Chatbot Fail

Why the new fad might be a waste of time, and why that doesn’t matter

The baroque superorganism known as Silicon Valley—spread across a 100-kilometer axis of California, the undisputed beating heart of the global tech industry and the dominant cultural and economic force of our time—has caught a new fever.

This itself is no novelty. The valley’s venture capitalists, entrepreneurs, titans, and pundits constantly—desperately—search for what Michael Lewis once dubbed “the new new thing.” But the most recent recipient of their collective affection is perhaps the strangest yet. Over the last few months, Silicon Valley has fallen in love with chatbots: software that converses with humans using so-called “natural language,” on messaging services such as Facebook Messenger, Telegram, and Slack.

How does it work? Let’s say The Walrus were to provide a Facebook Messenger search chatbot. You would add it as a contact on Messenger, just as if it were a person, and then talk to it in the exact same way you might text a friend:

Hello, I am Walrusbot! What would you like to find?
Articles about California.
There are 323 Walrus articles that mention “California.” Can you be more specific?
Recent articles.
There are seven articles that mention “California” from the last six months. Would you like me to list their titles so you can select which one to read?

This may seem inferior to simply Googling—but in December, the Wall Street Journal revealed that Google is building a new messaging service which uses chatbots as a search interface. Facebook just announced their “bots on Messenger” platform, which helps other companies run chatbots within Facebook’s wildly popular Messenger app. That same week, Line, Kik, and Telegram, competing messaging apps which each have hundreds of millions of users worldwide, introduced or announced updates to their own chatbot platforms.

Bot fever has now spread beyond the valley. New York City’s Betaworks announced it would invest $2 million in ten chatbot startups. Microsoft introduced its own chatbot, “Tay,” to Twitter where, within twenty-four hours, trolls hacked it into spouting neo-Nazi bigotry. Bloomberg proclaimed “Clippy’s Back: The Future of Microsoft Is Chatbots,” while Wired declared “Facebook Believes Messenger Will Anchor a Post-App Internet,” suggesting that instead of constantly using apps on our smartphones, we will soon spend that time inside of messaging apps, chatting with chatbots.

All this in the space of only a few months. Can it be true? Are chatbots the future? Might we be on the verge of a transformative sea-change to rival the rise of Google?

Almost certainly not. The arguments against a chatbot future are legion, compelling, and backed by both common sense and history. Chatbots are as old as modern computing. The first, Eliza, written at MIT in the 1960s, simulated psychotherapy so effectively that some interlocutors mistook it for human; its very simple software, which mostly just asked open-ended questions while matching a few words and phrases to make it seem like they were logically connected to the user’s previous answer, could be construed as eerily searching and sympathetic.

This is what Sherry Turkle christened the “Eliza effect”: humans unconsciously assume that software which communicates conversationally has much more intelligence and sophistication than is actually present. Inevitably, the software eventually fails to match that assumption, disappointing and frustrating the user who unconsciously expected more. Think of it as a linguistic Uncanny Valley, the notorious effect wherein, when simulated human faces look almost real, they provoke a deep unease, unlike less accurate representations.

But the fundamental flaw of chatbots is not excessive anthropomorphization. It is that they impose restrictive human limitations on software which should transcend those limitations. Until this year, the Valley accepted as an axiom that superb design, in terms of both aesthetics and the visual display of information, was the cornerstone of human-computer interaction. Consider the iPod and the iPhone, devices which were beautiful. It was this world-leading design that propelled Apple into its position as the world’s most valuable company.

Chatbots are a complete repudiation of that idea. A chatbot “interface” is essentially a command line. However intelligent its underlying software may be, that is a massive regression, a deliberate abandonment of visual interfaces in favor of a smarter DOS. We have spent twenty years exploring how computers can visually display information in compelling, effective, intuitive way, and a decade determining how touch interfaces can best serve human needs. Chatbots reject both of those fields completely in favor of a nineteenth-century epistolary model. This seems unwise.

It’s easy to see why their proponents favor them. There’s no need to try to rival Apple-class design quality when your software preemptively rejects the entire concept of design. Amazon, Facebook, and Microsoft are enormously powerful, but to succeed in a world of apps, they must deal with—and chafe at—Apple and Google, gatekeepers of the smartphone duopoly. If bots replace apps, though, this stricture vanishes.

Unfortunately for them, that if seems far too immense to be overcome. It’s true that voice interfaces, à la Amazon’s Echo and Apple’s Siri, are powerful and convenient. But conversational interfaces soon become little more than slower, lower-bandwidth, intensely annoying search engines.

The counterclaims tend to boil down to: “This time is different, because artificial intelligence!” That might sound plausible. After a long fallow period, AI has entered a golden age. So-called “deep learning” neural networks, which (to oversimplify) are structured like our own brains, and are trained rather than programmed, help power Facebook’s news feed, Google’s search engine, and Tesla’s self-driving cars. They perform feats of pattern recognition far beyond what is possible with traditional programmed software.

But a DOS prompt backed by modern AI is still a DOS prompt, needlessly shackled for the sake of useless human imitation, vastly inferior to that same AI powering a full-fledged multi-sensory interface such as your phone. It is instructive that while Facebook has made its bot platform available for others, its own AI-powered chatbot, “Facebook M,” announced with great fanfare last year, is still “a long way” from general release. (They haven’t explained why.) It is instructive that there are precisely zero examples of highly successful chatbots.

You may wonder: what triggered this sudden massive outbreak of chatbot hysteria, and caused Silicon Valley to believe that chatbots might replace Apple’s apps and Google’s search? Blinkered starry-eyed optimism? Madness? Hubris? A refusal to learn from the past?

None of the above. The truth is that the tech industry does not believe chatbots are the future. What it believes is that there is a small but nonzero possibility that society has changed enough, or technology has advanced enough, that this time, somehow, bots just might conquer the world. In Silicon Valley, that is sufficient reason to proceed as if their eventual triumph is a certainty.

What outsiders often perceive as reality-denying valley hubris is in fact a coldly rational response to the laws which rule it; the laws of what Nassim Taleb calls Extremistan, wherein success tends to be either mindbogglingly massive or nonexistent. Silicon Valley is a world in which skepticism has no upside.

Would you invest $5 million in a restaurant which has a 97 percent chance of complete failure? Of course not. But how about a tech startup with that same 97 percent chance of abject failure, but a 2 percent chance that your investment might be worth $100 million, and a 1 percent chance at $1 billion? You should make as many investments like that as you possibly can, because if you make enough, you will more than double your money.

That is the world of valley investment. One its most admired figures, Y Combinator’s Paul Graham, muses: “If most of your ideas aren’t stupid, you’re probably being too conservative . . . the most successful founders tend to work on ideas that few beside them realize are good. Which is not that far from a description of insanity . . . We’ll probably never be able to bring ourselves to take risks proportionate to the returns.”

A new craze hits every year. Last year it was Meerkat and Periscope, apps which allowed anyone to stream live videos to their contacts, which  dominated conversation for months before petering out into near-irrelevance. Bot fever is especially virulent because it’s not restricted to startups. Some massive companies (Amazon, Facebook, Microsoft) want it to succeed for strategic reasons, while others (Google) feel the need to prepare for that success, just in case.

In truth chatbots are almost certainly a deeply flawed concept with a very limited future. They are also an excellent example of how, if you can tell an even remotely plausible tale of world domination, everyone in Silicon Valley, from startups to venture capitalists to Fortune 500 titans, will feel compelled to take you seriously—because this is a land whose very real creation myths consist almost entirely of crazy ideas that conquered the world.

Jon Evans (@rezendi) has written for Wired, TechCrunch, and New Scientist.