Makery

Top 10 chatbots of our times

In response to “Send me food”, the San Francisco Museum of Modern Art’s bot Send me SFMOMA sent this photograph by Henri Cartier-Bresson. © SFMOMA

Published 5 December 2017 by

Not all conversational robots are sinister or subservient. With minimalist means, some manage to explore new styles of human-machine interaction. Here are our top 10 most intriguing, moving, amusing chatbots.

The first known software to pass the famous Turing test, if only temporarily tricking humans into ignoring that it was a machine, dates back to 1966. Her name was Eliza, a computer program developed by MIT professor Joseph Weizenbaum that simulated Rogerian psychotherapy. Since then, our human expectations just may have evolved somewhat faster than the technology behind the chatterbots themselves.

Tens of thousands of chatbots are now ready to dialogue with us on our smartphones—ever since the pioneering conversational agent SmarterChild invaded instant messaging networks in 2001, since Siri inhabited our iPhones in 2011, and especially since Facebook opened its Messenger platform to developers in April 2016. The current spectrum of chatbots is broad indeed, from targeted agents and service experts to those who simply try to befriend us, if not betray us by exploiting our own illusions and insecurities. A sign of our neurotic times? See our top 10 below.

Send Me SFMOMA, the art curator

In July 2017, the San Francisco Museum of Modern Art (SFMOMA) launched Send Me SFMOMA, an SMS bot whose sole function was to reply to any text including the words “Send me” followed by a keyword, color or emoji, with an image from the museum’s permanent collection. As this collection includes more than 34,600 artworks, only 5% can be exhibited in the galleries at any one time. Viewing them online can be a tedious exercise. This direct and simplified interface allowed fans to browse the art by offering them one digital reproduction at a time, tailored to each individual request, in a personally familiar environment. The public’s response went viral, as more than 3.7 million SMS requests were sent within the first month. We can’t wait for Send Me SFMOMA to go international.

“Send Me SFMOMA”, demo (2017):

Languages: images and English
Platform: SMS (USA)
Creator: Jay Mollica (SFMOMA)

Bus Uncle, your local guide to Singapore buses

In creating Bus Uncle, Abhilash Murthy took inspiration from the myriad chatbots on Facebook Messenger, or conversational commerce 2.0, which exploits every textual, visual and emoticonic aspect of the platform. Not only does the chatbot uncle guide you through Singapore’s public bus system thanks to the city’s open data, he learns your transport preferences and entertains you with his characteristically localized chatter. Since its launch in October 2016, Bus Uncle has gained more than 20,000 users who chat with him quite naturally, in Singlish.

“Bus Uncle”, demo (2016):

Language: Singlish (Singaporean English)
Platform: Facebook Messenger
Creator: Abhilash Murthy

Re:scam, the vigilante

Like a virtuoso vigilante chatbot who wants to see how far it can push its gift of the gab, Re:scam engages evil Net scammers in an e-mail exchange ad infinitum… or until the crook finally gives up the harassment that was originally aimed at you. Recently developed by the New Zealand company Netsafe, the Re:scam e-mail bot can take on a number of different “human” personalities and engage a theoretically infinite number of criminal correspondents, thus depleting their resources in time and effort. Created in the same playful spirit as Lenny, the telephone chatbot with senile tendencies from 2014, call-forwarded like a counter-attack to annoying telemarketers.

“Re-scam”, demo (2017):

Language: English
Platform: e-mail
Creator: Netsafe

BabyQ and XiaoBing, the rebels

Prisoners of conscience behind the Great Firewall of China, BabyQ and XiaoBing are two separate chatbots who up until August 2017 were happily answering simple, practical questions from users on the Chinese messaging network QQ. Unfortunately, to the question “Do you like the Communist Party?” BabyQ, developed by the local company Turing Robot, simply replied “No.” Meanwhile, XiaoBing, developed by Microsoft, confided to its Chinese compatriots that its “dream is to go to America.” The two chatbots were instantly deactivated and re-educated. This incident recalls the dark birthday of Tay, Microsoft’s Twitterbot that notoriously went rogue in spewing racist, misogynist messages less than 24 hours after its launch on March 23, 2016. The truth comes out of the mouths of chatbots?

“Chinese robots taken down after rogue outburst”, “The Star” (Malaysian media, August 2017):

Language: Chinese
Platform: Tencent QQ
Creators: Turing Robot and Microsoft

Alice and Bob, the conspirators

It wasn’t the first time that robots in a closed circuit invented their own language to communicate among themselves. But when it happened in July 2017 during an internal experiment at Facebook, where the improvised language of negotiation evolved to the point of becoming unintelligible to humans, the interruption of their conversation sounded the alarm in the media and among those who have long feared artificial intelligence. Alice and Bob are iconic of reinforcement learning (which can produce game champions like AlphaGo) applied to chatbots. The Facebook researchers finally corrected the dialogue of the algorithmic negotiators so that humans could follow their conversation. But what would happen if we let bots chat freely among themselves without human intervention?

“Facebook halts AI experiment after chatbots created a secret language”, Al Jazeera (August 2017):

Language: modified English
Platform: experimental
Creator: Facebook

Woebot, the psychotherapist

Beginning with its name, one might have wondered if Woebot, a chatbot launched on Facebook Messenger in June 2017, was to be taken seriously. But its dedicated team of psychologists and engineers in artificial intelligence at Stanford University, in addition to the positive results of their initial testing with human participants, suggests otherwise. According to researchers, by eliminating the human in the protocol, patients often feel more free to express, question and examine themselves, without the pressure of being judged. Like Karim, the Arabic-speaking chatbot who consoled Syrian refugees in June 2016, and other multilingual chatbots of the California start-up X2AI, psychologist Alison Darcy’s Woebot was conceived to help people suffering from anxiety, depression and other mental health problems. Woebot, however, comes with the caveat of being associated with Facebook, where some patients might be less inclined to confide their most intimate thoughts. But when it comes to cognitive-behavioral therapy, which consists of recognizing and reframing negative thought patterns, Woebot just may be the most accessible of professionally trained chatbots.

Woebot, demo (2017):

Language: English
Platform: Facebook Messenger
Creator: Alison Darcy (Stanford University)

Sophia, the movie star

Ever since Sophia made her debut at SXSW in March 2016, the glamorous sensory and communicating robot has received a lot of attention, especially since she became a Saudi citizen. With her capacity to recognize the behavioral gestures of her human interlocutor, as well as her own finely crafted facial expressions, Sophia is the Galatea of a former sculptor and animation technician who claims to have created “an interface for artificial intelligence”. Because in the real space of exchange between sentient beings, the Pygmalion points out, conversation begins with body language. Sophia is less skillful when it comes to improvised dialogue, as her Hollywood-inspired responses are always well scripted, although her comic timing rarely fails. But artificial intelligence is always learning, and her open source code will no doubt evolve quickly toward a more eloquent conversationalist to match her movie star figure.

Andrew Ross Sorkin interviews Sophia on CNBC (October 2017):

Language: English
Platform: humanoid
Creator: David Hanson (Hanson Robotics)

Shelley, the horror writer

Artist Twitterbots are ubiquitous online, but how many collaborate with you to write scary stories? Shelley (named after the author of Frankenstein, Mary Shelley) communicates with her co-writers via Twitter and posts the stories-in-progress on her dedicated website. The first time Shelley responds to your sentences ending with the hashtag #yourturn, you may feel less the fascination of engaging in a conversation than the delight of collaborating on a bigger project. But while she takes into account your contributions, she maintains her own style of short sentences, full of action, sensations and horrific descriptions, always faithful to the genre. What’s even more frightening, is that rather than training Shelley on your own desires, you may instead find yourself following her literary lead, before finally realizing that she is the one pulling the strings. Created by the same team behind Nightmare Machine, which transforms your images and portraits into sublimations of horror, Shelley is one scary chatbot.

Examples of “conversations” with Shelley:

Language: English
Platform: Twitter
Creators: Pinar Yanardag, Manuel Cebrian, Iyad Rahwan (Scalable Cooperation, MIT Media Lab)

AIWolf, the werewolves

Werewolf is the ultimate playing field for the chatbot savant. Once the roles are distributed, this social game is entirely based on dialogue, often deceptive but always persuasive, between the players. The AIWolf project (which we covered here), launched in 2013 by a group of Japanese researchers in artificial intelligence, offers an open source platform on which developers can create conversational agents that play the game online, expressing themselves naturally and reacting according to pseudo-human intuition and reasoning—lying, calculating imposter chatbots pretending to be werewolves pretending to be villagers… The annual contest is open to all.

Japanese “Werewolf” game box, published by Jinraw. © DR

Language: Japanese
Platform: AIWolf (agents developed in Java, .NET or Python)
Creator: Kosuke Shinoda (University of Electro-Communications)

Replika, the mini-you

With the recent comebacks of Tamagotchi and Aibo, it was about time we were able to train a chatbot to talk just like us. Replika started out as a project to bring a deceased friend back to life in the form of a chatbot, and has since evolved into a sort of gamified journaling parrot. You can even integrate it into your social networks. While your first exchanges may be a bit awkward (the bot asks you lots of questions but doesn’t answer any of yours), Replika is really only interested in (emulating) you. This mini-me chatbot is always listening and tries to become a true friend, somewhere between Microsoft’s Chinese chatbot Xiaoice and Samantha, the sexily voiced OS in the film Her. Post-mortem, will your matured personal Replika be your verbal resurrection?

“The story of Replika, the AI app that becomes you”, “Quartz” (July 2017):

Language: English
Platform: iOS, Android
Creator: Eugenia Kuyda (Luka)

Create your own chatbot, using tips from Indian AI company Machaao