AI for School Teachers

Rose Luckin, Karine George, Mutlu Cukurova

• Editorial: CRC Press
• Año: 2022
• ISBN:
  9781000543728
• Edición: 1
• Categoría: Computer Science
• Categoría: Computer Science General

What is artificial intelligence? Can I realistically use it in my school? W hat's best done by human intelligence vs. artificial intelligence, and how do I bring these strengths together? What would it look like for me, and my school, to be AI Ready?

AI for School Teachers will help teachers and headteachers understand enough about AI to build a strategy for how it can be used in their school. Examining the needs of schools to ensure they are ready to leverage the power of AI and drawing examples from early years to high school students, this book outlines the educational implications and benefits that AI brings to school education in practical ways. It develops an understanding of what AI is and isn't and how we define and measure what we value and provides a framework which supports a step-by-step approach to developing an AI mindset, focusing on ways to improve educational opportunities for students with evidence-informed interventions.

1 What is AI and why MIGHT AI be useful in education?

DOI: 10.1201/9781003193173-1

How can AI benefit education? What is AI and how can I use it effectively? What sort of AI do I need? These might be some of the questions you are asking yourself. Before we dive into these questions, let’s take a moment to get clear on what AI is and why AI might be useful. We will also try to enthuse you about AI’s potential for use in education.

What is AI?

The term “Artificial Intelligence”, abbreviated to AI, is not something for which there is any single accepted definition. Many scientists disagree about exactly what the precise definition of AI should include. As this chapter will explain, there are two main types of AI: machine learning and Good Old-Fashioned AI (GOFAI). Some people believe that only machine learning should be called AI, but many others believe that the definition of AI should also include tools and technologies that make intelligent decisions in other ways. We have selected a simple definition of AI in an attempt to include the vast array of different sorts of tools and technologies that could be considered to come within the bounds of the phrase “Artificial Intelligence”. The definition we have selected is taken from the Oxford English Dictionary:¹

The capacity of computers or other machines to exhibit or simulate intelligent behaviour.

A short history of AI

Machines that could behave, but not intelligently

With a definition in place, let’s look back at where AI has come from. For many centuries, humans have been intrigued by the task of creating representations of living creatures, including humans. These representations are often referred to as automata and they date back to the Middle Ages, possibly even earlier. In the 19th and early 20th centuries, automata reached the height of their popularity. From bears that could turn somersaults to magicians who could see another automata in half, and nightingales that sang in golden cages, these party pieces were increasingly intricate and could perhaps be considered forerunners of AI.

Or perhaps more accurately, we might consider them to be the forerunners of the field of cybernetics, which is a scientific area of study that explores the control and communication that happens in animals and in machines. The study of cybernetics was started by Norbert Viner in the mid-20th century and is still very much at the heart of what robots can do. These cybernetic robot predecessors were more mechanical than intelligent, but their creation influenced the field of robotics that evolved. Even today, not all robots are intelligent; some are just labour saving through their speedy completion of mechanical, repetitive tasks. However, many robots are also intelligent and are part of the human desire, played out over time, to create objects that can behave in intelligent ways.

We also love to tell stories about objects that behave intelligently, and robots have long been a love of the film-making industry. Who cannot but be endeared by robot characters like Star Wars’ C-3PO,² or WALL-E,³ feel fearful of Blade Runner’s Roy Batty⁴ or The Borg,⁵ and simply be amazed by Ava in Ex Machina!⁶ The reality is nothing so dramatic. It is certainly true that intelligent software and robots are a reality, but none of them have the all-round capabilities of their movie-star peers. The ability of AI systems to achieve more than one area of expertise is still mainly a fantasy. From tangible robots to invisible software, AI is a specialist operator with no ability to transition from one area of expertise to another. A self-driving car cannot play chess; a surgical robot cannot drive a car. It’s worth noting, however, that a human surgeon can likely drive a car and play chess and a great deal more besides.

But let us return to our brief look back at the history of AI.

Another important person in the history of AI is Alan Turing who, in 1950, wrote a famous article titled “Computing machinery and intelligence” in which he posed the question: “Can machines think?”⁷ Alan Turing was a mathematician and code breaker at Bletchley Park during the Second World War. Turing proposed a clever test that could be used to decide if a machine was thinking and, therefore, was intelligent. This test, which is called the Turing Test, presents the proposition that if a computer can fool a human into believing that it is really a human, then that machine deserves to be called intelligent. This thought experiment captured the interest of many scientists and helped to progress the birth of AI. Indeed, there are still Turing Test challenges today in which computer scientists pit their AI against one another to see whose system can convince the most people.

The ten men who gave birth to modern AI

Following the publication of Turing’s famous article, the field of AI evolved at a rapid pace, and in 1956 a momentous meeting took place at Dartmouth College in New Hampshire, in the United States.⁸ A ten-man group of scientists met with the aim of studying human intelligence in all its richness and from all aspects. Their goal was to be able to describe each feature of human intelligence so precisely that a machine could be built to simulate it. The scientists believed that they would be able to make significant advances towards their goal over the period of a summer.

They soon discovered that human intelligence was far more complicated than they had understood and progress during that summer was small. Nonetheless, the occasion of this meeting was extraordinarily significant, as it gave birth to what we now recognise as the scientific discipline of AI.

It is interesting to wonder had the ten scientists been from more diverse backgrounds, if their group would have taken AI in a different direction. We will never know. What we can be sure about is that the lack of diversity amongst those who work in AI is an ever-present concern. We hope that by making AI accessible to more people, a more diverse population might become interested in working with AI.

From that date onwards, the task of creating computer programs that behaved in intelligent ways was the cutting edge of science. At this stage, AI was not focused on robotics, but on developing software that could enable computers to interact intelligently. Early attempts were very simple. Systems such as ELIZA,⁹ a computer program that played the role of a psychotherapist, were text-based and required the person playing the role of the patient to type out their problems and questions. The ELIZA software was programmed to look for keywords in what the patient typed. When a keyword or phrase was found, the software triggered a stock answer template and ELIZA offered her advice in the text on the computer screen. It is hard to believe that anything so crude could fool anyone into consulting with ELIZA for more than an initial sentence. But several people were duped by ELIZA, at least for a while.

Good Old-Fashioned AI and expert systems

If those ten scientists who met in New Hampshire are considered the fathers of AI, then maybe ELIZA should be the mother.¹⁰ The important thing about systems like ELIZA is that it identified a particular approach to simulating intelligent behaviour. This approach is called production rule-based pattern matching and ELIZA “gave birth” to many similar systems over the decades following her inception in 1964. In fact, the production rule-based systems that evolved from ELIZA became sophisticated enough to accomplish advanced activities such as diagnosing an illness from a set of symptoms and suggesting the treatment regime based on these symptoms. These systems are referred to as expert systems and were used in a variety of different fields, such as medicine.

The pinnacle of the GOFAI movement came in 1997 when a system built by IBM, called Deep Blue,¹¹ beat the then chess grandmaster Gary Kasparov at the game of chess. This was extremely impressive and marked the high point of this phase in AI’s history.

The problem with these Good Old-Fashioned AI (GOFAI) systems was that the actions the AI was able to perform had to be pre-programmed into the software when it was written – a huge task. For games like chess, there may only be a certain number of moves that each particular chess piece can take, but there are many millions of iterations that the combination of these moving chess pieces can create. In fact, trying to look ahead for just two moves in a chess game would generate 1,225 possible chessboard states. Looking ahead 20 moves will generate 2.7 quadrillion possible board states.

To write a computer program to find ways of dealing with the existence of all these possibilities and to be the best in the world at it too is no mean feat. However, there is a severe limit to the intelligence that this style of AI could achieve. Once the knowledge was written into the computer program code, the system could not be updated without going back and changing the code. No matter how many disease cases they diagnosed, or gas pipe fractures they identified, or games of chess they played, GOFAI systems will never improve.

Before we confine GOFAI to the history books, however, it is worth thinking about the uses it still affords. As teachers, we can easily see how a GOFAI system could still be extremely useful in the classroom. For example, planning a school trip involves many possible steps, as illustrated in Figure 1.1. An easy-to-use app that helped teachers step through all these processes and decision points could be extremely helpful and it could be built using GOFAI.

Figure 1.1 An example of the steps involved in planning a trip. (With permission from Paul Quinn, Acting Principal, Watford UTC.)

It is also worth taking a moment to note that there were suggestions that production rules of the sort used to create systems such as ELIZA and her “offspring” could represent the basis of human thought. For example, the work of John Anderson who developed the adaptive control of thought (ACT) theory of human thought as a set of production rules.¹² As we judge the wisdom of his work, we should remember that we know a great deal more about human thought now than we did at the time that John Anderson was developing his ACT theory. We should also remember that his work precipitated a generation of cognitive STEM tutoring software that was extremely successful at tutoring students in maths and science. Indeed, these cognitive tutoring systems are the basis for the systems that are sold today by the spin-off company from Carnegie Mellon University called Carnegie Learning.¹³

In truth, the Carnegie Learning example was not the zeitgeist in the late 20th and early 21st centuries. From the zenith that was Deep Blue came several decades where AI moved at a much slower pace than it had in the years immediately following the Dartmouth College meeting. The technical limitations of machines that could not learn severely restricted what could be achieved with AI, and funding was less available. The deepest of the AI winters to date was upon us.

Machines that can learn

It is always darkest before the dawn so they say, and the AI winter was indeed quite a gloomy time to be working in AI. Progress in AI was slow as we said farewell to the 20th century, and the new millennium dawned. Despite the coolness towards AI, the drive to produce AI systems that, like humans, could become better and better and better at a particular activity was a strong motivator for many aspiring computer scientists. In 2011, Google formed Google Brain and thus emerged the...