10 Impressive Things Artificial Intelligence Does Better Than Humans
Quick: What do you think of when you hear the words “artificial intelligence?” You might think of Siri or Alexa. Maybe you picture robots that are coming to steal your job. Or perhaps you think of technology that turns against its inventors and spells the end of the human race.
Whatever your personal stance may be, according to recent headlines, the population is split when it comes to their belief in AI. Some individuals think the premise of it is over-hyped, while others think that you shouldn’t underestimate its significance.
So who’s right and who’s wrong? For now, there is no clear answer. However, AI isn’t just something that’s going to happen in the future. It’s happening now. With numerous approaches to advancing the field, AI is sure to impact our lives in some way. In fact, it can already complete many tasks better than humans can. Don’t worry, though. In its current forms, it won’t make your life seem like a science fiction film overnight or take your job away.
While you’re probably familiar with the high-profile milestones achieved in this category, you most likely haven’t heard them all. Some may even surprise you. With that intent, we’ve put together this list of the most impressive things that AI can already do better than humans.
1. Playing “Jeopardy!”
Let’s start with the AI win that just about everybody has heard about. IBM’s Watson computer winning Jeopardy! in 2011. According to TechRepublic, Watson won the game show with $77,147, leaving human champions Brad Rutter and Ken Jennings in the dust with $21,600 and $24,000, respectively.
As Steve Lohr reports for the New York Times, the triumph seemed to suggest “limitless horizons for artificial intelligence.” But as illustrated by IBM’s struggles to convert Watson from science project to a commercial technology, progress in AI “typically comes in short steps rather than giant leaps.”
2. Playing Go
Another game at which AI has bested human champions? The complex board game Go.
Joon Ian Wong and Nikhil Sonnad report that AlphaGo — an artificially intelligent player developed by Google DeepMind in London — didn’t only effectively win the game. They said it also “came up with entirely new ways of approaching a game that originated in China two or three millennia ago and has been played obsessively since then.”
Lee Sedol, one of the world’s best and most experienced Go players, lost the first three games in AlphaGo. And he won the fourth one only because he adopted “some of AlphaGo’s strategy by pursuing less expected and riskier maneuvers that proved successful in the end,” he says. This episode, according to Quartz, illustrates that “the true value of artificial intelligence reaches far beyond the simplistic narrative of man versus machine. Instead, AI’s potential may be in teaching humans new ways of thinking for ourselves.”
3. Playing video games
It’s not just game shows and board games that AI can beat you at, though. Elizabeth Lopatto reports for The Verge that Google’s DeepMind can also learn to play video games. Specifically, it taught itself to play dozens of Atari games. Out of 49 Atari games, AI performed better in 43 games than previous technologies, and better than humans in 29 games.
So what’s the big deal? These are pretty simple video games to maneuver, after all. Well, they’re a better model of real world chaos than the games that previous computers have mastered, like chess, for example. And as Nicola Twilley points out in The New Yorker, chess has an “extremely limited ‘feature space.'” All a computer needs to consider is the positions of the pieces on the board during its hundreds of turns. But an Atari game has much more information and hundreds of thousands of turns. “In this sense, a game like Crazy Climber is a closer analogue to the real world than chess is, and in the real world humans still have the edge,” Twilley states.
4. Playing poker
Recently, AI is besting humans at poker. Libratus, an AI system built by computer science researchers at Carnegie Mellon University in Pennsylvania, beat four of the world’s top players. “Libratus relied on three different systems that worked together, a reminder that modern AI is driven not by one technology but many,” Wired reports.
Poker has been a difficult game for AI to crack. That’s because there’s no single ideal move, and because the computer needs to randomize its actions so that opponents aren’t sure when it’s bluffing. But as Libratus’s success illustrates, AI can now out-bluff even the best of champions.
5. Playing with building blocks
Artificial intelligence can do more than just beat games, according to ComputerWorld. It can also learn to play with traditional toys, like building blocks. A 2016 study conducted by Facebook revealed that AI can learn the intuitive physics of wooden blocks, a toy that enables infants to develop motor skills and learn about the physical behavior of the world.
The researchers used a 3-D game engine to create small towers of wooden blocks whose stability was randomized, rendering them either collapsible or standing upright. With that data, they trained convolutional network models to accurately predict the outcome by estimating the blocks’ trajectories. By generalizing the learning experience to new physical scenarios, the AI has successfully achieved performance comparable to human subjects.
6. Transcribing audio
Want to know what useful tasks AI can do when it isn’t playing games? For one, it can transcribe audio. According to ComputerWorld, Microsoft’s researchers “have been tweaking an A.I.-based automated speech recognition system so that it performs as well as, or better than, people.”
And the proof is in the tests. During a test implemented by the National Institute of Standards and Technology, the AI system had an error rate of 5.9%, the same as human transcribers that Microsoft hired. When the test was replicated, it had an error rate of 11.1%, nearly matching the human score of 11.3%. This shows us that AI leaves little room for error in audio transcription, and is certainly not any worse off than we are.
7. Lip reading
Another surprising task that AI has learned? Lip reading. Quartz reports that a lip reading AI system developed at the University of Oxford is more accurate than humans. Professional lip readers can figure out only 20% to 60% of what a person is saying. But LipNet, a system developed with funding from Alphabet’s DeepMind, can watch a video of a person speaking and match text to the movement of their mouth with 93.4% accuracy.
Some of the AI’s extraordinary accuracy “might have to do with the fact that it was trained and tested in extraordinary conditions,” Quartz reports. But the researchers tested human lip readers on 300 random videos, like the ones they fed the AI. The result? The humans had an average error rate of 47.7% — and the AI’s was just 6.6%.
8. Reporting news
Another useful skill that artificial intelligence is learning? Writing and reporting news. “AI is particularly quick at turning structured data into words, as a Financial Times journalist found when pitted against an AI called Emma from California startup Stealth,” ComputerWorld reports.
Another example of this technology in play consists of a homegrown AI system writing hundreds of reports about the Rio Olympics for The Washington Times. Automated Insights also had success in this area when they launched a free service based on Wordsmith, the technology it uses to generate stories for the Associated Press. So far, such AI may be best at formulaic writing, like weather forecasts based on meteorological data, or sports narratives based on score sheets. But artificial intelligence is incredibly fast at turning data into news stories. Even if they’re a little more dry than the stories penned by their human counterparts.
9. Diagnosing diseases
It’s not just journalists who can get help from artificial intelligence. Doctors, too, will increasingly be able to use AI to more accurately diagnose diseases — at least in the long run, once the risks and models are worked out.
Simon Parkin reports for MIT’s Technology Review that artificial intelligence from Babylon can check a patient’s symptoms against a database of diseases. Taking into account the patient’s history and circumstances, it will then recommend an appropriate course of action. Similarly, IBM’s Watson moved on from Jeopardy! to using “600,000 medical evidence reports, 1.5 million patient records and clinical trials, and two million pages of text from medical journals to help doctors develop treatment plans tailored to patients’ individual symptoms, genetics, and histories,” Parkin states.
10. Using human intuition
Sure, computers can crunch numbers. But they can’t beat human intuition when it comes to looking for important patterns, right? Wrong.
The Washington Post reports that an algorithm created by Max Kanter at MIT’s Computer Science and Artificial Intelligence Laboratory can identify patterns in unfamiliar data sets, just like human data scientists. For instance, it can predict when a student is most at risk of dropping an online course, or indicate the reason someone turns into a repeat buyer. The algorithm “seems to do a decent job of approximating human ‘intuition’ with much less time and manpower,” The Washington Post states. While it’s unlikely that such algorithms would replace human intuition, according to Gizmodo, “it seems plausible that they could help make the analysis of large pools of data a little faster.”