How to apply the Uncanny Valley Theory to virtual agents or CallBots?

by | Jul 15, 2021

The Uncanny Valley Theory: methodology and subsequent interpretations.

The “uncanny valley” or “inexplicable” theory was developed in 1970 by robotics expert Masahiro Mori and basically argues that, when a bot shares a greater number of characteristics with a human, our empathy increases, triggering a positive reaction. However, there comes a time when, faced with this humanoid figure, this feeling changes drastically, triggering rejection – a disturbing valley – or an aversion to the degree of similarity of an entity that is not considered “human”. This rejection becomes positive again when we approach what could be considered a healthy person, as can be seen in the following graph:

To corroborate this controversial theory, in 2011 a team from the University of San Diego conducted an experiment based on a robot (R), a human-like android (A), and a human (H). This experiment consisted of analysing the effects on the brain when shown images of the three figures, using magnetic resonance imaging.

The answer was clear. While the robot (R) and the human (H) barely triggered any reaction, the android (A) produced a great alteration in several areas of the brain. Therefore, these scientists concluded that, faced with a human appearance but with robotic behaviour, we experience a different reaction that does not fit in our brain.

Gráfico valle inquietante

This particular reaction can be associated with a certain cognitive dissonance, since our brain has innate expectations about the appearance and characteristics of a human being, which this type of bots (by having a human appearance but with the “psyche” of an AI) these expectations are not fulfilled. On the other hand, religious motives and standards of how a human being should be can also greatly contribute in this circumstance.

Reacción del cerebro.

Application of the Uncanny Valley Theory to conversational AI or CallBots

According to the Theory of the Uncanny Valley, a limit of “humanization” will arrive for our AI that supposes rejection, which in the particular case of CallBots or Conversational AI could have its origin in the shock of having on the other side of the line an Artificial intelligence capable of handling low value-added requests just as a human would, but with a notably “robotic” voice.

To tackle this problem, the best option is to act directly on this cognitive dissonance, in order to modify one’s expectations about what “a bot should and should not do”, so that we will flatten the curve. Two alternatives are proposed:

  1. Increasing our knowledge of the operation of CallBots to avoid the reaction caused by the fear of an unknown entity (in this case without physical form) performing customer service functions that are often associated with a human.
  2. Providing our virtual agents with efficient voice technology to avoid the “gap” produced between the human-like attention but with a

In terms of the first point, a very interesting similarity to understand the aversion towards Artificial Intelligence whose operation we cannot understand can be found in aerophobia, or the fear of flying, as this irrational fear is often caused by the unknown, the aeroplane, its handling, the reliability of the technique, and the training of the pilots. However, when the individual affected by this phobia attends flight education classes and understands the operation of an aeroplane, the fear usually subsides or even disappears.

In the case of the androids from the previous experiment or our CallBots, a very similar situation can occur. We must understand what they are designed for and their main functions, and we will lose our fear of the unknown, shifting or even flattening the curve.

The function of CallBots is to make life easier for the agents carrying out repetitive and low-value tasks. For example, they are able to identify the customer’s needs, and refer them to the department or agents that best correspond to their request. They can also draw up a profile based on predefined setups, which the telephone agent will have before picking up the call. Having a prior source of information saves time and improves the communication with the customer. Seen like this, it doesn’t seem so disturbing, does it?

Agentes virtuales

Regarding the minimisation of the robotic voice, our virtual agents must have neural voices, whose resemblance to human voices makes them more pleasant in any conversation compared to other more outdated voice systems. In addition, to effectively solve each phase of the conversation correctly and quickly, and to create a perfect user experience, the integration of more than 5 AI providers is almost mandatory, taking advantage of the best qualities of each one at every stage of the CallBot’s conversation with the customer.

Finally, although natural language processing is improving considerably, we must bear in mind that CallBots cannot understand certain characteristics of human language such as sarcasm, double meaning, or complex moods, which can confuse the user. In this sense, we must be advised by specialists in the programming of these cognitive systems who can determine exactly what a CallBot can and cannot do, to avoid unnecessary deviations from their interaction with the client.

Conclusions and expectations of the evolution of the curve

As has been seen, the Uncanny Valley Theory can be used as a guide to improve the interaction mechanisms between humans and Artificial Intelligence, in this case CallBots, and not as an impediment, thus increasing the level of knowledge we have of its functionality and application, while eliminating the “android” effect caused by robotic voices.

That said, we must bear in mind that, since it was developed in 1970, we have experienced a remarkable evolution in the field of technology, and we now live in a world in which we interact more and more frequently with Artificial Intelligence, the tendency of which is to be present in all aspects of our life, from our homes to our moments of leisure. We must therefore consider that this curve might have already flattened and even tends to disappear.

At Vozitel we generate your custom CallBots for collection functions, appointment management, medical appointments, data protection, etc. We work with the main cognitive services in the market and, simultaneously, all thanks to a team of specialists who will advise you right from the very start so that you make the most of your virtual agents.  For more information.