top of page

8 things even your grandma should know about AI


AI and robots are not the same. No matter how cute. Photo by Alex Knight on Unsplash


From online dating to ordering shopping online from Alexa, to being given suggestions by Netflix of shows we might like, we, the human race, are increasingly interacting with AI all the time.


The intelligence of machines is developing exponentially according to experts in the field. Due to this explosion in machine intelligence, it’s important for everyone — from your grandma to a 5th grader — to know the basics of AI. This is not only because we use so many different applications of AI in our daily lives, but because the influence of AI on the future of humanity is only going to get greater for all of us at current pace.


We, the users, are a crucial part of this grand experiment, with the majority of us increasingly spending more of our work and personal lives online and uploading content and data to a wide range of services and websites.


1. What does artificial intelligence mean?

Artificial intelligence refers to the intelligence of machines. Artificial intelligence is often viewed as machines that are capable of learning and problem solving, similar to a human or an animal’s brain. Unlike a human or an animal, an incredible amount of programming is required for machines to recognize objects and understand how they relate to other variables but the capacities of machines to do these types of tasks — object recognition, facial recognition, gesture recognition — is increasing rapidly. Machine learning refers to


a programming approach in computer science in which the behavior of a program is not fully determined by the code but can adapt its behavior (i.e., learn) based on the input data (Brunn et al., 2020)

The term artificial intelligence was coined by American computer scientist John McCarthy. He organized a meeting in the summer of 1956, known as the Dartmouth Conference, that would later be viewed as initiation of AI as a field of science. Other recognized proponents included Alan Turing, Marvin Minsky, Allen Newell and Herbert A Simon, who collectively championed the approach known as “symbolic reasoning”. Alan Turing, known for many important achievements — such as code breaking during World War 2 — also invented the Turing Test in order to set the standard for an intelligent machine. He came up with the idea that rather than copy an adult brain, it would be better to simulate a child’s brain and then teach it to learn. Nifty programmers have been teaching machines to learn increasingly complex tasks ever since. Today, when applications ask us to verify if we are human users, we are asked to do the opposite of the Turing Test when we select pictures in the CAPTCHA test.


Want to read this story later? Save it in Journal.

How is artificial intelligence different from biological intelligence?


Biological intelligence refers to the somatic intelligence that characterizes human and animals, existing independently of our conscious control. Our hair grows, our cells repair, our blood becomes oxygenated whilst breathing. This means that we have multiple systems operating in parallel: our immune systems, cardiovascular systems, hormonal systems, muscular systems that are interconnected and self-regulating. Biological intelligence is highly complex and geared towards the survival of the species. Robots, in contrast, do not have the same systemic complexity.


2. What is an algorithm?

The algorithm is the buzz word of our current zeitgeist. The word algorithm originated from the works of 9th century Persian mathematician Muḥammad ibn Mūsā al-Khwārizmī. An algorithm refers a set of rules or set of instructions that defines a set of operations. A computer will follow these instructions to solve a given problem, there are many algorithms for each problem. Computers optimize then the find the most suitable algorithm for reaching a goal most effectively. However, not all instructions are the same as there are many different languages. Each language has their own way of organizing commands which are called syntax. Common programming languages include Javascript for game and app development, Python for AI and machine learning, R for statistics and C++ for game development and graphics.


3. What is symbolic reasoning?

In layman's terms symbolic reasoning refers to the ability to do the math. In other words, the cognitive ability to represent numbers and mathematical rules in an abstract manner (Landy et al., 2014). In the context of AI, a reasoning is an operation where rules and definitions are followed to produce new knowledge. This process can become automated when performed by an algorithm. When you type a question into google, symbolic reasoning allows the answer to come back so quickly. For example, I typed the question: what is is the healthiest food to eat for breakfast? Google gave me about 14,700,000 results in 0.90 seconds, the top result being eggs.


Symbolic reasoning was a large part of Cambridge Analytica’s massive analysis of Facebook’s ‘Knowledge Graph’ of people’s friendships and beliefs.


There is an important distinction between symbolic reasoning and deep learning methods. Symbolic reasoning is often considered Good Old Fashioned AI (see John Haugeland), and contrasted to robotics and deep learning methods. Symbolic reasoning (I keep trying to think of a more user- friendly term of this but let’s go with it) is based on logical reasoning, algorithms, and representation of knowledge as formal propositions. Deep learning is based on mathematical analysis of large data and patterns, sometimes using an approach based on the way neurons in the brain are wired together (propagating strengths of connections between them as a way of improving performance and learning).


4. What is deep learning?

Deep learning can be defined as “a technique of machine learning that involves multiple, hierarchically organized layers of artificial neurons”. Deep learning involves brain like structures called artificial neural networks that are given huge amounts of data and are trained to recognize patterns. The deep aspect of deep learning according to Jeff Dean is that these neural networks have multiple layers.


Neural networks typically have multiple layers that include an input layer, hidden layers, and an output layer. In the hidden layers, data from the input layer undergo many changes, multiple times (Brunn et al., 2020).


5. What has been achieved through deep learning?

There are a range of different domains that deep learning has achieved massive scientific and commercial achievements in practical application. One of the most advance examples is speech recognition. Many of us use speech recognition software to help with tasks in our daily lives, whether it is Siri on an iPhone or Alexa in your home, speech recognition software is becoming increasingly accurate. This technology has been applied for applications such as deafness and disabilities and in military contexts. An application that features heavily in the news is game playing. Deep learning enables computers to play and win at games are a much-cited example, with the computer ‘Deep Blue’ beating Gary Kasparov in Chess in 1997 and more recently Alpha GO beating Lee Sedol at the game of Go (see picture).



The Game Go, a Chinese board game that involves abstract strategy to gain control of the board


Deep learning requires access to data on a huge scale, without a lot of data, the AI is likely to be inaccurate. Thus, through processing medical records, it is possible for computers to detect the risk of a disease through identifying early risk factors and therefore there is much scope for applying deep learning in healthcare. Through the early identification of disease, it is proposed that the use of AI could reduce medical costs and in theory prevent disease. When Facebook when you would be asked to tag pictures of yourself or your friends, but now the name of the person is suggested to you? That again is a deep learning algorithm. From analysis of many pictures of your friends with their names and profiles, it is now possible for AI to recognize faces. Occassionally there have been bugs and the underlying architecture becomes clearer to the user.


6. What are neural networks?

A human brain cell is called a neuron. Human brains are highly sophisticated at processing information. We can effortlessly recognize numbers from writing on a page. Recognizing a written number on a page for a computer is much harder. To achieve this task, neural networks have been created to enable computers to learn things such as recognizing hand written digits. Neural networks have been used for the analysis of images, they can detect patterns and make sense of them. Input is taken into the network and then abstractions occur at multiple layers before the output is given.


7. What is reinforcement learning?

The idea of reinforcement originally comes from psychology. Pavlov’s experiments with dogs established the ideas of different types of conditioning and responses (unconditioned and conditioned stimuli and responses). Back in the 1930s Skinner established that animals and humans make an association between a behaviour and a consequence Early psychologists found that animals could be trained to do things in return for rewards. Animals, such as dog, or humans can be taught not to undertake a behaviour through a punishment. Reinforcement learning is also studied within economics, neuroscience and as relevant here, within AI.


In the context of AI, reinforcement operates on the same principle: algorithms can achieve a goal and obtain rewards too. Incredibly, the reward can be delayed and an algorithm can undertake a range of complex steps to achieve it. Computers will optimize to achieve as much reward as possible. Computers repeat actions where they are behaving in ways that increase winning and decrease actions that prevent winning. Thus, the agent or the computer has to learn that set actions led to winning the game. It is noteworthy that algorithms require a huge amount of data to learn how to win the game.


A comprehensive introduction to reinforcement learning in AI can be found here in this video by David Silver, who has combined deep learning with reinforcement learning in a program that learns to play video games from pixels.


8. What is the difference between robotics and AI?

When we talk about the rise of robots and the rise of machines, are we talking about the same thing? No.

“In short, a robot is a machine which may or may not require intelligence to perform specific tasks and has a physical form. Whereas an AI is a program so it doesn’t need to be physical.”

AI is a program that complete tasks that would otherwise be performed by a computer.

Whilst AI’s and robots are different things, there is a sweet spot in the middle, the artificially intelligent robots.


Final thoughts

This blog has conveyed 8 things about AI: What does AI mean? What is symbolic reasoning? What is an algorithm? What is deep learning? What has been achieved through deep learning? What is reinforcement learning and what is the difference between robotics and AI? The aim is to give a brief overview of some key questions.


Whilst AI is highly technical and an ever-expanding, ever-deepening field, it should not stop the general public from being aware of some of the key concepts, a small fraction of which has been covered here.











8 views0 comments

Comments


bottom of page