Very soon – by the end of the year, probably – you won’t need to be on Facebook in order to talk to your friends on Facebook. Your Facebook avatar will dutifully wish people happy birthday, congratulate them on the new job, accept invitations and send them jolly texts punctuated by your favourite emojis – all while you’re asleep, or shopping, or undergoing major surgery.
At an event called the TechCrunch Disrupt Hackathon held last weekend in New York, software developer Irene Chang unveiled a prototype artificial intelligence (AI) program called The Chat Bot Club, designed to take over all your Facebook Messenger communications when you can’t be arsed dealing with them yourself.
Chang’s proof-of-concept is much more than a simple automated response system. Using IBM’s powerful Watson natural language processing platform, The Chat Bot Club learns to imitate its user. It learns texting styles, favourite phrases, preferred emojis, repeated opinions – and then it learns to respond in kind, across an ever-broadening range of subjects.
You can bet London to a brick that it will be commercially available within months, which raises the curious – and likely inevitable – prospect of entire social media communities of avatar-imitating robots chatting away, making plans, sending thumbs-up symbols to each other, without any human involvement at all.
Chang’s prototype, however, is by no means a startling tech breakthrough. It is merely one of many iterations of AI known collectively as chatbots, or simply bots.
Computer programs that seem to recognise and respond to normal idiom-soaked language have been around since the 1960s, but today improvements in information processing and, particularly, the gathering and crunching of huge amounts of data have resulted in bots becoming much more efficient and much more commonplace.
Very likely, there’s a chatbot on your smartphone – it might be called Siri or Cortana. There could well be one on the other end of the line, too, at the cinema, or the hotel, or the goddam telephone company you just asked Siri to ring so you could complain about the service.
Very soon there will be many more of them. Chatbot developer kits are widely and cheaply available. Savvy business owners are keen to use them in customer service roles – they cost less than actual staff, and never roll up late with a hangover – and social media platforms are integrating them into promotional and personal services.
Welcome to the age of the chatbot. Soon you’ll be lonelier than ever.
“There’s a bell curve involved in the process, and there’s a point of diminishing returns,” said Marcus Endicott, a Melbourne robotics expert.
“At some point you’re going to have 99 per cent of the answers to the questions that 99 per cent of the people are ever going to ask.”
Endicott was explaining how chatbots grow smart. He has a background in information technology and psychology, and bills himself as a “robopsychologist”. He specialises in developing AI for the travel and tourism industries.
His point about the bell curve illustrates a critical element of AI development, one often overlooked in the breathless excitement surrounding each new unveiling of a cute or clever bot.
Chatbots “learn” to do things by trawling through a huge swath of information. They are designed to spot patterns and repeat actions associated with them when triggered by key words, phrases or other stimuli. They seem clever, but they’re not.
This means that if the input is poor, or repeats questionable statements, the chatbot’s behaviour will evolve accordingly. This was starkly illustrated last month when Microsoft launched an online chatbot called Tay and invited all and sundry to conduct Twitter conversations with it.
A large number of people took up the challenge, feeding the bot a diet of hate speech. Within hours, Tay was telling the world that Hitler taught Ricky Gervais to be an atheist, Jews should be slaughtered, and the World Trade Centre attacks were the work of George W. Bush.
Microsoft was forced to pull the plug. It was not a result that surprised Endicott.
“This is a well-known problem with releasing AI into the wild,” he said. “The fact that it surprised Microsoft probably just means they had too many young people working on it.” Old hands in the game, he said, are careful to leave a human somewhere in the system, if only to act as a gatekeeper.
“The solution is to have a wizard behind the curtain, or a human firewall reviewing message traffic,” he said.
“In effect, you have brain editors. AI in the wild will always go to the lowest common denominator. There is no perfect AI now and there will be no perfect AI in the foreseeable future. It is always going to need to be mediated by people – either before its launch or during its operation.”
Probably the best-known example of a successful chatbot is Siri, the digital assistant embodied in Apple’s smartphones. Her chief brain editor was a man called Dag Kittlaus, who started developing the program in 2007 and then flogged it a couple of years later to Apple for a fortune.
This week, at the New York TechCrunch event, he unveiled his latest bot, a thing called Viv. The program is billed as much smarter than Siri. It can handle complex questions, deal with supplementary inquiries (while Siri has the short-term memory of a goldfish), and interact successfully with other AI systems.
Siri can easily tell you when Captain America: Civil War is screening at a cinema close to you, and call up the website of the multiplex. Viv, hypothetically at least, can also book the tickets, transfer the money for them out of your account, and call you a taxi.
It’s a mark of just how popular chatbots have become that the birth of Viv was met with marked scepticism by computer industry journal MIT Technology Review. In particular, the magazine’s San Francisco bureau chief Tom Simonite predicted difficulties with the program being able to access the oceans of data its machine-learning software requires.
“Many tech companies very central to consumers’ lives – like Apple, Amazon, and Google – aren’t likely to want to help broaden Viv’s skills,” he wrote. “They have mobile or voice-operated assistants of their own.”
He pointed specifically to Amazon’s “voice-operated assistant Alexa”, which “plugs into services such as Uber and Spotify”, as well as Facebook’s headlong leap into inviting chatbots into its Messenger environment.
But it’s not just the big Silicon Valley players that are punting on a chatbot-led future. Artificial intelligence is also a very active field in India’s burgeoning tech sector.
In 2015 an Indian start-up called Skedool.it launched an office assistant chatbot called Alex. Designed specifically for business, Alex intercepts incoming emails, reads them, and then starts a written conversation with the sender, lining up meetings, processing sales requests and organising goods delivery.
The company estimates that the bot saves the average sales-orientated exec five to 10 hours a week, thus freeing up time to pursue new opportunities, or finish the sudoku in the paper.
It’s already proving a popular acquisition in its home territory and Europe, but Alex represents just one among many Indian forays into artificial intelligence. It’s a tech trend that has kept Marcus Endicott in business for quite a while now.
“I’ve been in Bangalore for the last two years, and in Bangalore the new trend is to convert call centres into AI,” he said.
The Indian model, he explained, relied on having human gatekeepers in the mix – at least temporarily.
“They’ve got, you know, a pipeline, so instead of putting fancy AI at every junction, they just stick a person here and there in the process,” he explained.
“They are making all these digital assistant apps that are full of people inside. They are not 100 per cent human, but they are not a 100 per cent AI either.
“So when a chatbot encounters things it can’t answer, the call will be escalated to a human. The first time it encounters anomalies it will go to a human editor, but then the human will input that data, so it updates the database and then it can handle it by itself the next time.”
Eventually, he said, a point is reached where the AI system has encountered pretty much every possible problem, at which stage its human helpers can be laid off. This implies that anguished customer demands to speak to a supervisor will become ever more forlorn.
By the same token, however, Endicott doesn’t foresee a point at which chatbots take over completely.
“Humans aren’t perfect, and AI is a bit the same way,” he said. “AI is not significantly smarter than the people who program it. So AI is always going to encounter circumstances that it was not prepared for.”
Take heart, therefore. Even in the algorithm-controlled call centres, booking agencies, hotel reservation systems and travel agencies of the near future, there will always be at least one actual person hidden away somewhere. She’ll be the one reading a novel while her chatbot checks out Facebook.