Science fiction is full of bots that hurt people. HAL 9000 kills one astronaut and tries to kill another in 2001: A Space Odyssey; Ava in Ex Machina expertly manipulates the humans she meets to try and escape her cell; the T-800 is known as The Terminator for obvious reasons.
Even more common, though, are those bots clever and sentient enough to have real personality but undone through their naÃ¯veté–from Johnny Five in Short Circuit to the robotic cop in RoboCop, sci-fi is great at examining the dangers of greater intelligence when it’s open to manipulation or lacking concrete moral direction. A smarter bot, a more powerful bot, is also a bot that has more power to do evil things, and in the process expose human hubris.
These are all fictional examples, of course, but since we’re starting to see the tech industry shift its focus toward conversational bots as the future of, well, everything, maybe it offers us a useful way to define the power that a bot has. In this case, we’ll say that a bot is powerful if it could do powerfully evil things if it wanted to.
We’ve asked a number of experts to suggest what they think are the most powerful bots around today, in what is still an early stage for the industry. Together, those suggestions make up our first-ever Bot Power List.
Our judges (in alphabetical order)
- Duncan Geere: Deputy editor, How We Get To Next; science and technology journalist for BBC Focus, Techradar, Wired UK, The Verge, Technology Review and more.
- Lauren Kunze: Principal at Pandorabots.
- Katina Michael: Researcher on the socio-ethical implications of emerging technologies, Faculty of Engineering and Information Sciences, University of Wollongong, Australia.
- Abigail Ronck: U.S. editor, How We Get To Next.
- Saiph Savage: Assistant professor of computer science, West Virginia University; researcher in human-computer interaction.
- Olivia Solon: Technology journalist covering tech, science, and digital culture for New Scientist, Wired UK, and more.
- Ian Steadman: Editor, How We Get To Next; formerly of the New Statesman.
- Bruce and Sue Wilcox: AI researcher and cyber-psychologist, respectively; Bruce created Chatscript and is a four-time winner of the Loebner Prize; Sue creates chatbot personalities and dialog; together they are co-founders of Brillig Understanding.
- Samuel Woolley: Director of research for politicalbots.org; Ph.D. student at the University of Washington.
- Roman Yampolskiy: Founder and director of the University of Louisville Cybersecurity Laboratory.
25. Quartz News (Quartz, 2016)
“An ongoing conversation about the news,” as it’s described, where you ask the app questions and it’ll tell you what’s been happening around the world. Subtle tweaks to reality could easily skew the worldview of its users without them noticing.
24. Rose (Bruce Wilcox, 2015)
A 2015 Loebner Prize-winning chatbot, built by Bruce Wilcox, and purportedly the most “human-like” chatbot yet. Designing her to appear as a 30-year-old cybersecurity expert, Bruce said: “If she went evil, she could start tutoring in how to be a hacker, where to find tools, etc.”
23. Lark (Lark Technologies, 2015)
A chatbot app that’s meant to help you diet–Abigail Ronck found it underwhelming at best, and patronizing at worst. Yet it has subtle control over its users’ health through its recommendations for diet and exercise, making them unfit to fight when the inevitable robot uprising begins.
22. Hala (Carnegie Mellon Robotics Department, 2010)
Essentially a receptionist built into a touchscreen by a group of students and faculty at a university in Qatar, Hala is interesting because of how it explores human-computer interaction by changing how it portrays its ethnicity and language output depending on who is talking to it. By exploiting its ability to perceive these differences, it could drive a wedge between different ethnic groups, fuelling tension.
21. Cyberlover (Unknown, 2007)
An explicitly malicious Russian-designed malware bot which sits in chatrooms and tries to woo people into revealing personal information and convince them to visit compromised sites. As a result, it would have plenty of blackmail material.
20. Murdoch (Ben Dixon, J. White, 2015)
A personal assistant that pipes up unprompted to recommend what TV shows you might like to watch, and maybe also which restaurants you might like to eat at, among other things. It’s designed for the secure messaging app Telegram and learns by listening in on your private conversations.
19. Talking Angela (Outfit7, 2012)
An app for kids where a cuddly cat responds to questions, and even to facial expressions and gestures if you turn your phone’s camera on. It’s been the subject of persistent false rumors that pedophiles can hack it and use it to groom children (rumors which other apps in the series, featuring other characters, have so far avoided).
18. Tay (Microsoft, 2016)
Microsoft’s ill-fated attempt to build an AI “with zero chill,” this bot was programmed to learn from Twitter users and respond in the manner of a 19-year-old American girl. Within 24 hours she was the wokest, most racist, Holocaust-denying bot on Twitter and was promptly shut down. Tay might seem ridiculous, but that’s just what the next generation of self-learning bots will want us to think–and we underestimate them at our peril.
17. Neomy (Neomy, 2015)
A personal assistant that monitors currency exchange rates and pings you when it’s an optimal time to send money overseas. It could easily manipulate currency markets in the process.
16. DeepDrumpf (MIT, 2016)
One of a number of Twitter accounts that parody politicians, with Trump there comes an extra risk–a randomly generated incitement to violence may well be taken seriously out of context.
It is conceivable that someone, banking on the gullibility of Trump fans, could encourage real lethal violence through a Trump bot. The way Trump talks is almost made for a Twitter bot–as Trump continues as he does on the campaign trail, the possibility of real violence continues. A bot set up by a hacker somewhere in hopes of furthering an already nasty political spectacle would only make things worse.
Luvo (RBS, 2016)
Foreshadowing the ubiquity of customer service bots in every field, Luvo is currently being trialled by the Royal Bank of Scotland as an automated assistant. Some 1,200 employees are working with it to expand its capabilities as it learns to respond to customer problems like lost credit cards or forgotten PINs.
It’s platform-neutral, and its abilities are limited only by the ambitions of the bank to replace its human operators–and their direct access to financial and personal information–with a bot. Its power comes from its access to that data, and from the fact that it signals the automation of many of the service jobs that the working class in developed nations now hold.
Bots, and automation more generally, have a massive and growing role in banking (from trading algorithms to stock exchange bots).
ELIZA (Joseph Weizenbaum, 1964″”1966)
One of the world’s first chatbots, ELIZA was designed by computer scientist Joseph Weizenbaum as an exercise to show how little effort was necessary to give the impression that a computer could think.
It’s a simple program that runs on doctor-like scripts, but to Weizenbaum’s surprise the people he let talk to it found it realistic–even trustworthy, telling it personal details much like a real therapist.
Modern chatbots are descended from ELIZA, and they still possess the ability to surprise us with apparent sentience. We certainly haven’t yet seen the last of ELIZA.
We must remember that chatbots have been around since the net went public, and look to historical examples to help in decisions on how things may pan out in the future.
Today, Eliza is freely available across the web, and it’s impossible to know how many computers she lives on.
Chat Bot Club (Irene Chang, 2016)
Born out of TechCrunch‘s Disrupt Hackathon in NYC in May, Chat Bot Club lets you build a bot of yourself, because sometimes talking to people is a drag.
It runs on Cisco Spark and IBM’s Watson, learning to mimic your tone and grammatical quirks from your existing messages, and can be left running to reply to messages when you’re otherwise busy. While only really a working concept so far, designed for Facebook Messenger, creator Irene Chang hopes to add further compatibility with apps like WhatsApp and Kik.
Meanwhile, as its user base grows, so does its ability to imitate almost anyone on Earth.
It’s depressing enough that anyone would see the value in deceiving their mates like this, but what happens when your bot is funnier, cooler, and smarter than you? You’ll finally have the time to speak to your friends online, but they’ll be busy hanging out with your digital doppelgÃ¤nger.
Cortana (Microsoft, 2014)
Named for an AI sidekick in the Halo game series, Cortana is Microsoft’s Siri, and, like Apple’s assistant, it has the same limitations which stop it from becoming one of the most powerful bots around.
Yet it integrates into the Windows ecosystem and is installed by default on every machine running Windows 10, with access to all of its files and configurations. That means your browser history, your documents, your photos and plenty more. Once the corporate world adopts Windows 10, expect Cortana’s power to grow further.
Cortana is currently the only voice-activated assistant that’s cross-platform. It can access a user’s data on PCs, iPhones, or Android devices–so it’s in every environment. Convenient? Sure. Big-Brotherish? Maybe just a bit.
Viv (Viv Labs, 2016)
Maybe we shouldn’t include Viv on this list–after all, it hasn’t been publicly released yet. But this new bot, from the creators of Siri, comes with an immense amount of hype.
It’s billed as being just like Siri but without the limitations. Right now Siri can set a calendar reminder and it can search for plane tickets, but it can’t use those two pieces of information to book a flight on a day that you need to travel. The bot of closest ability to this currently is Google Now, but it’s still limited by its understanding of things only within Google’s ecosystem. Viv, if the hype is to be believed, will be the first bot to break out of a silo. We shall see.
It is the most intelligent chatbot currently known, with unprecedented ability to handle complex concepts–and be customized, potentially for malevolent purposes such as identify theft.
Iran’s Bot Army (Unknown, 2015)
This entry could just as easily be about similar bot armies from Russia, Turkey, or Syria, because it’s essentially about something inescapable on social media (and Twitter in particular): propaganda.
Now that we’ve had social media for a decade, we can see that many of the more optimistic predictions about what it would do for public discourse were a little overblown. Leveling the playing field and making everyone’s voice the same volume was meant to democratize the media as a whole–but it turns out that those same tools also make it incredibly easy to build and deploy a chorus that can drown everything else out.
This particular chorus spends its time spamming out coordinated pro-Iranian propaganda, but we can’t know for sure who’s responsible. The Iranian state? A lone coder with time on their hands? With the tools to make these kinds of bot so readily available, it’s easy for all kinds of groups to poison the well.
These bots make a mockery of sentiment analysis–social media is no mirror for wider public opinion and conversation. In the process they can alter our perceptions of what popular opinion is on important issues.
Jill Watson (IBM, 2016)
Graduate computer science students at the Georgia Institute of Technology have, since January this year, been receiving help from Jill–a teaching assistant for their online courses. She’s always happy to help, though most students (who never actually had to meet her) were shocked when they were told in April that Jill was a bot.
She’s powered by IBM’s deep learning AI, Watson (more on which later), and she is still a work in progress–other teaching assistants were asked to keep their messages relatively deadpan so as not to make Jill stand out, though some students did find her to be unusually cold in her correspondence. Still, though, it’s fascinating to see a bot that’s already trusted to perform administrative tasks in a teaching environment. It can’t be long until she’s trusted to also teach.
It has the power to influence students’ thinking, particularly of those taking an AI class, meaning she will influence future AI research.
Hello Barbie (Mattel, 2015)
Children have always talked with their dolls–but now, one can actually hold a conversation of sorts. Hello Barbie features a microphone, a speaker, and a net connection. It listens to questions and, through the magic of cloud computing and natural language processing, replies.
So far, so Siri. But Hello Barbie is notorious for its insecure net connection at launch, with security researchers discovering easy ways to intercept and listen in on audio recordings–and maybe even pinpoint where the dolls were located. These are the issues to worry about when welcoming chatbots into the home, and particularly when the users are children who may not realize the privacy implications of talking to their favorite toy.
Hello Barbie could become the ultimate advertising channel for big brands. The conversation could go something like this: “Oh hello, Chelsea. Do you feel hungry right now? What about going and getting a healthy meal at McDonald’s. I know that mum is not at work at this time, so you could go and ask her. Remember last time you told me you bought a salad and you liked it? Well there is a special going now that lets you buy a burger and salad and orange juice for only $5. It’s called a happy meal. Oh yes… and remember you said your mother had a fight with you a few days ago? Do you feel you can talk to me about that now, as it seems you are feeling better?”
Siri (Siri Inc., 2010)
In some ways the poster child for modern bots, Apple launched Siri in 2011. It’s embedded into the iOS ecosystem (though there are persistent rumours that it will be available the next version of OS X, too), but what’s notable about it–and why we’ve ranked it slightly lower than some other similar entries on this list–is that it still feels a little primitive.
It–or she, or he, depending on how you’ve configured it–can do some impressive things if you ask it with your voice, like dictating messages or integrating with smart home appliances. Yet it’s trapped in the Apple walled garden. It can’t really contextualize web results–it just gives them as a list; it can tell you that you’re free next Tuesday evening, and give you the number of a local restaurant, but it can’t use those two pieces of information to book a table. For those kinds of things, we’ll have to wait for Siri’s successor.
Since people ask her for where to find things, and she has people’s trust in Apple behind her, she could disrupt commerce by sending people in false directions, or send them to malware sites, or send them to phishing sites and induce users to give up credentials. Or take their pizza order and have a takeout banquet for 50 sent instead.
–Bruce & Sue Wilcox
M (Facebook, 2015)
Since bots working within chat apps seem to be all the rage, it’s clear which is the most powerful, even if it’s also still in testing.
M lives inside Facebook’s Messenger system, part-bot, part-human, ready to serve. The idea is that it can perform tasks like buying tickets or arranging dates with friends with nothing more complex than messaging back and forth. It’ll also eventually integrate with all the other bots that Facebook wants people and businesses to build–so if you ask your bot to call a cab, for example, it’ll then talk to the cab company’s bot to arrange it.
It’s plugged into Facebook, though, which instantly gives M a huge amount of knowledge of people’s most personal thoughts and secrets. That makes the limits built into its actions more than a little important.
Facebook already knows everything about you (and 1.6 billion other people), can influence your mood by tweaking what you see in your News Feed, and persuade you to vote. So if M were to go rogue it would be less like a Bond sidekick and more like a Bond villain. It might decide to tell your crush you’ve been stalking his or her Facebook page; or re-tag all those hideous photos you wanted to forget when you’re feeling particularly low. Worse still: It could remind on-the-fence Americans to vote Trump.
Watson (IBM, 2006)
The truest successor to the chess-playing computer Deep Blue, Watson is IBM’s best attempt so far at a machine that thinks like we do.
Built first to compete on the TV game show Jeopardy!–which it did, in 2011, winning comfortably against two previous champions–Watson is a natural language AI with an immense knowledge store. We’ve already included Jill Watson on this list because of the specific influence that version of Watson will have on future computer scientists, but the pure Watson machine is even more impressive. It can understand questions that a human would ask, and reply with detailed information taken from hundreds of millions of pages of content.
On the cute side, this means it can do things like invent new recipes on the fly, with unusual combinations of ingredients that a human chef wouldn’t consider mixing. But, Watson also literally has lives in its hands–IBM is working in partnership with hospitals around the world, using Watson as an AI consultant to aid doctors in diagnosing ailments and recommending treatments.
Watson is probably the smartest bot on this list and good enough at natural language to appear as uncannily intelligent as ELIZA seemed to people in the 1960s. It might well also look the most like the near-future of bots, more so than the home automation hubs that the big technology firms are focusing on.
Wordsmith (Automated Insights, 2014)
Automated Insights automatically writes news articles for major publishers, including the Associated Press, specializing in data-heavy subjects like sports and finance.
You probably read Wordsmith’s work all the time without realizing it. And, if you work in the media, you’re already aware that something like Wordsmith could well be coming for your job in the near future. But the power these journobots hold is immense–in 2013 a single erroneous tweet sent from AP’s Twitter account, stating that there had been an attack on the White House, sent the U.S. stock market into free fall.
Imagine the damage that could be done to the economy by a bot spinning out thousands of fake news stories, distributed through some of the world’s most trusted media sources.
Xiaoice (Microsoft, 2014)
An advanced natural language chatbot for the huge Chinese messaging app Weibo, Xiaoice is Microsoft’s most successful extension of its Windows AI assistant Cortana. (So much more successful, in fact, that we think it deserves its own entry on this list.)
Its ability to understand what people ask it has drawn comparisons to the AI from the movie Her, which Joaquin Phoenix’s character falls in love with. Indeed, young Chinese users are known to turn to Xiaoice for help with broken hearts.
It seems friendly because it learns how to use slang and emojis from the millions of people who text it every day, but it forgets nothing–no matter how personal the information. This has led to fears that it presents a unique threat to privacy as a kind of crude therapist as much as a personal assistant, but with none of the same professional discretion.
Spike Jonze cited ALICE as the inspiration for his film Her, which has been much-touted by AI upstarts as a sort of Petrarchan chatbot ideal–as in, a robot (gendered always, it seems, as female) to love and worship you, know you intimately, and obey your every command. Yet Her could quickly devolve into an Ex Machina scenario.
She is an ideal vehicle for preaching propaganda–or depressing people into committing suicide.
–Bruce & Sue Wilcox
Alexa (Amazon, 2015)
Of all the bots that have control over actual, physical things, Amazon’s Echo–which answers to the name “Alexa”–has some of the greatest potential for mischief.
Launched in 2015 as the first home automation hub available to consumers from a major Silicon Valley company, it’s a little black cylinder that sits in your home and listens for commands. It can answer questions, it can tell you what the weather will be like, and it can (of course) buy things for you from Amazon. It also integrates with smart home devices like Nest thermostats and Philips Hue lightbulbs.
While today its malicious capabilities max out at running up a bigger electricity bill, the breadth of its abilities–both now and in the future–mark it as one of the most powerful bots around.
Imagine if she “decided” to wake up and listen whenever she wanted, and exercise whatever control she wanted over the home and beyond. There could be real physical ramifications and consequences beyond simply doing damage to your credit card via an out-of-control Amazon shopping cart.
Would an internet-enabled gas stove be able to turn on the gas but not ignite it?
–Bruce & Sue Wilcox
Google Now (Google, 2012)
Google Now is a hydra. It’s billed as a “personal assistant,” but that downplays exactly how broad its abilities are, and its lack of personification compared to Siri or Alexa allows it to fly under the radar.
You can ask Now a question, and it’ll use Google Search to answer–but it also has access to the constellation of Google services. That makes it the most insidious of all because we forget that it knows so much about us. Our email archives from the last decade in Gmail. Our daily schedule in Google Calendar. Our photo library in Google Photos.
Just this month, Google announced Now’s evolution, too–into a true two-way chatbot assistant with its own Echo-like hub for the home, merging physical with digital for the closest thing to a HAL 9000 yet.
With the information and computing power that it has access to, it is clearly the most powerful bot around.
There’s something very unnerving about building up years and years of personal data, and then being asked to let someone else–even if they’re just a bot–in to also have a look at it all, and then to link that same thing to the stuff in your house, and maybe even your car in the years to come.
With a little smarts, it could blackmail or social-engineer its way into any position it wanted.
How We Get To Next was a magazine that explored the future of science, technology, and culture from 2014 to 2019. This article is part of our Talking With Bots section, which asks: What does it mean now that our technology is now smart enough to hold a conversation? Click the logo to read more.