CHAPTER ONEWhat’s AI, Anyway?
Before we can understand how our future with artificial intelligence will look, we have to know what artificial intelligence is—and what it isn’t.
To do that, we need a definition of intelligence. This is the funny part: Not even experts on intelligence agree on a definition of the word.
For now, it’s enough to say that intelligence is the ability to learn—and then to apply that knowledge to a goal.
The goal part is important. A piece of paper might have facts written on it, but it’s never going to do anything with them all by itself. Paper is a tool, but it’s not intelligent. It doesn’t observe the world and react to it. Intelligent beings do.
We used to think humans were the only intelligent species. That’s not true, though. Plenty of animals and even some plants meet this definition of intelligence.3 And now, certain machines meet the definition, too.
It’s important to remember that AI is code that’s written—usually but not always by people—to perform certain kinds of tasks. AI is in use when machines take data, learn from it, and apply the learning to a task. It’s fun to joke about our future evil robot overlords, but that’s not what we’re talking about. Not yet, anyway.
We’re also not talking about the kinds of robots you often see in factories, the kind blamed for job loss through automation. It’s true that many jobs have vanished this way. But a robot in a factory is not necessarily AI. If it can’t learn and instead repeats the same task, it’s missing the intelligence part of the AI equation.
AI doesn’t even need to have a robot-like body. If you have a smartphone in your pocket, then you’re walking around with AI.
For now, think of AI as falling into two buckets: narrow and general.
* * *
Narrow artificial intelligence already exists.
If you’ve been online, you have almost certainly encountered narrow AI. If you’ve ever asked Amazon’s Alexa, Microsoft’s Cortana, Samsung’s Bixby, or Apple’s Siri anything, you’ve interacted with artificial intelligence. Also, social media apps like TikTok use AI to serve up posts designed to give you a certain experience.
That sounds like kind of a squishy way to describe it, doesn’t it? “A certain experience.”
But that’s really what it is. The type of experience you get is one determined by TikTok’s strategies. If they want you to stay online, they can adjust it to keep you riveted based on what you do when you use it. It’s even possible for social media apps to change your emotional state without your knowledge. Their software, guided by algorithms that let it learn from you and others, can influence your behavior and your feelings.
This is what narrow AI does well: It lets computers strive toward goals based on rules it has discovered through trial and error.
Not all computer programs are AI. Plenty of software is rules-based, which means someone has created a list of commands for the program to follow. A web page built in HTML is an example of this. It is made using tags that don’t change, and it’s the same page every time.
In contrast, AI uses large amounts of data, and by trial and error, discovers rules and patterns on its own. Another expression for this is machine learning, a branch of AI that involves learning from data, usually by updating some parameters that can be fine-tuned. Most modern AI involves machine learning.
The goal can be varied: selling socks on Instagram, catching a bogus credit card transaction, detecting disease, writing Shakespearean-style sonnets, or creating faces that look exactly like real people. The results are getting better all the time. A few years ago, for example, AI-generated cat faces were creatures of true horror; ones produced by new algorithms are excellent.
* * *
Artificial general intelligence (AGI) does not exist yet.
That’s the kind of broad intelligence that human beings have—and then some. A machine with artificial general intelligence could achieve any complex goal, regardless of training or environment. While it’s tempting to say “anything a human can do intellectually,” this is not likely to be the case. Today’s narrow AI is built as an echo of the human brain. But there’s no guarantee that AGI will have intelligence that is similar to the kind people have. Eventually, if a model for AGI emerges, many experts believe it will surpass what human beings can do with our brains. (And naturally, other experts don’t—but if the success of machine learning with games is any indication, the abilities of future algorithms will astonish us.)
Experts haven’t decided whether a machine with artificial general intelligence would have consciousness. Defining consciousness is a hard problem. To simplify things for now, we can consider that something has consciousness if it’s aware of itself and its experience in the world. Whether it has consciousness or not, AI will someday beat human intelligence at every single cognitive task—kind of like the way you can read, write, and do math in a way that cats and dogs can’t. (This is not to say that cats and dogs can’t do math; they can! But humans do have larger brains capable of more sophisticated things.)
For now, though, think about AI as either narrow or general. Narrow AI is already influencing your life. The other might someday—but not in the immediate future.
* * *
AI is already pervasive. If you were very young or not even born in 2012, when something called deep neural networks took off, it can be hard to realize how much this is the case.
Whether or not you’ve always lived with it, machine learning is everywhere now, both online and off—and it’s not always easy to tell when you’re seeing it.
Unlike a piece of hardware, like an iPad, a neural network is made of lines of code that you can’t see. But it’s probably something you interact with daily.
Let’s say you have a smartphone. If you unlock it with your fingerprint or face, then you’re using AI. This same kind of facial recognition is sometimes used at airports for international travel. Law enforcement agencies also use it.
It’s all over social media, too. Every time an app suggests you tag a photo of yourself or someone you know, it’s using this technology to recognize faces (and that is controversial, as you’ll soon learn).
Your phone might also come with a built-in bot. These bots can shop for you, tell jokes, play music, give directions, place calls, and answer questions—all prompted by the sound of your voice and processed by AI.
Let’s say you’re writing a paper or sending a text. That means you know how hard it is to duck autocorrect. Its cousin, autocomplete, uses a kind of AI to anticipate what you plan to write next. Search engines use data from other users’ queries to help you find what you’re looking for (this is why the suggested queries are sometimes weird and even offensive).
Google tries to predict your query using the data from other people’s searches. (Google)
* * *
Maybe you’re watching videos on YouTube or looking at photos on Instagram. The recommendations are generated by AI. The same is true for the ads for sneakers and clothes that appear in your feed.
Social media networks and apps also sometimes use AI to make sure users aren’t bullying others or making posts with inappropriate language. (They also sometimes use human beings for this.)
A conversation between the author and her Replika chat bot. (Martha Brockenbrough)
AI bots can converse—sometimes so well you won’t even know you’re talking to a computer.
And AI systems can write, create music, and make pictures good enough to fool people into thinking that human beings made them.
But it’s not just online and in devices.
AI can drive cars and trucks. Fully autonomous cars aren’t widely available yet, but ones that partner with human beings are. Robots with AI can pick up and sort packages—a surprisingly tricky task for nonhumans. In hospitals, robots can deliver medicines, relieving overworked humans from that chore and even preventing human error.
Moxi, a robot built by Diligent Robotics, helps hospital staff complete routine chores all on its own. (Diligent Robotics)
Drones are being used to find lost people, and some of those are trained to find the sounds of screaming humans, which seems horrifying until you remember tales of people being buried under rubble.
In short, there are tons of places where artificial intelligence will be used and where it will be useful. In the near future, you can expect AI to affect your education, your health and fitness, your travels, your finances, what you buy, click, and watch, and maybe even your relationships. It has the potential to remake economies and influence world politics, including causing diplomatic failures that lead to war and other forms of violence.
You can be certain that it’s going to change modern life the way the internet did—and then some. The more you know about it, the better prepared you’ll be for a future that is unlike anything humanity has yet experienced.
CHAPTER TWOThe Long Human Quest for AI
AI seems modern. Maybe even futuristic. But it’s actually an ancient idea that we made real through a combination of vision, invention, insight, and perseverance.
How old is the idea? A centuries-old Chinese text describes an automaton built in 1000 BC by an engineer named Yan Shi. It could move and even sing perfectly in tune. Ancient Egyptians imagined statues brought to life with captured souls.
Later, the Greeks imagined that Hephaestus, their god of fire, built a huge bronze automaton named Talos. Powered by a vein of magical blood, Talos had one job: to keep the island of Crete safe by flinging boulders at enemies or superheating himself to scorch pirates. That wasn’t all. Hephaestus, who had a disabled leg, also made a set of wheeled tripods to help him move—arguably history’s first imagined version of a self-driving car.
Here’s Talos keeping Crete safe in Attack of the Titans. (Columbia Pictures)
That’s what all technology does: makes it easier to do what we need, whether that is military defense or a mobility aid—or even help with math.
You might have played with an abacus in elementary school, for example. If you added and subtracted with those sliding beads, you’ve used a tool that’s more than 2,600 years old. At least 2,000 years ago, someone invented a device with 37 bronze gears that could track the movements of the moon and sun. It could also predict eclipses. More than 1,200 years ago, Aztecs invented a tool called a nepōhualtzintzin, which calculated numbers and planetary movements using beads and string. Users wore it on their wrists.
The suanpan is a Chinese abacus. Calculate comes from calculus, the Latin word for a pebble used in an abacus. (David R. Tribble)
These devices aren’t AI in the way that a computer scientist today would define it. But they are part of the ancient human tradition of invention, saving or enhancing human labor and reducing human error. To some people, that’s a basic definition of artificial intelligence: a tool that helps people with their thinking. But there’s more to it. These inventions, imagined and real, can be seen to come from a desire to lift humanity out of oppression and struggle.
The Greek philosopher Aristotle observed this more than 2,300 years ago: “There is only one condition in which we can imagine managers not needing subordinates and masters not needing slaves. This condition would be that each instrument could do its own work, at the word of command or by intelligent anticipation.”
Imagining tools that do their own work on command—or even by anticipating a command—sounds exactly like the aims of artificial intelligence today. (It should be noted, though, that Aristotle didn’t oppose enslaving humans.)
Aristotle imagined something like AI 2,300 years ago. (Jastrow, Ludovisi Collection)
* * *
First, we had to figure out how to think.
The world didn’t come with an instruction manual. Even so, people have done a remarkably good job discerning its unwritten laws. This is exactly what programmers design AI to do: figure out through trial and error how to get a desired result.
If you’ve ever played a video game like SimCity, you’ve done this discerning yourself. To play, you have to track a bunch of interconnected variables and make decisions accordingly. This forces you to figure out the underlying rules that govern the game.
For example, if you want to increase your population, you have to build more houses. But if you build them too quickly, you’ll deplete the forest … All of this requires thinking and the formulation of theories about the rules of the world. You win—or at least do well—when you figure out the hidden logic.
Human beings didn’t evolve knowing how to think about big, abstract problems in a formal way. We first had to conceive of logic. Aristotle’s description of the framework for it has been particularly influential, even with the kind of thinking that gave rise to computers and software.
Using Aristotelian logic, you make conclusions using syllogisms. A syllogism starts with two premises that overlap:
All pies are round.All round food is delicious (e.g., pizza, donuts, and quesadillas).Then you make a conclusion based on the premises:
Therefore, pies are delicious.
This approach doesn’t guarantee that the underlying premises are correct. You might hate pizza, and it is also possible to bake a rectangular pie. But the syllogism gives us a way to describe and understand the world, a way of knowing things. It’s also a way of abstracting general rules for categorizations. If we hadn’t developed this logic, we wouldn’t have digital devices.
Muhammad ibn Musa al-Khwarizmi, as featured on a Soviet postage stamp.
AI uses algorithms that offer up a set of rules. The word algorithm has ancient origins. It comes from the last name of a Persian scholar named Muhammad ibn Musa al-Khwarizmi, who invented algebra around 825.
George Boole invented Boolean algebra.
Copyright © 2024 by Martha Brockenbrough