Far from being installed in evil robots or charged with guiding a spaceship through a treacherous part of the galaxy, artificial intelligence (AI) in 2017 is rather more boring. You’re unlikely to find something that can ask the important questions but you might stumble upon a piece of computer code that can tell you the difference between a banana and a carrot; that, or give you accurate directions to the Post Office.
Alice, Bob, and Eve
Google (who else?) has made AI a major concern over the past few years. Its most famous creation is, of course, DeepMind, a program that the tech giant claims will one day be able to solve problems without the necessity of prior teaching; or, to put that another way, it’ll be able to learn all on its own. Founded in 2010, DeepMind has its sights set on healthcare; specifically, to help doctors diagnose patients with greater accuracy and to bring down statistics that claim 230,000 people wait too long for scan results. In a bid to get more companies involved in producing AI, Google has created a fund from which startups can earn a financial boost, with Algorithmia, a firm that calls itself an “open marketplace for algorithms” and has the ultimate goal of building more intelligent apps, taking $10.5m from the first round of investment. Algorithmia sells plug-and-play AI to developers (it currently has more than 3,500 available), as well as much larger solutions to firms involved in machine learning. Also at Google HQ, a second AI project, Google Brain, is designed to answer questions about how intelligent computers and robots deal with sensitive data, chiefly by creating neural networks – 3 of them, Alice, Bob, and Eve – that can interact with encrypted messages. The trio of cryptographer AIs worked both with and without “keys” to encrypt and decipher the messages, and, over 15,000 attempts, Alice and Bob managed to create cyphers that Eve couldn’t crack.
Arguably the biggest boon to AI research is its increasing accessibility and the growing number of programs that do silly things like make their own music, recognize drawings of objects, or organize 1000s of bird calls, as in a project created by Manny Tann, Kyle McDonald, Jessie Barry, and the Cornell Lab of Ornithology. The kind of freedom at which interested parties can experiment is key to cracking the more difficult problems faced by AI researchers. A popular learning tool for aspiring Google coders is the creation of blackjack bots that can understand complex strategy like doubling down and pair splitting, blackjack strategy in which it means making two hands from one in the event that the player is dealt two cards of the same value. Such rules are already part of the AI programming, but the idea is that the bots play each other to see which one is the smartest overall, learning in the process. From there, pattern recognition and language processing, the code that powers parts of Siri and Cortana, Apple and Microsoft’s respective AI assistants, are logical next steps. The fact that Cortana attracts 133m users a month and 19% of people ask Siri for help daily is a testament to the importance of language processing. AI is difficult though. Considering that only two species on earth have ever asked an existential question – humans and Alex the grey parrot (“What color am I?”) – the idea that we might be able to program something to do the same sounds more than a little arrogant; self-awareness is extremely rare in nature. An AI must also be able to process visual information like the brain does and speak or type naturally to be considered at least equal to humans.
That latter point is perhaps better known as the Turing Test, the requirement that a computer can communicate in a way that’s indistinguishable from humans to be considered intelligent. Experiments with “chatbots” have had mixed results; Siri is advanced enough to have something approaching a sense of humor but Microsoft’s Twitter AI, Tay, went rogue and was removed from the web after 16 hours. Insane Twitterbots aside, AI isn’t going anywhere. By 2020, Gartner expects robots to power 85% all of customer service scenarios, with content, press releases, and similar copy likely to come from the virtual pen of AI at 20% of businesses from next year. It’s obviously not an ideal situation for flesh and blood workers but the elimination of human error, laziness, and pesky things like health and safety are an obvious boon for companies.
It’ll take some time before intelligent robots can ask questions about the color of their circuits but a breakthrough in AI is coming, and it’ll be one of the most significant events in human history to date when it does.