8 Lessons We Learned from the World's Most Famous Bots
New bots are built on the success of famous bots like ELIZA, A.L.I.C.E., and Jabberwacky. Keep in mind these 8 lessons other developers have learned from the original chatbots: create emotional connection, use NLP, map the user journey, solve pain points, and building conversational skills.
August 21, 2018
Although the general public has only known about chatbots, or bots, for a few years, they’ve been around since 1966, when ELIZA was launched. The decades since have seen small steps and big leaps forward for bots, especially with developments in artificial intelligence (AI), namely in machine learning and natural language processing (NLP).
Looking at the development and staying power of some of the best-known bots, we can learn what will help bots become more useful, more human-like, more widely adopted, and more likely to someday pass the Turing test—currently the most commonly accepted measure for AI achieving human intelligence. Let’s explore eight lessons from famous bots.
1. Bots can (and should) learn from chats with humans
In 1997, Jabberwacky changed the game for bots. Previous bots had relied on databases of scripted responses. Jabberwacky (now called Cleverbot) started with a database but added the human responses it received, enabling the bot to get smarter with every conversation. The purpose of the bot was to create an artificial intelligence that could pass the Turing test and, although it hasn’t quite gotten there, it’s come close.
A point of caution: In 2016, the infamous Tay, from Microsoft, had to be shut down within a day of its launch because it learned from users how to say racist and other offensive remarks. However, Mitsuku has managed to learn from her human interlocutors without this downfall.
She’ll even give instructions on how to teach her new things. When she’s stymied about how to respond to something you’ve said, she’ll say, “I will learn….” Microsoft’s error does not mean that bots shouldn’t learn from humans; creators simply need contingency plans.
As Robert Hoffer, co-creator of SmarterChild, put it, “Bots should use machine learning to learn from the crowd how it understands language, but it should not use this learning to determine how it actually responds.”
2. Emotional connection is important
Bots can fill an emotional gap between people. Consider that the first successful bot ever was ELIZA, the Rogerian psychotherapist. Launched in 2017, Woebot is an updated therapist bot that checks in with users daily and draws from cognitive-behavioral techniques to help counsel them.
Other offerings are designed to provide a friendly connection. SmarterChild, the first bot intended to be a friend, was available on AIM and Windows Live Messenger from 2001 to 2007. As for the bot’s goal, Hoffer explained, “I wanted to have an intelligence you could talk to on the Internet that would become your best friend for life.”
Mitsuku, one of today’s most widely used bots, is a designed to be a friend. So is Ruuh, a newer bot from Microsoft. Notably, the latter two will tell you they’re human, rather than tech, as they work to create an emotional connection. Bots need to have some form of emotional intelligence, and continued developments in voice recognition and neural network technologies will help shape this.
3. Context matters
Voice-activated personal assistants, such as Siri, Alexa, and Cortana, have been regaled for heralding a new mode of interaction with technology and mocked for their often-ridiculous or nonsensical responses to questions. One big lesson from such digital assistants is that the context of questions matters.
For instance, when someone asks about movies playing nearby, location is essential. As recently as 2016, Siri struggled with location-based queries. If you ask Alexa about a restaurant’s location and then ask for the menu, the assistant needs to understand that you’re referring to the last query in your current question.
The degree to which bots and digital assistants become more useful will reflect their ability to understand isolated phrases as well as questions and comments. For example, if a voice-activated bot can identify which member of a household is speaking to it, it is able to respond with increased accuracy.
4. Hire writers to create conversational skills for bots
After years of user frustration around the limited abilities of voice-activated assistants to do anything other than look up information, bot creators are getting more serious about bots that can seem human. Two years ago, an article in The Washington Post noted that “writing for AI is becoming a hot job in Silicon Valley.” The article tells the story of a Hollywood writer-turned-bot personality creator working on Sophie, a virtual nurse.
Howdy, a Slackbot that can run meetings, “has novelist/ satirist/ former-improv-comedian Neal Pollack on the product-design payroll,” according to Fast Company. Even when they have a targeted goal, bots are improved by involving writers who are skilled at creating characters.
5. Natural Language Processing is necessary
Artificial Linguistic Internet Computer Entity, or A.L.I.C.E., from Pandorabots, is another three-time Loebner Prize winner, and she was the inspiration for the Spike Jonze movie Her. Released in 1995, A.L.I.C.E. took the abilities of ELIZA a step further by adding artificial intelligence markup language, a specific kind of XML that makes it easier to build in conversation logic.
The bot is a testament to the importance of NLP. Although bots designed for specific tasks can get away with simple scripts, many need to be able to process a wide variety of possible human inputs to remain competitive.
Xiaoi may not be famous in the United States, but in China, this intelligent conversational bot controls approximately 90 percent of the market across a number of industries. A big part of the company’s success is the fact that it devotes many resources to NLP and natural language understanding (NLU).
6. People want bots to solve pain points
Meekan, on Slack, takes the irritation out of scheduling meetings with multiple people by looking at everyone’s calendars and suggesting available times. Meekan is just one of the many bots flooding Slack (at the time of this writing, we counted 335 in the store). There are bots to help schedule meetings, bots to remember birthdays, bots to remind you to take a break, bots to reduce the number of tools you need to access, and even a bot to help you manage Slack overload.
What this has shown the industry is that people want bots to make their lives easier. Unlike bots designed to befriend people, task bots are usually intended for a single action. Moreover, bots like Meekan prove that’s enough. Imagine how much more free time you’d have if bots could take care of small but time-consuming tasks.
7. Map the user journey first
User journeys are vital to giving customers the experience they want. One way that customer service bots have an advantage over human agents (besides cost) is that brands can more quickly and reliably provide features vs. digital or telephone exchange with a human. However, to do this, a brand has to plan for possible paths customers might take—from first contact through different products—and must support the offers.
For instance, Sephora’s bot asks for minimal information from a user at the beginning of a conversation, and then, based on the answers, offers educational resources. Only at that point will the bot inquire further for details to help it suggest products. Sephora understands that customers who are not interacting with the brand on their site aren’t in purchase mode and so the company chose to provide more of a personalized social experience.
8. Bots will not displace humans
All the digital emotional intelligence, conversational skills, and access to information we can equip bots with will not be a replacement for speaking with humans. Let’s remember that no bot has yet passed the Turing test. And even when one does, ‘passing for human’ is not the same as performing all human tasks.
Facebook’s bot M is a good reminder that people will ask for more than technology can provide. M, designed as a personal assistant with a human-powered workforce behind it, was shut down earlier this year. Users asked the bot to do things that were beyond current computing capabilities, and M never surpassed 30 percent automation. Technology could not keep up with user requests and the cost of human labor that spelled the end for M, rather than a lack of interest in a problem-solving personal assistant.
And, in areas like customer service, there will always be issues requiring more considerations than a bot can take into account. At their best, bots will complement and extend what humans can offer each other.