It’s been a big couple of weeks for chatbots. First Microsoft’s teen-inspired bot gets turned from flights of fancy to Nazism and racism. Then Facebook CEO Mark Zuckerberg falls under the chatbot sway, predicting a rosy future for these in-messenger plug-ins.
For those who’ve been computing for more than a few years, chatbots bring to mind the fevered natural-language meanderings of psychbot ELIZA, and a maze of twisty little passages, all of them different.
The former was an early artificial intelligence experiment, crafted at MIT 50 years ago, while the maze comes from an early text-based adventure game called, wait for it, Adventure. Both attempted to parse basic text commands to mimic a therapist, and provide a diverting escape.
Neither worked really well, but like the singing frog, back then you didn’t worry about how well it sang, but instead marveled that it could sing at all. Today’s chatbots are somewhat more convincing.
But as Microsoft found with Tay, advances in Natural Language Processing (NLP) don’t necessarily translate into success. It took less than a day for Tay to “learn” from its interactions and evolve from playful and vacuous to racist and homophobic.
Although much was written about Tay’s descent into darkness, it’s just the beginning of the chatbot revolution, where computer-based entities will interact with us via text – and ultimately voice to enhance our lives. I’m wildly optimistic about where this technology will take us.
It’s not likely to pass the Turing test anytime soon though. Although eventually for text and voice to become the dominant man-machine interface, it’s going to take time. And until then chatbots will be more specialized assistants – and occasionally objects of ridicule – rather than the long-hoped for guy or gal Friday.
The most successful over the next year or two will be special purpose – those designed to replace a directed task with a limited array of actions or initiatives baked in. The classic example is of a travel bot that helps with reservations or fixes problems when you’re on the road. KLM’s new Facebook app does just that – helping with reservations on the Dutch airline, and presumably freeing up humans to handle more complex routings and problems. It’s a mechanized version of the DeltaAssist twitter account, one that over the next few years will replace much of the airline’s interaction between customers and human reservation agents.
The next few years will see a plethora of chatbots focused on the lower levels of Maslow’s hierarchy of needs. These will be focused on physiological (food, clothing, shelter, where’s the potty) and safety (health, financial and physical).
But the holy grail remains more generalized conversational assistants that fulfill all of our needs, from love and belonging to transcendence. Google, Apple, Microsoft and Amazon are setting themselves up as the Four Horsemen of the Chatbot Apocalypse with (respectively) Google Search/Now, Siri, Cortana and Alexa. And yes, these four efforts are impressive. But they’re still in the “singing frog” category today. They sound perhaps a bit more polished than ELIZA, but still provide hours of maddening exchanges regularly lampooned across digital and traditional media.
However, as processing power advances – and quantum computing goes mainstream – fuzzier and more generalized computer-based language processing will become far more successful. By 2021 I expect we will be happily conversing with Tay’s descendants without even thinking about whether it’s a person, or a computer. And that, finally, will finally deliver on the promise of “natural”, and the singing frog will finally transform into a mellifluous princess – finally delivering us from the tyranny of the keyboard and the glowing rectangle.