What is artificial intelligence?

Since the 2022 launch of ChatGPT, artificial intelligence (AI) has rarely been out of the news. That particular kind of generative AI is called a large language model (LLM) and is built using an actual neural network that allows it to interact with us in increasingly human-like exchanges. For this reason, it is much more engaging and “likeable” than other forms of AI. But it’s important to remember that LLMs represent only a fraction of the AI-powered solutions, tools, and devices that increasingly augment and enhance modern business operations. From optimizing supply chains to flagging anomalies in complex systems, AI is already embedded in the platforms and processes that keep modern industries running. Some of these tools work quietly behind the scenes. Others help frontline teams work faster, reduce risk, or make better decisions. But wherever we find them, AI-powered solutions are increasingly changing how work gets done.

Artificial intelligence (AI) refers to the ability of machines to solve problems, perform tasks, and simulate human cognitive functions such as perception, reasoning, learning, and decision-making. Unlike traditional software, which follows fixed rules, AI systems improve their performance over time by analyzing data, identifying patterns, and adjusting their behavior based on outcomes and feedback. Some AI systems rely on rules or logic trees. Others use statistical models trained on large datasets. The common thread is that AI empowers machines to handle complexity, ambiguity, and variation.

The idea that machines could think predates computers and started with questions like: Can logic be automated? Can reasoning follow rules? Can a machine ever learn? As early as the 1950s, pioneers like John McCarthy and Alan Turing were exploring programmatic ways to quantify and formalize thought.

By the 1990s, rule-based systems could capture decision-making in code. They worked well in narrow domains but couldn’t easily scale or adapt. It was this limitation – combined with the rapidly-growing memory and capacity of computers – that inspired the developmental strides in deep learning. This saw models beginning to learn patterns directly from data, rather than from being hand-programmed to do so. 

Today, AI is an exponentially advancing phenomenon. And while it obviously requires human oversight, training, and management, it has become historically unique among technologies in its ability to use data and experiences to fine-tune and improve its own performance. 

There’s a lot of hype around AI. Particularly, its output is becoming increasingly indistinguishable from that of humans. It is often described in tiers – from narrow, task-specific systems to speculative forms of intelligence that rival or exceed our own. Here's how those types break down, and where we stand today.

1. Narrow AI
   The type of AI that most of us use every day is called narrow AI, but that doesn’t mean it’s unsophisticated. Trained on often billions of data exemplars, these systems are only “narrow” because they are designed to perform a single or defined set of tasks, such as identifying defects on a factory line, recommending a product, or responding with natural language to a prompt or question. These models are typically built to spot patterns, make predictions, or assist with decisions in a specific context. Large language models (LLMs) like ChatGPT are still examples of narrow AI. And even though the deep learning neural networks they run on can make it seem like they’re “thinking,” they’re generating responses based on statistical patterns rather than understanding or intention.
2. General AI
   General AI (or artificial general intelligence, AGI) refers to an as-yet non-existent form of AI that actually understands the information it learns and can apply that knowledge across a range of unrelated tasks. For example, an AGI model observes a toddler playing with matches. It was never trained that this was a risk. However, it uses its knowledge of kids’ behavior and the properties of matches to reason out that there is danger and that it should intervene. But despite media stories about this or that AI model being sentient or having its own hair-raising agenda, AGI does not currently exist. While narrow AI continues to grow more advanced, most experts believe we’re still far from achieving AGI.
3. AI superintelligence
   This kind of AI would surpass human intelligence in nearly every area, but for now, it remains firmly in the realm of science fiction. In theory, it could learn independently, perceive its environment, and even develop self-awareness or its own motivations. At that point, predicting how humans and such machines might coexist becomes nearly impossible. Fortunately, most computer scientists agree that this dystopian future is unlikely to arrive within our lifetimes.