Do embodied agents improve usability?

Question

When you search a web site from outside using a search engine and having the addendum site:mysite.com you have a good chance to find what you want. At least you get close to what you want to find and can either improve your search string of keywords or navigate to related content. The internal search engine work pretty much in the same way, even if the quality of the search result may vary from those from an external search engine.

Link to the agent (swedish)

There are sites that do have embodied agents where you can ask real questions (see image: question on the left, the answer I wanted highlighted on the right) instead of a keyword search. In this way the user have multiple choices of accessing content (regular navigation, external or internal search and the embodied agent). But from what I know, the embodied agent isn't very common on web sites. It might be that it is a tradeoff between cost and gain. Too few users and high cost is often a factor, and there might be other usability improvements, with better effect, than the embodied agent for the same amount of spending. Or there might not be any gain at all having the agent, and that is why they are rare on web sites.

Still one wonders... Do embodied agents improve usability?

---

More to read on Wikipedia:

• Embodied agent

• Intelligent agent

• Artificial intelligence

Answer 1

I'm not qualified to assess whether "putting a human-like face improves usability" independent from the type of the underlying search algorithm.

However, as for the quality of search results, there are efforts to create semantic search tools, that are to better understand what the user is actually searching for. In this paper* several approaches for interaction with semantic search engines are compared with regard to the usability:

• Keyword-based systems
• Form-based systems
• Natural language systems, especially question answering (QA)

The latter can, but does not have to be realized with a human-like wizard. Presumably, a human agent will animate the user to write a complete sentence or a complete question rather than simply typing a keyword. In this sense, the agent visualizes that the underlying search engine can process human language.

Furthermore, the agent can unnoticeably guide the user to ask his questions in the right way. The effictiveness of semantic sarch depends on the format of the question. For example, the agent could animate to give instructions like "give me a list with all books published this month!", by simply asking: "How can I help you?" and he can ask for additional restrictions: "What should I take care of?" Summarizing, in case of semantic search the use of an agent can be seen as an attempt to ensure that the semantic search capabilities are a) really used at all and b) efficiently used.

*Uren, Victoria; Lei, Yuangui; Lopez, Vanessa; Liu, Haiming; Motta, Enrico and Giordanino, Marina (2007). The usability of semantic search tools: a review. The Knowledge Engineering Review, 22(4), pp. 361–377.

Answer 2

My impression is that the benefits of embodied agents are largely theoretical so far, largely due to the current limits in the AI back end. In theory an embodied agent encourages the user to interact with the application as a social entity. User input will naturally include features of human-to-human communication, such as semantics, prosody, and emotional expressions. This is richer input but it’s helpful only if there’re means to capture that input (e.g., microphone, cameras) and an AI that can correctly use such input. The AI needs to then present the embodied agent in a manner that is socially harmonious with the user’s state. That kind of AI is still being developed in the laboratory. When an AI can respond to sarcasm correctly and reliably, it’ll be ready.

Without an effective AI backend, an embodied agent is just a pretty face to distract the user and clutter the display. Worse, it can induce user to engage in social expressions and rituals (e.g., thanking the agent) that are ignored by the UI. That’s wasted effort, a reduction in usability. An embodied agent might encourage more emotional engagement in the content, but then I suspect relevant static pictures of people also accomplish this.

There’s also the risk that users will end up playing with the embodied agent rather than using it. That is what I see happening with most embodied agents that have been tried. This may be potentially gratifying for the user (at least for a while), so in that sense it’s a positive user experience. However, it also distracts the user from the intended task or content of the site or app, so that’s bad for UX.

Finally, there’s the question of whether users want to be social with their machines. I suspect most people just want their ATM to give them some cash. Users may prefer the certainty of hitting the “No” button for “Another Transaction?” over relying the on the machine to infer the end of the session by the user’s body language. I don’t think people need an ATM to share in their excitement of getting a new job. They’ll do that with their human friends at the pub down the street.

Answer 3

Thanks for the interesting topic. I have found a variety of terminology ascribed to the subject, including: chatbot, chatterbot, conversational agent, AI agent, and my favorite.... virtual agent.

There are quite a few services out there to provide your website with a virtual agent, but the question of ROI is harder to track down. I think that (as usual) the real answer to the question is: (drum roll) "It depends." It depends on what type of website you have; what type of users you have; and the goals of both the site and the users.

I did find some articles published in 2012 on the subject. I did not yet find one specific to usability of a virtual agent on a search, but this one may be close: Link Their approach is a little different, namely to install a search agent that would integrate with Facebook, but the findings may be relevant.

Here is one that speaks directly to something I was thinking as I read other answers to your question. Specifically, it hypothesises that older technology users may benefit from a more human-like interaction: Link They found that the more social approach (using a virtual agent) engendered more trust and familiarity for the older users in the study to engage in online shopping.

Here are a few more promising studies: http://www.springerlink.com/content/g87225n9477pg060/

This one explores the concept of giving humor to the AI agent: Link

This one found that the apparent gender of the AI agent affected how users related to it, but not the content of the interaction: Link

This study focuses on the context of medical industry: Link