3 Ways You Can Make AI Seem More HUMAN
Human beings aspire toward perfection. You may have heard the popular phrase, “be the best version of yourself.” Wouldn’t it be great to be flawless and never make mistakes? But part of what makes us human is our flaws. “To err is human,” so the saying goes.
The human imagination is incredibly powerful and has produced science fiction that depicts nearly omnipotent, metahuman AI that are vastly more powerful than people. But that type of AI is off in some distant future and as conversational AI designers today, we ask ourselves, “how can we make our AI avatars seem more human?”
It’s a big world out there for a little bot
1. Use Filler Words in Voice Enabled AI Avatars
There are specific ways AI designers have been exploring to provide human touches to their AIs. Google’s Duplex, for instance, has a varied vocabulary for “filler” words. These non-word sounds, like “umms” and “uhhs” add little context clues which communicate to the audience that the speaker is thinking about what to say next. Human beings understand this even without it being explicitly communicated; so, when Duplex’s designers intentionally have their bot verbalize these sounds while it is formulating a response, these design choices add to the illusion that the user is interacting with a real person.
2. Use Pauses and Simulated Typing in text based Bots
Another Human touch AI designers add to their bots is natural pauses. Human beings, when communicating, tend to have natural pauses in between each statement. People are either thinking about what to say next, taking a breath, giving the other person a chance to speak, taking time to type out their statement, as well as a multitude of potential reasons why. AI chatbots make good use of this concept, sending out a statement, and then pausing for a few seconds before writing the next statement. This simulates the natural pauses a human would make, and thus, aids in the illusion of realistic AI.
3. Add balance to give depth to Character-driven AI bots
One way designers can add texture and depth to their AI is by balancing common desirable traits like helpfulness and amiability with quirks and human touches like a slight talkativeness or a mercurial nature. It is key to only add “negative” traits that act to define the AI’s persona and give distinguishing, human-like contrast to the user. The conscientious designer should only create very minor drawbacks that do not contribute to a negative user experience. An AI designer should avoid any negative qualities that give a frustrating interaction with the user. For instance, one could design a likable, rough around the edges, rakish character, who is perhaps slightly “rustic and earthy” at times. If this character is still helpful to the customer, while being interesting by way of not being the generic smiling happy assistant, the customer will want to interact more with the AI. This is the type of realistic AI that users want to interact with; a balanced, character-laden AI. The designer should not give qualities that impinge on the quality of service the customer expects, i.e. helpfulness. However, to design powerfully realistic AI, one must balance the bad with the good. The key is understanding the bad qualities’ purpose as a design choice is to add contrast to the bot’s personality, while maintaining all the tremendous usefulness an AI can provide.
Human touches help us feel real
Flaws are what makes us abundantly human. When we as designers seek to make an AI which simulates a human being, we need to add some flaws, some human touches to aid in the perception by the user that the AI they might be interacting with, might be real. If we could make a bot with very realistic flaws, it would go a very long way in beating the Turing Test. People understand flaws, they live with flaws everyday and they are apparent throughout living life. Perfection is confusing and unrealistic, and so when we encounter fragments of it, it throws humans off.
If you are interested in a beautifully flawed AI persona, please check out Liz, our charming and helpful AI, just don’t really get engaged in her stories, she’ll talk your ear off!
Talk to Liz
