Blog: 5 Ways to Design an AI Assistant that isn’t Socially Awkward


Voice assistants, while intelligent, helpful and fast, demonstrate socially awkward behavior: they have to be told when they can enter a conversation, they cannot perceive non-verbal cues and they are emotionless. Just like in the real world where people avoid potentially awkward situations at all costs, so too will they avoid AI assistants if they make us feel weird when we engage with them. The true value of AI assistants will be unlocked only when they can better understand human emotions and possess situational context to not only respond to, but proactively satisfy our needs.

Now that this next generation of AI assistants are becoming embodied by humanoid avatars that are better at understanding people, such as our AI-driven virtual character Millie, it’s more important than ever to eliminate awkwardness in social AI. When you perceive an AI with eyes looking at you, it’s a whole new ball game in terms of potential risk and reward. Given these promising, exciting technological advancements, we’ve compiled a list of 5 ways you can ensure you’re creating socially savvy, non-awkward AI assistants.

1. Avoid a character design that’s too realistic

Who says that your genius assistant must look like a human being? From talking toys to joke-cracking animals, our entertainment culture has primed us to accept artificial characters that can communicate with us. Add to this that humans get uncomfortable when seeing their embodied likeness and you’ve got a strong case to avoid creating realistic-looking humanoids.

Back in the 1970’s, roboticist Masahiro Mori found that even though there is a positive relationship between the realness of virtual avatars and their acceptance by humans, there’s a point where we become overly-attuned to the artificialness of the character despite its realism and sense that something is “off”. It is here that we’ve reached what Mori termed, the uncanny valley. As long as characters are consistently unrealistic, we’re ok — but as soon as ultra-realistic features blend with other unrealistic features, it confuses our sense of reality and the resulting sensation is an uncomfortable, weird feeling. This is a trap to which virtual characters can easily succumb because, as the technology currently stands, we cannot yet render each element of the character’s makeup evenly, causing mixed realism. Therefore, you should avoid blending realistic and non-realistic features (i.e. detailed skin with large, non-humanlike eyes, etc.).

2. Use Childlike Features

Studies show that it is a good idea to include some childlike features in your avatar design. It turns out that humans respond positively to evolutionary features connected to survival and reproduction. These features evoke protective and empathetic instincts in humans even if they originate from non-human entities such as avatars. The feelings that these features cause form the basis of mutual understanding and cooperation. Therefore, why not include snub noses, small chins and round heads in your avatar design?

3. Focus on the eyes

In our last post, we discussed the emphasis humans place on eye contact when interacting with fellow beings and how it helps create emotional connections between people. That’s why it’s critical that you animate your virtual character’s eyes so that it doesn’t get confused for a zombie. Your character should be able to express behavior, intent and positivity from its eyes, as humans do, to help with seamless communication. The next step is figuring out how to render the communicative abilities of the eye in virtual beings. This will be a key development as the technology isn’t there yet.

4. Embed diversity in your design process

Avoiding the awkwardness of the uncanny valley is being sensitive to small subtleties. There are few things less awkward than noticing insidious deviations when you interact with avatars. For example, we wrote about how the team at Microsoft who built Clippy weren’t sensitive to the views of women, most of whom felt he was leering at them — a subtlety missed by the male-dominated team. To avoid issues like this, make sure your team is diverse to ensure you are thinking through all of these design parameters. You can also opt to design a gender-neutral character.

5. Privacy is Key

As our AI assistants become increasingly humanlike, trust is becoming an ever more important trait that will determine the successful adoption of your product. Soon, when we are interacting with our assistants as we do with other humans, we’ll expect them to be trustworthy as we do of our close friends. Make sure your product isn’t exploiting your customer’s trust by withholding its functions or performing secret ones. This destroys trust and, thereby, adoption of AI assistants in general.

Socially awkward AI, while mushy sounding, can be eliminated with scientifically-based product development. As our AI assistants begin to understand us better, product designers must fastidiously ensure that, rather than create a dispassionate tool that responds exclusively to commands, deliver a helpful assistant that interacts with us just like one of our friends.

For a scientific analysis of how to avoid the uncanny valley in your product design visit: https://interactions.acm.org/archive/view/september-october-2018/avoiding-the-uncanny-valley-in-virtual-character-design

Source: Artificial Intelligence on Medium

Leave a Reply

Your email address will not be published. Required fields are marked *