The impact of Artificial Intelligence (AI) on the world must surely be one of the greatest contemporary puzzles. The spectrum of risk and the gamut of possible applications are significantly complex that almost any scenario can be envisioned, from robot apocalypse to workless utopia. The only assurance is that change is coming, and in my opinion, it is likely to be a revolution of a scale seen only during the onset of history-shifting events such as industrialization or farming.
The timing is auspicious: Google AI unit DeepMind’s AlphaGo has recently beaten South Korean professional game player Lee Sedol at the ancient Japanese board game Go—a game of huge potential complexity based on simple rules and considered one of the biggest challenges in AI, since it defies brute-force planning. As a milestone in machine intelligence (and good PR), it sits up there with IBM Watson’s victory on Jeopardy in 2011 and IBM’s earlier AI DeepBlue’s victory over chess grandmaster Garry Kasparov in 1997.
What AlphaGo does is worth briefly exploring, since it relies on the same core principles as all machine learning and resulting AI. The system is structured as a deep neural network, mirroring the connections made as humans learn. This system is then fed with huge quantities of data until the machine can identify patterns, allowing for complex feats like facial recognition and semantic natural language processing. It’s this structure that allows the same capabilities in Facebook, Google, and the NSA.
However, for DeepMind this kind of learning is limited in part by what the human programmers tell the machine it is looking at. In other words, if you tell the AI it’s looking at a dog, it will learn that it’s looking at a dog after multiple repetitions of the training. It will learn, but it still needs to be taught that it’s looking at a dog until it learns it is, indeed, looking at a dog; for these feats humans are the teachers.
Where DeepMind took things a step further is that it allowed AlphaGo to learn from itself by experimenting and playing against variants of its own intelligence. This not only reinforced the learning from human input, but also allowed AlphaGo to discover its own unique ideas—what might appear as creativity. In the case of Go the success in creativity can be clearly shown— games are won or lost on brilliant, out-of-the-box moves. As seemed to be proven in Game 2 and Game 4 of the AlphaGo-Lee showdown, two games pivoted on moments described in various superlatives (e.g., “one in a million”) by those in the know. Most interestingly, while one of these superhuman leaps of creativity was from Lee, the other was from AlphaGo.
If repetition creates learning, and creativity can be extracted from experimental self-learning—a process that might be similar to thought or reflection—then AI can surely provide creativity. AlphaGo showcases a specific, and relatively narrow (although very impressive) use of AI, but the ability to ape human thought will be more far-reaching than we realize, and is already starting to happen. Journalism is going through a related process of steady automation as AI, such as Narrative Science’s Quill, learns to write informative text, and also increasingly understands how to stylize and build a wider narrative. A similar creative leap has been made by the European What-If Machine, which has penned London’s West End musical “Beyond the Fence.”
A review of work likely to be lost to automation places roles in telemarketing, legal aides, accounting, auditing, and news writing all set to be replaced heavily by automated systems in the coming years. Factory work has already been supplanted in large part by assembly-line robotics, but more complex pattern-oriented tasks including those done by warehouse workers and delivery drivers are also under threat from developments, such as Amazon delivery drones and automated warehouse machines tendered by the likes of Kiva Systems. Even highly cerebral jobs, such as medical diagnostics, are being offered by IBM Watson, and driverless cars and planes are hot on the heels.
AI will allow a degree of automation and robotization of the world that will make many things unimaginably different and likely spark a wave of societal and economic reform not seen for generations. And we’ll look back at seminal moments like AlphaGo, DeepBlue, and Watson as just the beginning.
At what point can you be sure this article wasn’t written by a machine?
Tom Morrod is the Senior Director for Consumer Electronics, Broadband and Video Technology at IHS
Posted on 21 April 2016