Understanding Artificial Intelligence (2021) aims to demystify the subject of Artificial Intelligence (AI) for everyone, including those who don’t have an IT or mathematical background. It will give you a basic understanding of how AI works and why sometimes it makes mistakes or offers imperfect solutions.

Introduction: Gain an understanding of AI and why it isn’t so intelligent after all.

These days, everyone seems to be talking about AI. But what is it? What are its limitations? Should we be scared of what it might be capable of?

The reality is that humans have always been inventors. In prehistoric times, we created harpoons for hunting – when agriculture came along, we created pickaxes, sickles, carts, and even tools to create tools. And then came machines, and some people were afraid of their rapid development. But those machines, too, were only tools.

And now there’s AI and things are moving up a notch. Well, forget what you’ve seen in science fiction. AI won’t replace or enslave us. It doesn’t have free will. AI is also just a tool. But what we must do is learn how to harness AI for good and not misuse it.

In this summary to Understanding Artificial Intelligence by Nicolas Sabouret, you’ll find out what AI is; why solutions provided by AI may not actually be the perfect ones, but will still be good enough for most purposes; and where the future of AI might be heading.

What exactly is AI?

So let’s clear something up right from the get-go: computers are machines. AI doesn’t make them intelligent. They still do no more than what we tell them to do.

Computers have come a long way, though. At first, they were just simple calculators dealing with numbers and math. Then they progressed from only numbers to dealing with words, then images, and then sound. Nowadays, we even have computers – in our smartphones, for instance – that can listen to our requests and convert them into actions.

They’re capable of all that thanks to algorithms. A simple algorithm that you used at school was the process you used to add large numbers together. You can also think of an algorithm as being like a recipe. Just as a cook knows how to follow a recipe, a computer follows the instructions in the algorithm to give you the required result.

In the early nineteenth century, Charles Babbage was the first to produce a machine capable of following algorithms. And by 1936, Alan Turing had shown that computers can, in theory at least, follow any algorithm, however complicated it might be.

So where does that leave AI, now? For starters, we probably should all be using the term AI programs rather than AI. All AI does is apply an algorithm written by a human to give responses that appear to be “intelligent.” Programmers also use AI itself to write programs using a technique known as machine learning. This is a bit of a misnomer and causes some confusion as actually, the way AI produces new programs is dependent on the quality of the data it receives. So, as has always been the case, the old adage “garbage in, garbage out” holds true.

So is AI really intelligent?

That’s a very good question, but to answer it, we first need to understand what intelligence is. And that isn’t as easy as you might think.

Is it the opposite of ignorance? Consider this: if you were asked when the city of Istanbul was founded, would you know? Most likely not. But does that make you unintelligent? And would you define Wikipedia as being intelligent because it can tell you the answer? (It’s the seventh century BCE, by the way).

What about the ability to do complicated math? Can you give the answer to 24,357 x 527, for example? You probably can, given time, but a simple calculator can do it much faster. Is that calculator more intelligent than you?

In reality, neither the calculator nor Wikipedia is intelligent. Computers may be capable of dealing with tasks associated with calculation and memory, but they’re not intelligent by human standards. After all, humans can reason based on past experience, make decisions in situations too complex to describe accurately to a computer, learn new skills, have ideas, and communicate using complex and abstract ideas.

So how can we assess the intelligence of a computer? Well, remember Turing? He came up with what is known as the Turing test. Here’s how it works:

A human is in one room and a computer with AI in the other. You can communicate with both using a keyboard and screen for each, but you don’t know which is connected to the human and which to the AI. A delay is built in to ensure you can’t tell which is which from speed, but only from the responses themselves. The test is to determine how close the AI can answer in a manner that resembles human intelligence.

Since 2006, there’s been an annual competition to determine the chatbot which comes closest to fooling the judges. The judges, though, are very adept at getting the chatbots to respond incorrectly, some needing only five questions to identify the AI correctly.

But the Turing test is, in any case, flawed. If we want to test the intelligence of an AI program, we should really test its performance in the task it has been developed to perform – playing chess, for example, not whether it can have a philosophical discussion about chess.

We should remember that in creating AI, we’re not trying to create an artificial human. It was Edsger Dijkstra, a computer scientist, who summed up the difference between human intelligence and AI rather neatly when he said, “The question of whether machines can think is about as relevant as the question of whether submarines can swim.”

Machines don’t think. We might think they’re intelligent – but they’re not.

What is an AI algorithm?

So now that you know that AI isn’t intelligent in the same way as a human is, let’s take a look at what AI algorithms are.

You might be surprised to learn that they’re nothing special. They’re just like the cooking recipe we mentioned earlier – a step-by-step approach to solving a task. But it’s taken some time and years of research to develop them. And there’s no standardized approach, either, which means there are many different AI algorithms. But they do have some common elements: to overcome the limitations in memory and processing capacity of computers.

To put this in perspective, a personal computer created in the 2010s can do billions of additions per second. And that number has kept increasing. But, if you need 10 billion steps, you’ll have to wait some seconds to get your result. If you need a trillion, you might have to wait 15 minutes. And for 100 trillion, you’ll need a day.

That 100 trillion might seem a lot, but the number of transactions needed can climb quickly. For instance, imagine a school head who needs to schedule students, classes, and teachers. If the school has just ten rooms with 15 classes, for each hour there are around 3,000 possibilities. For each eight-hour day, that means six billion billion billion possibilities – that’s six with 27 zeros. Considering each one of those with a computer, even one that processes a billion operations per second, will take far too long!

These computational limits are very important to AI and are known as complexity. We need to distinguish between the complexity of algorithms and the complexity of the problems the algorithms are solving. The complexity of an algorithm depends on how big the problem is and how much data is needed. The complexity of the problem itself is the minimum number of operations needed to solve it. The latter is often a theoretical number and other, less elegant algorithms with higher complexities are often found while looking for the best solution.

All a bit complicated? Well let’s cut a long story short: some problems have a theoretical complexity that’s so high that, assuming we could even write the best algorithms to solve them, the number of operations they’d require far exceeds what even a computer of the future could manage – even if it were millions of times faster.

So when it comes to AI, the algorithms deployed are sometimes less than perfect – a face recognition algorithm will sometimes make a mistake or a chess algorithm make the wrong move, for example – but often they give us the best possible solution in a reasonable amount of time.

How does AI come up with a solution?

There are many AI methods that can be used to solve problems. As we’ve already established, they’re not intelligent and they don’t always provide the best solution to the problem being addressed. Sabouret provides many examples of these methods but here, let’s examine a first principles AI method called exploration.

Imagine you’re in Alexanderplatz on a trip to Berlin and you want to visit Museum Island. You can follow the route on the map and make a note of street names and where you need to turn. You build the route using your intelligence. Most of us can do that, but some might find themselves hopelessly lost. Thankfully, these days, we have GPS to help us. But how exactly does that work?

Well, GPS determines your position on Earth by referencing 28 satellites and the time it takes for each of those to send a signal. With this information and a map of where you are, GPS can calculate a route for you. It does this using your current position, your final destination, and all points in between, by creating a “graph” of all the points you must pass through en route.

This might sound simple but it’s a bit more complicated for a computer; it has to consider all possible neighboring points as it progresses, storing in memory the others as it goes. An alternative is to spiral out from the starting point. Neither method is particularly efficient. So when the area that the computer has to consider becomes too great, it becomes almost impossible to solve the problem in a reasonable time.

This is where heuristics come into play. A heuristic algorithm is created that approximates the solution. In this case, it would be something like “go in roughly the right direction and stay roughly on the right path.” If we consider our recipe analogy, instead of weighing an ounce of butter, we’d probably just cut what looks like an ounce off our eight-ounce block. It’s not precise but it’s near enough.

So the route you’re given to get from Alexanderplatz to Museum Island may not actually be the best, but it’ll be good enough and get the job done.

Where is the future of AI heading?

You might be wondering if we’ll ever build a machine that’s truly intelligent, one that learns as a child does, is aware of the world around it, can feel emotions, and with which we can build a future.

It’s a difficult question to answer, but although some researchers believe so, Sabouret says that there’s no current evidence to suggest that it’ll be possible.

Back in the 1970s, John Searle coined the term Strong AI to mean AI that imitates the human brain perfectly. Now, we also use the term Weak AI for AIs that have a specific purpose only – such as winning at Go. They are, of course, anything but weak as they far outsmart your average human.

But turning back to strong AI, we can further subdivide this into general AI and artificial consciousness. General AI means creating a machine capable of resolving many problems over a wide range such as getting a university degree, passing the Turing test, and making coffee in an unfamiliar kitchen. Artificial consciousness is something else. Such a machine would be conscious of the world around it, and yes, even of the fact that it was a machine. But the jury is still out on how to define consciousness for a machine and what tests it would need to pass to be “conscious.”

Are either of these on the horizon? The honest answer is, Who knows?

There’s no doubt that AI is changing the world and machines will become more and more impressive. The rate of progress makes it difficult to know exactly what will come next. AI can already beat humans at Go and Poker but not at complex video games, for example. And although it may be able to access medical information quicker than a human, it can’t use that information to diagnose what’s wrong with a patient. And the human factor necessary for some tasks, like hiring people, Sabouret argues, is nigh on impossible to capture in any AI system and humans would easily be able to use “antagonist data” to fool these systems.

So should we be concerned that AI could end up enslaving humanity as we see in some sci-fi movies?

Right now, no, weak AI has been created for specific tasks, it can’t imagine things and it can’t create. Perhaps strong AI will exist one day but we’re nowhere close to creating artificial consciousness.

The biggest worry right now is the misuse of AI. For example, in a world where social media feeds have become the main source of information for some age groups, a totalitarian dictatorship could theoretically take control of what we’re allowed to know. Overnight. And although self-driving cars won’t attack their drivers unless programmed specifically to do so, there’s always the possibility that such technology could, one day, be used to create a machine that can identify and eliminate its target subject.

AI is already being used to commit crimes of a much lower order – to hack cyber security protocols, for example. So it’s clear that AI can be misused – but it’s not going to do that spontaneously, it has to be programmed to do so.

But whatever the future holds, there’s one thing we can be certain of: AI has brought humanity toward a better understanding of itself. We’ve first had to understand how a human does a task to find a way to do it better. Perhaps, as Sabouret posits, understanding ourselves better is AI’s greatest gift to humanity.

Summary

Although we may think that machines and AI are intelligent, in fact, they’re not. They’re only capable of doing what we program them to do. You’ve seen that by using heuristics, AI can provide a solution to a given difficult problem within an acceptable time frame. That solution may not be perfect, but it will usually be a “good enough” solution. And finally, you discovered that creating a machine that has consciousness isn’t currently on the horizon, but in the wrong hands, AI can be used for the wrong purposes. One thing AI isn’t going to do, though, is rise up and rebel against us.

About the author

Nicolas Sabouret is a Doctor in Artificial Intelligence and full Professor of Computer Science at Université Paris-Saclay, one of the highest ranked research universities in France. He has been teaching Artificial Intelligence for twenty years to undergraduate students, graduate students, engineers and PhD students. Nicolas has supervised many research projects in this domain. His goal is to help give people the keys to an informed thought on Artificial Intelligence.

Lizete de Assis is a computer scientist and artist. She obtained a Master in Artificial Intelligence. Her knowledge of the field and her stroke allow Lizete to illustrate even the most difficult notions in a simple and diverting manner.

Genres

Technology and the Future, Computers, Artificial Intelligence (AI), Science, Engineering, Robotics, Computer Graphics, Machine Theory

Table of Contents

Introduction.
1 What is Artificial Intelligence? 2 The Turing Test.
3 Why Is It So Difficult? 4 Lost in the Woods.
5 Winning at Chess.
6 The Journey Continues.
7 Darwin 2.0.
8 Small but Strong.
9 A Bit of Tidying Up.
10 Taking an AI by the Hand.
11 Learning to Count.
12 Learning to Read.
13 Learning as You Draw.
14 Winning at Go.
15 Strong AI.
16 Is There Cause for Concern? 17 To Infinity and Beyond! Even More AI! Acknowledgments.
They Made AI.

Overview

Artificial Intelligence (AI) fascinates, challenges and disturbs us. There are many voices in society that predict drastic changes that may come as a consequence of AI – a possible apocalypse or Eden on earth. However, only a few people truly understand what AI is, what it can do and what its limitations are.

Understanding Artificial Intelligence explains, through a straightforward narrative and amusing illustrations, how AI works. It is written for a non-specialist reader, adult or adolescent, who is interested in AI but is missing the key to understanding how it works. The author demystifies the creation of the so-called “intelligent” machine and explains the different methods that are used in AI. It presents new possibilities offered by algorithms and the difficulties that researchers, engineers and users face when building and using such algorithms. Each chapter allows the reader to discover a new aspect of AI and to become fully aware of the possibilities offered by this rich field.

Video and Podcast

The post Book Summary: Understanding Artificial Intelligence – A Straightforward Explanation of AI and Its Possibilities appeared first on Paminy - Information Resource for Marketing, Lifestyle, and Book Review.