2
Understanding Artificial
Intelligence

POPULAR-LEVEL CONVERSATIONS around the implications of artificial intelligence (AI) vary widely between enormous optimism and deep skepticism-even fear. For our purposes, we need to temper both extremes to be able to more objectively examine what AI even is, and only then establish a foundation to think well about it.

First let's look at the nature of intelligence itself.

To some extent, the term artificial intelligence is a bit misleading in that to many it implies a level of internal sophistication that current iterations of AI simply do not have. For centuries philosophers have debated about the nature of intelligence-what is it, who possesses it, how do we recognize it, etc. Growing up, I thought they gave advanced degrees to people for their intelligence. However, now having earned a doctoral degree and still routinely struggling to spell rhythm correctly on the first try, I can tell you this is not always the case. There are different types of intelligences, and the definition of intelligence itself is a concept open for debate.

Mid-twentieth-century mathematician Alan Turing gave the world a way of looking at intelligence as something that should be measured "not [by] the mechanism, but [by] the manifestation of intelligence."1 In other words, Turing asserted that intelligence can only be measured by the external behavior of a thing, arguing that "if a machine operated so proficiently that observers could not distinguish its behavior from a human's, the machine should be labeled as intelligent."2

This approach, later named the "Turing Test," has become a gold standard by which we measure the intelligence of machines-in terms relative to the performance of their human overlords (you and me) to perform a similar task. Today's current iterations of AI are dubbed "intelligence" not because they themselves possess intelligence in the way that you or I might have spoken about the smart kid in school, but because they are made to mimic forms of human intelligence.3 In this way, we might say that as it pertains to AI, there is a difference between mimicking intelligence and possessing intelligence. AI is in the business of mimicking intelligence more than it itself is intelligent in an ontological sense. But for those of us not immersed in the world of technological development, we consider AI "intelligent" nevertheless because of its capacity to perform tasks in a manner similar to human beings. In no small way does AI fool us in causing us to believe it is intelligent in a way that we are intelligent, yet it is only a mimic of intelligence rather than something that possesses intelligence in a proper sense.

There are several ways technology experts classify and group types of AI. The categories used by industrial research giant IBM, however, are among the simplest to understand for those who aren't immersed deeply into the world of AI. For our purposes, we will use these categories throughout the book.4

Category one: Narrow AI. Simply put, narrow AI (sometimes also called "weak AI") is the sort of technology we have today. Narrow AI is programmed to execute a defined task; it leverages the curation of information to perform within defined parameters given to it. Narrow AI can perform tasks better and faster than humans. While OpenAI's debut of ChatGPT caused many to assume that AI is new, this sort of AI has been around for decades, though it has improved by leaps and bounds the last several years. Even chatbots and relationship bots, which are designed to mimic human speech and conversation, are simply a form of narrow AI. This broad category encompasses other terms like "generative AI," "large language models," and more.

Within the scope of narrow AI, IBM outlines two subtypes of technology, based on functionality. First, there is "reactive machine AI," which uses currently available data to predict or recommend future decisions. Have you ever been in a conversation where you mention something that you'd like to purchase, only to find the next time you're scrolling Instagram your ads are magically tailored to that purchase? That, and other recommendations by "the algorithm" (whether it be suggested friends on Facebook or the "shows you might like" section on Hulu), are reactive machine AI. Reactive machine AI has been a regular part of our daily lives for years now, often directing our purchases, social media engagement, and streaming preferences. It shapes our worldview more than we would ever like to admit.

The second subcategory of narrow AI is "limited-memory AI." The advances in limited-memory AI have sparked public interest so widely since the debut of ChatGPT by OpenAI. Unlike reactive machine models, limited-memory models can retain past information to inform their current and future tasks.

ChatGPT and the virtual assistants on our smartphones are examples of limited-memory AI. The rapid evolution of AI technologies in this space is awakening public imagination over the potential AI has both to massively disrupt the norms of our lives as well as to add expediency to them.

In my own work, AI has played a significant role in streamlining my efficiency. I use an AI-powered productivity planning tool to schedule and rearrange the work I complete on a daily basis. This allows me, someone who is woefully ill-suited to break down large tasks into smaller, time-bound action items, to maintain a high level of productivity while still maintaining a quality harmony between my work and the rest of my life.

Following OpenAI's release of their custom GPT feature in 2023, I created a custom GPT, a limited-memory AI version of "myself." The KorpiGPT bot has learned from my previous writing and other publicly available content and can make recommendations about future projects with ease. While I don't use the bot for the actual writing process, KorpiGPT has been an invaluable tool to help me create outlines, edit content, and be a brainstorming aid that has saved me from countless hours of staring at "the drawing board."

On the back end of a lot of my larger writing projects, I use Grammarly's masterful AI-powered resources to make recommendations in the editing process. I tend to be verbose, so Grammarly helps me trim down my word choice. I also tend to be needlessly snooty in my word choice, and Grammarly helps me temper that tendency in favor of a better blend of readability and sophistication.

So, while a bit oversimplified, these things can reasonably be said:

1. 1. Narrow AI is today's AI. Everything else is theoretical.

2. 2. Of narrow AI's subcategories, reactive machine AI has been giving order to our lives for a while.

3. 3. Limited-memory AI, from personal assistants to self-driving cars, is what is revolutionizing our lives today.

Category two: General AI. "Artificial general intelligence" (AGI, or sometimes called "strong AI") is theoretically the next step in the evolution of artificial intelligence. AGI is a theoretical concept in which AI technology could possess the intellectual capabilities of a human, including its need to grow into its own intelligence, much like a child grows intellectually over time. Fictional depictions of AGI, both by Paramount, include Cortana on Halo and Zora on Star Trek: Discovery-two AI that are sentient and grow in their capacity to not only expand knowledge but expand in the application of that knowledge.

Unlike narrow AI, which can only perform tasks within its defined parameters, AGI would have the ability to perform new tasks outside of its original context, without human intervention. So, hypothetically speaking, an AGI tool designed as an expert in eighteenth-century French art could decide to also develop an expertise in international poverty development or could learn to compose music.

One type of AGI in its experimental phases is in what's called "emotion AI." If actualized, emotion AI would be able to detect, analyze, mimic, and respond to displays of human emotion. Think about getting home from work after a long, hard day, and Alexa recognizing your facial features, gait, and the tone in your voice and responding by proactively offering empathy, turning on what it knows to be your favorite mood-booster playlist, and perhaps even offering to DoorDash you dinner so you can kick your feet up and relax. Of course, with every possible upside, there are cautious dystopian scenarios, such as what is showcased in the episode "Help" of the 2019 Netflix series Creeped Out, which describes an AGI who takes license with two unruly children while their parents are away, trapping them in the house until they get along in order to teach them a lesson their distracted parents are unable to teach.5 But I much prefer the scenario where Alexa knows that I'm sad and orders me Taco Bell to cheer me up.

Category three: Super AI. "Super AI" is a theoretical future event wherein artificial intelligence not only has intelligence capabilities that match humans, but actually exceeds human capability. In this future, inaugurated by what many refer to as a technological "singularity," super AI will have the ability to take up new tasks, create new things, self-diagnose and improve, and essentially exist without the need of human...