What is…artificial intelligence?

A monthly tech explainer series about the technology shaping our world today, from the Garage.

By Poornima Apte — May 5, 2022

The term “artificial intelligence” can conjure up dark images of cyborgs, ready to vanquish the human race. But the real-life version is much more benign. Done well, AI has the potential to help us work, play, and accomplish many other tasks in much smarter ways than we could by ourselves. Instead of supplanting human intelligence, AI  supports it — and can be used for good: It has helped sift through large numbers of potential COVID-19 vaccine candidates, aids in the rebuilding of earthquake-damaged structures, helps blind and low-vision people, and dismantles human trafficking networks.

What is AI?

Illustration by Eric Chow

How it works

Artificial intelligence is created by software programs to do things that normally require human intelligence. While AI’s implementations vary, one particular kind is its bread and butter today: supervised machine learning. In supervised ML, a model trains on a large number of example datasets to recognize patterns. 


RELATED: What is robotic process automation and chatbots? 


You’ve probably seen it at work “manning” airport security. To recognize a firearm in luggage, the model feeds on thousands of X-ray images of firearms. An agent teaches the model which images have a firearm and which ones don’t. An algorithm improves or “trains” the model based on this information; this is why this type of AI is called “supervised” learning. At work, the trained model compares X-ray images with what it has learned to spot suspicious packages. 

The a-ha moment

AI research first captured the public’s imagination when a chess program defeated grandmaster Garry Kasparov in 1997. Since then, the criteria for what qualifies as true intelligence keeps changing. We went through an AI “winter” in the late ‘80s when the technology fell short of promised claims and funding dried up. Today’s renewed appetite dates back to the early 2000s when a group of Canadian scientists refined an old technique, neural networks, to tackle realistic pattern-matching problems. Soon, the pattern-matching, anomaly-finding kind of AI was off to the races. By 2011, new software algorithms and AI-specific tools that could take advantage of massive parallel computing engines like graphics processing units (GPUs) combined to make the deep learning form of neural nets emerge as the leading practical solution for AI. 

What AI is used for today

In the past decade, deep learning has become the dominant form of AI used in most commercial applications, such as computer vision and natural language processing. When you use your credit card on your beach vacation and your bank sends you an alert, that’s AI in action, looking for anomalies in spending patterns. AI also helps you pick the next movie to watch on Netflix and tailors your social media feeds. HP service agents use AI to resolve customer problems faster. Faster than any human can, AI crunches through vast numbers of datasets to find pathologies in chest X-rays, stay a step ahead of wildlife poachers, or identify fraud in financial data. 

How AI might change the world

AI is under the hood of many promising technologies like surgical and elder care robots. It will also steer fully autonomous driving, personalized medicine, and the metaverse

Decades from now, imagine requesting a self-driving car to take you to the hospital so a robot can conduct your surgery. Science fiction? More likely just science, in the future.