Published on <a href="http://ift.tt/2nSFSnC">| Prototyping: From UX to Front End &mdash; Medium</a> Zoe K 
<h4>Why chatbots may not fly</h4>
Chatbots, an explosively popular concept of a conversational interface, are designed to leverage the most ubiquitous and natural human capability: human conversation and make the input as seamless as possible. With help of artificial intelligence and natural language processing chatbots cater to various user needs, including: search queries, navigation, task requests or customer support.
Despite conversational agents being a half a century old concept, real life chatbots as a service are a novelty, and as such they go through what may seem as a series of understandable developmental setbacks. <a href="https://chatbot.fail/">[1]</a>
However instead, it may be an indicator that, with the chatbots, Tech industry have bitten more than it could possibly chew. When we talk about chatbots and their opportunities, we fail to evaluate the whole scope and to mention the underlying complexity of human interactions. The same complexity that we, humans, take for granted.
Essentially, human conversation is so complex, and we parse it with such ease, that it sets up a very high expectation for the chatbot and leaves an extremely small margin for error.
The complexity of human conversation
Conversation is one of the core capabilities that defines us as intelligent beings. It was perfected in the course of over 300,000 years <a href="http://ift.tt/2n6devl">[2]</a>, and its sophistication reflects its age.
Conversing is such a natural activity that in the majority of cases we don’t even have to navigate it consciously.
Indeed, the levels of organisation involved in the conversation are multiple; for example: psychological (perception, understanding and emotion), motivation (rationality and agency), social: (culture, role, activity, language et cetera). <a href="http://ift.tt/2nSMen3">[3]</a>, <a href="http://ift.tt/2n6vpB4">[4]</a>
<img alt="" src="http://ift.tt/2nSKBWc">
This is a task that Tech industry was working on resolving for decades, and it is still far from synthesising a human-like enough conversational agent.
<blockquote>To summarise, engaging in complex language behavior requires various kinds of knowledge of language: …</blockquote>
<blockquote>Morphology — knowledge of the meaningful components of words</blockquote>
<blockquote>Syntax — knowledge of the structural relationships between words</blockquote>
<blockquote>Semantics — knowledge of meaning</blockquote>
<blockquote>Pragmatics — knowledge of the relationship of meaning to the goals and intentions of the speaker</blockquote>
<blockquote>Discourse — knowledge about linguistic units larger than a single utterance <a href="http://ift.tt/2n6kWpv">[5]</a></blockquote>
The scope of work is so big that even the most robust AIs can only tackle this scope by parts. To solve such a huge problem, we need to have multiple people developing multiple AIs and connecting them to work together as a system. This suggests a research budget that is more of a game for Google, or DARPA, rather than any given startup team.
Our Expectations
It is because chatbots are trying to simulate the conversation, we subconsciously expect them to understand our messages like human beings do, and react accordingly. This sets the bar very high, and obviously, not having their expectations met increases the frustration of the users and may jeopardize their whole experience with the service.
Factors that elevate users’ expectation are exactly those things that enable us, human beings, to converse so effectively, and that computer systems are often unable to reproduce.
Predicting things and taking turns
<img alt="" src="http://ift.tt/2nSMgLH">
In Human-Human Interaction, we do a few things that chatbots seem to fail at reproducing: we send messages in tiny bundles, we take turns in exchanging them, and we are very good at predicting the timing of next turn and even content of the next turn of our peer.
<blockquote>The bulk of language usage is conversational, involving rapid exchange of turns.</blockquote>
<blockquote>…(T)his transition between speakers is generally more than threefold faster than language encoding. To maintain this pace of switching, participants must predict the content and timing of the incoming turn and begin language encoding as soon as possible, even while still processing the incoming turn. […]</blockquote>
<blockquote>This fast pace holds across languages, and across modalities as in sign language. It is also evident in early infancy in ‘proto-conversation’ before infants control language. Turn-taking or ‘duetting’ has been observed in many other species and is found across all the major clades of the primate order. <a href="http://ift.tt/2n6mJuE">[6]</a></blockquote>
Being learning systems
<img alt="" src="http://ift.tt/2nSY6VR">
We largely take the human ability to record and recollect the information we share with them for granted. During the conversation, a human being intends to communicate the message in a way of a progressive disclosure and expects the counterpart to memorise previous messages as the conversation progresses. We don’t expect our peer to forget our name and other information the moment after we told them.
Expecting novelties
<img alt="" src="http://ift.tt/2n6eBKK">
In Human-Computer Interaction we want the system to behave exactly as we expect it to. These expectations are conditioned by our previous experiences with various computer systems and our experiences with this particular system. Whenever the system behaves incoherently, we consider it a bug.
In Human-Human Interaction, however, the margin for novelties is quite stretchy. We are used to receive seemingly incoherent or unexpected bits of information from our counterparts. These bits may signal the change of topic or be completely random and dismissable, but we don’t reckon something went wrong with our counterpart. <a href="http://ift.tt/1yomBGQ">[7]</a>
In that sense, the conversational context of some chatbots may be confusing, or present a mixed message.
Anthropomorphic Design
Apparently, talking like a human could be a part of the problem, because it sets up an unfair expectation.
<img alt="" src="http://ift.tt/2n6bUJb">
But not talking like a human could be an issue, too.
<img alt="" src="http://ift.tt/2nSHz4m">
<blockquote>People prefer to collaborate with a robot that is able to respond socially. On the other hand, robots that overuse anthropomorphic form, such as humanoids that almost perfectly resemble a human but still remain unnatural copies, can have a contrary effect and evoke fear or rejection. <a href="http://ift.tt/2n68mGM">[8]</a></blockquote>
In this particular case the author refers to a well-known phenomena of the Uncanny Valley <a href="http://ift.tt/L6R05i">[9]</a>. Arguably, this could be applied to chatbots, too, despite the differences between facial expression pattern recognition and conversational pattern recognition.
In other words, the chatbots better not try to imitate humans, but should not behave all robot-like.
<img alt="" src="http://ift.tt/2n6nOCK">
Recap
<ul>
<li>Human conversation is incredibly complex and is currently impossible to simulate realistically.</li>
<li>Users subconsciously believe that they are talking to human-like entities, and have matching expectations for their counterparts. These expectations cause frustration if not met.</li>
<li>Chatbots should not behave entirely human-like but should not be entirely robot-like, either.</li>
</ul>
All these obstacles pose quite a challenge for both designers and engineers: the use-case that allows to work around listed limitations is extremely narrow and navigating it to the sweet spot of user satisfaction here is going to be extremely hard, whereas the price of users’ frustration is going to be high. Nobody wants to design the next Clippy.
Some people argue that chatbots will replace 99% <a href="http://ift.tt/2nSSx9R">[10]</a>of the existing apps, probably meaning chatbots might replace their GUIs.
But the question should sound as — would the chatbots on a large scale actually offer any return on the investment at all?
I would say not at the moment. Maybe in the course of next 5 to 10 years, and by then everybody will forget the term “chatbots” and will come up with some other name for Hals 9000.
By the way…
If you already have built a chatbot, and it’s causing user frustration, try teaching it to apologise when it misunderstands something.
We, humans, tend to be very understanding and forgiving when someone apologises for their sub-optimal competence. <a href="http://ift.tt/2n6deLR">[11]</a>
<img alt="" src="http://ift.tt/2nSMeDz"><img src="http://ift.tt/2n6rCUr" width="1" height="1" alt="stat?event=post.clientViewed&amp;referrerSou">
<hr>
<a href="http://ift.tt/2nSUHq1">I’m sorry Dave, I’m afraid I can’t do that.</a> was originally published in <a href="http://ift.tt/2cAqCpx">Prototyping: From UX to Front End</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.

I’m sorry Dave, I’m afraid I can’t do that.

Why chatbots may not fly

Top Stories