Chatbots have always been a partial representation of AI development in general. It’s common knowledge that chatbots will get “smarter” over the years. To put it another way, we’ve been expecting chatbots to more or less flawlessly pass the Turing test at some point in the not-so-distant future.
But, how do we determine the smartest chatbot? In this article, we will define the qualities that make a chatbot smart in the most humanly convincing way and showcase some of the most ideal services that represent this design.
This will inevitably also include economic and business design influences, of course.
Individual Traits of a Smart Chatbot
“By 2029, computers will have emotional intelligence and be convincing as people.”
– Ray Kurzweil, SXSW Interview in 2017
Even for casual users of chatbots, there are very specific attributes and action patterns that can be observed and evaluated in order to judge whether a certain conversational AI is smart enough. This has been the image presented by almost all the technological predictions about natural language processing so far. In general, it has been popularly divided into these categories:
Adaptability – many regular folks will immediately point out the essence of being situationally aware, at least algorithm-wise, when it comes to evaluating chatbot responses. So even if the responses are artificially coded (one by one), if the communication method is adequately varied, then the chatbot is considered “smart” enough.
Personalization – high natural language processing aptitude doesn’t always equate to well-crafted personalized messages. However, it helps to make the conversational AI’s actions feel more fluid. From simply asking straightforward questions to subtly including detail or two about the client when recommending stuff, a personal touch always adds to the overall “smartness” of any chatbot.
Understands Needs and Wants – on a related note to personalization, one very highly evaluated criteria in determining the smartest chatbot is its ability to analyze what the user actually needs. This is as opposed to what is apparently being asked at that time. Knowing the difference provides multiple classification-related benefits, though most notably for assessing high-probability leads on the designer side.
Attention to Detail – simply referred to as the strategic use of all user-related information when either deciding on an action or suggesting actions. For example, the smartest chatbot for this criteria might be capable of seamlessly combining transaction histories with regional profiles to create very specifically targeted recommendations.
Flowing Diction and Tone – in the case of voiced chatbots, sounding “smart” may be as straightforward as having a naturally sounding flow to words and phrases. While not exactly a product of coding or AI adaptation, it is still a considerable point to notice.
Interactivity is technically also considered to be something that makes a chatbot smart. However, we do understand that chatbots are nowhere near being sentient, so the term is never really used in its literal sense.
The Human Response for Smart Chatbots
“Chatbots are important because you won’t feel stupid asking important questions.”
– Petter Bae Brandtzaeg, Why People Use Chatbots (2017)
Second, to how the chatbots themselves react is how the human user treats the chatbot. Like how gamers imagine the virtual world of games in terms of codes, human users of chatbots typically tend to simplify the conversation when faced with an artificial partner. Only important keywords/keyphrases (commands) are used when it seems like the chatbot is nothing more than a glorified user interface.
Thus, the smartest chatbot should be capable of achieving these somewhat subtle prerequisites:
Should be mistaken for a human agent, at least ideally – from greeting up to the recommendations, the chatbot should feel like an operator that just happened to be on standby when the query is sent. This means responding with adequate word fluidity, either with a combination of pre-made speech quirks or adapted sentence patterns. This ability is also vitally important when the user would eventually be rerouted to an actual, real operator (which would then become a seamless, unnoticeable transfer).
Must encourage the user to interact with natural language – along the lines of treating the chatbot like a person, its design should also make the user think that they should talk naturally to interact with it. This is as opposed to the aforementioned codes and keywords, which are likely only used to trigger specific functions of a “text-glorified user interface”. This is even more important when the chatbot is voiced. The “conversation” would never seem like it’s a typical communicative exchange without integrating the chatbot’s NLP features efficiently.
Fallback options should be quick and seamless – this combines the key interactive aspects of the previous two. Instead of the chatbot using a few more extra steps, it should be capable of immediately switching over to a human representative at the most appropriate time. To make it seamless, it first requires the chatbot to be mistaken as a human. Then, if no direct fallback notification is used, the NLP should be applied “without skipping a beat”, until the human operator sends the next reply.
Simulate emotional callback to convincing effect – empathy can be a powerful tool in significantly improving the interactive experience provided by chatbots. This is a more complex method of integrating natural language processing features and AI, and as such, the feature would largely depend on the industry or type of business that the chatbot is being implemented. For example, chatbots designed for elderly services can project inquisitiveness by asking follow-up questions or showing curiosity in other written/spoken details that would encourage the user to be more open.
Disappears into the background when unneeded – lastly, if the service is no longer needed, a smart chatbot can automatically gauge the conversation flow and simply go back behind the curtain. Not as prevalent for service chatbots, but it can be quite convenient for (information) assistant chatbots used internally.
The Smartest Chatbots in Action
As for how exactly the smartest chatbots operate, we have a few examples, both past and current, that might be relevant to the ideas you are currently brainstorming (reference only):
- Google Duplex – using a combination of neural networks and cloud-based services, Duplex was the prototype AI system that was meant to “accomplish real-world tasks over the phone.” Basically, the assistant chatbot provides a platform of interaction for business agents, to which products and services are added automatically. Some of the concepts developed in this project were later incorporated into Google Assistant updates.
- Endurance Bots – dementia may not cause physical disability, but its impairment of mental functions can cause just as much damage. The developers of Endurance sought to mitigate such physiological restraints by communicating with the patients. Each conversation is recorded and then automatically analyzed for certain deviations that might be a symptom of specific memory loss issues. This also provides a temporary emotional relief, as the chatbot is also designed to engage in conversations empathetically.
- Casper (Insomnobot 3000) – having the urge to call or text someone suddenly when you can’t sleep? Well, Casper covers that for you without bothering your soundly sleeping buddies. In this instance, smart is used as a techno-slang term, though, rather than an actual metric of its aptitude; it is designed with a “sassy” attitude. The bot is built to initially amuse insomniacs trying it out for the very first time, as it can throw out some really snarky buddy responses when it interacts. At the moment, it has its limitations, but definitely worth checking out again in a few years.
- Roof.ai – this one is a real-estate business bot. Vehicles and homes are probably some the most tedious things that representatives are always expected to show and explain to customers. Hence a chatbot that not only provides but also gives good rough estimates on why the potential client would actually need them is instantly a candidate for the smartest chatbot. Of course, it’s not completely perfect, with a few generic “in-between” responses here and there. But you’d be convinced that it is almost perfect when you start talking to it.
Sadly, JARVIS is Still Out of Reach
With all the advancements that we have in AI, software development, and the ever-increasing number-crunching power of computers, one would think that something like Tony Stark’s JARVIS or FRIDAY is nearly within reach. Sadly, the level of independence shown by these two fictional “chatbots” is still impossible with current technology.
Perhaps Google or Amazon will unleash something similar within the next decade. But right now, all integration in chatbot design would have to take a few initial steps in connecting all-related analytics information. As the media often concludes, chatbots smarter than humans are yet to be poised to supplant us completely… at least in the foreseeable future.
So what is the smartest chatbot for the layperson? Anything that can dynamically adapt its interaction to your intended user base.
