Although modern AI systems still have trouble deciding whether or not to flip that stranded tortoise in their path, they’re already outpacing the intellectual capabilities of their creators in a wide variety of fields. From beating grandmaster Go players to outguessing cardiac surgeons, lipreading to audio transcription, neural networks and machine learning have already surpassed humans — and that list is only going to grow longer.
In fact, SoftBank Group CEO Masayoshi Son told attendees at Mobile World Congress in February that he fully expects computers running AI to exceed human intelligence within three decades. “I really believe this,” he said, joking that soon even a pair of sneakers will possess more computational power than the person wearing them.
“I believe this artificial intelligence is going to be our partner,” Son said. “If we misuse it, it will be a risk. If we use it right, it can be our partner.”
We’re already beginning to see benefits from that partnership, especially across the field of medicine. Researchers from the University of Nottingham in the UK, for example, recently developed a machine learning algorithm capable of predicting a patient’s propensity for a heart attack or stroke better than American College of Cardiology/American Heart Association (ACC/AHA) guidelines.
Similarly, last December a team from the University Hospital of Marburg’s Centre for Undiagnosed and Rare Diseases leveraged IBM’s Watson platform to automate the analysis of its patient data. Not only can the AI rank a patient’s likelihood of having a specific disease as well as human doctors can, it does so in a matter of seconds. It’s even found infection vectors that the humans missed.
Thanks to Watson’s exhaustive patient-intake form that “even asks them about their childhood and what pets they have,” Dr. Tobias Müller, of the Marburg Centre, told New Scientist. “We had one patient with inexplicable gut symptoms who, it turned out, kept an aquarium. He had caught the tropical disease bilharzia from his water snails.”
Of course, humanity’s relationship with AI does have a competitive side too. Well, “competitive” in the broad sense of the word given how often Google’s AphaGo AI has spanked the world’s best human players in recent years. AlphaGo has even been moonlighting online, dominating players across the internet while further developing its game.
Now, if the idea of a machine embarrassing you at ancient Korean boardgames makes you want to kick over the nearest tower of building blocks, you’re also out of luck because Facebook’s already built an AI that’s just as good at that as you are. Granted, it doesn’t have the necessary physical feet to actually do the kicking, but it can predict how those blocks will fall just as well as people can.
And it’s not just competitive winner-take-all sorts of games like Go. Researchers from Brigham Young University managed to train an AI system to play a digitized version of the prisoner’s dilemma. That’s where two players who are “accused” of a crime can receive the lightest sentence of 1 year if they cooperate, 2 years if they both rat on their accomplice or walk free while the other player gets 3 years if only one of them turns.
The Brigham Young algorithm, dubbed S# (S-sharp), quickly learned that cooperation was the key. By the end of the experiment, machine-only teams coordinated their efforts nearly 100 percent of the time while their human counterparts only worked together in about 60 percent of the games.
This AI-ownage carries over to the business world as well. Last May, a reporter for the Financial Times went up against a digital journalist named Emma from AI startup Stealth. Both journalists were given the same UK employment data and tasked with filing a story as quickly as possible. Emma smoked her opponent by completing her assignment in just 12 minutes, compared to the human’s 35. However, speed isn’t everything in the newsroom, and despite being able to include added context such as the effects that Brexit would have on employment, the AI failed to notice the crux of the post, that the number of job seekers had risen over the past year.
But despite that flub, the world’s largest corporations and most prestigious universities are betting big on using AI and machine learning to boost productivity. Microsoft is already developing automatic translation AIs that will convert your PowerPoint presentations into any language in real time. The Seattle-based company also recently made headlines when it debuted a spoken-language translator that outperformed its fleshy counterparts. Chinese tech giant Baidu is similarly integrating AI into its SwiftScribe audio transcription app so that the system can actively learn and improve based on the edits the user makes. Meanwhile, Oxford University has developed a lipreading AI called LipNet (pdf) that is up to 10 times more efficient at the task than your average lipreader.
Google is making even more ambitious inroads into AI development, by training neural networks to design and teach other smaller neural networks. The primary network takes a series of “candidate” networks then trains them using reinforcement training (the same way one trains a dog) to develop and select the most efficient one. “There are important caveats, we do have higher false positives,” Alphabet CEO Sundar Pichai told the crowd at this year’s Google I/O Conference, but the results are still promising enough that the company is looking to apply it to DNA sequencing and cancer research before making the technology ubiquitous. “We want it to be possible for hundreds of thousands of developers to use machine learning,” Pichai said.
So given that AI seems to be better at everything from video game animation to recognizing musical genres, is there any subject that humans still reign supreme? The list is shockingly short. Outside of naming paints (though, let’s be honest, “Dorkwood” and “Stanky Bean” are both pretty great), there doesn’t seem to be much that AI can’t beat us at if given enough time and source materials to properly train. And that’s just a little terrifying.