: