Il case of Tengai, which we also extensively discussed on FIRSTonline, prompts us to make some general considerations on our relationship with tools equipped with artificial intelligence. Tengai can indeed be a paradigm case of a social use of artificial intelligence.
It does not take away any jobs, it does not impoverish any family, it is not an instrument of social control, it does not work for the Chinese government. In short, it has none of those components that arouse so much concern in an increasingly broader public opinion, as regards anguish about the future.
Tengai is undoubtedly an artificial intelligence service applied to the social-relational context. It helps to better perform an activity that can be conditioned by arbitrary actions and suffer from discriminatory attitudes. That of personnel selection. In evaluating personnel, Tengai tends to be objective, assuming the merit, competence, motivation and ability of the candidate to perform the task for which he is at stake as the only parameters of his work.
It is not entirely wrong to say that technology can be born neutral from the work of scientists, inventors and research teams, but when it meets the animal spirit of the market this neutrality tends to separate like a great river in a swampy delta. In the case of Tengai this evolution, typical of that type of ecosystem, for the moment seems far from happening. We are still in the good.
From science fiction to reality
Until recently, an automaton conducting a job interview was a subject reserved for science fiction. Today, with Tengai, it has become a reality. We have already discussed extensively about Furhat's robot, nevertheless the time has come to ask ourselves concretely what Tengai is called to do and what his role will be within the selection process.
(All quotations that follow, unless otherwise indicated, are taken from the site www.tengai-unbiased.com)
TNG promotes its recruitment solution thus:
I held Select offers clients quality candidate selection through robot blind interviews, data-driven interview analysis [date-driven] and the transcripts of the interviews. I held Select will also help enhance the candidate experience, [representing] a more credible brand for employers and is time and cost effective. The interview robot is integrated as part of the initial interview phase and conducts blind interviews, which means that all candidates get the same skills-based questions and treatment. It will also allow more candidates to be interviewed and all interviews will be transcribed.
The interaction
Tengai's first goal is to establish some empathy with the candidate.
Furhat's automaton will therefore be used in the initial phase of the interviews, that of screening based on skills. Candidates will likely have already passed a preselection, especially if the numbers at stake are important. The interviews in which Tengai will be involved, therefore, aim to identify the skills of the candidates for a specific job position. A task in some ways routine, but not trivial.
Gabriel Skantze, professor at KTH Stockholm and Chief Scientist of the Tengai project at Furhat Robotics explains:
I held will assist recruiters and conduct interviews with candidates to assess their skills and competencies, based on job requirements and profile.
He goes on to underline how the robot's work will ultimately not differ much from that of a human being: "The robot will conduct the interview in a very similar way to a human recruiter, using questions based on competence".
And this is where Tengai's peculiarity lies, namely in the robot's ability to evoke, arouse and eventually establish an empathic relationship with the person in front of it: «The robot will return feedback (nodding, smiling and saying "mhm") , in order to encourage the candidate to provide elaborate answers,” continues Skantze. Tengai will dialogue with the candidate and, if necessary, will also produce a more in-depth discussion: «If the answer is too vague, the robot could ask the candidate, for example, to provide more concrete examples».
Finally, concludes the Chief Scientist of Furhat:
After the interview, the robot will present a summary of the interview and some objective recommendations (objective) so that a human decides on the candidate.
Unpredictability and indeterminacy
Tengai is very different from voice assistants like Siri in that Tengai tries to process even unexpected situations.
Precisely the colloquial nature of the task entrusted to Tengai and the marked interaction to which he is called, expose him to having to deal with unpredictable situations and events. This will happen to a much greater extent than with other devices with which we currently communicate (think, for example, of the personal assistants of the various digital devices or even the GPS navigator), where the conversation remains mostly confined within a well-circumscribed perimeter.
Gabriel Skantze explains the difference:
In the case of Siri, the human user asks a question and Siri searches for relevant information and presents it in a fun and social way. To the At the moment, however, Siri appears easier to program than I held. Our task is considerably more complex. In our case, we need to transfer an acquired human behavior to a robot that has exactly zero previous experience of human social interaction.
The latter remarks how conducting a job interview is far more demanding and inevitably gives rise to indefinable scenarios a priori. To address and manage this uncertainty, the predictive nature of the latest generation of artificial intelligence is used and we actually rely on machine learning:
For this purpose, I held will be developed using machine learning, which means that we will collect data from human-conducted interviews, where the potential for bias is as low as possible, and then train the system to replicate the conversational behavior of the human recruiter.
The ambiguity
Language can be very ambiguous to a non-biological intelligence.
Tengai will have to deal with indeterminacy on a daily basis, as well as constantly struggle with ambiguity. Indeed, one of the most demanding challenges for the automaton to overcome is constituted precisely by ambiguity. The one concerning language and its innumerable facets and the one inherent in the understanding of the multiple codes we use to communicate.
Sinisa Strbac, project manager and robot trainer of the Tengai project knows it well:
We have put a lot of effort into analyzing the questions that the robot should ask. It has turned out to be very important that the questions are as clear and concrete as possible. I held struggles with ambiguity, so the less vague a candidate's response, the better I held in assessing and validating the mix of competencies.
The robot, like human recruiters, will follow a consolidated pattern of questions, within a verified competence framework (verified competencies framework), to be submitted to the candidates. Nonetheless, the AI will have to be able to fully understand the answers of the latter, in order to decide the direction in which to continue the interview.
It still detects Strbac
A candidate's response can describe any imaginable situation, be referred to in a myriad of different ways and in an individual choice of words. The robot must understand and analyze this information and then determine which supplemental question is more relevant.
In order to be able to conduct the interview in full autonomy, Tengai has been equipped with the very latest acquisitions in terms of AI, the state of the art in the field of speech recognition, natural language understanding and the meaning underlying the language. Thanks to this, the robot is able to collect a considerable amount of information and process it profitably, conversing with its interlocutor:
In recruitment, particular attention is paid to providing and collecting information. And this also applies to I held. It shares information in the form of a dialogue about the [job] interview, how it will be conducted and finally collects the answers. As it captures information, it understands what the candidate is saying, regardless of the number of words and phrases used. So yes, it can handle open-ended questions and indeterminate answers.
Nonetheless, when it comes to language and human interaction, the unknowns turn out to be innumerable. Any errors or misunderstandings could even affect the much sought-after objectivity of the machine. Gabriel Skantze is aware of this:
There may be a risk that speech recognition (which translates speech into words) may malfunction for speakers with a foreign accent or for a particular gender. This could potentially affect the outcome of the interview.
data and training
Tengai was designed to acquire data with a method called "The Wizard of Oz". This method places control of the process in one person's hands.
In order to make Tengai efficient, before and even more than training, the data used in the machine training process are fundamental.
Gabriel Skantze explains how the data was first acquired. It is a methodology that goes under the name of “Wizard of Oz”, which leaves control of the robot to a human, although those who interact with the machine are unaware of it.
I held it was initially monitored by a human recruiter sitting in another room, who triggered the robot's behavior, which allowed us to record human-robot interview data. This data will now be used to develop two different models: one that can replicate the Magician's behavior during the interview and one that can perform the evaluation of the interviews afterwards.
The diversity, underlines Skantze, also involved those who temporarily took control of Tengai, in order to avoid preconceptions of any kind, even if they were unconscious: «[Tengai] is learning from different recruiters so as not to acquire the behavior specific to an individual recruiter'.
The diversity, this time of the robot compared to the human counterpart, is also expressed in relation to the meticulous control that can be exercised over the behavior of the automaton: «Unlike a human recruiter, we can control the robot's behavior in detail, down to the micro movements of the corner of the mouth,” notes Skantze.
The data acquisition and machine training phase took months and counted beyond duecent interviews with volunteers offered as guinea pigs.
All mimics, sounds and facial expressions are designed by studying and filming several hundred interviews of different recruiters, thus creating an interaction where job seekers feel safe, at ease, for a smoother candidate experience. straightforward and avant-garde.
In fact, let's not forget that the peculiarity of Tengai lies in the automaton's ability to fully interact with its interlocutor.
Hospitality
As Minority Report slightly anticipated something that from science fiction would become reality, so Her/She, Spike Jonze's film, starring Joaquin Phoenix, anticipates another, not distant advent. Tengai may already be something he announces Her.
Sitting in front of a robot and having a conversation with it is (still) an experience in itself out of the ordinary. It is even more so if the stake is important, such as the search for a job. A condition which, even before and regardless of the presence of the car, is a source of stress and tension.
Detect Strbac:
We're asking our candidates to place themselves in a particularly unusual situation: sitting in a room with only a robot and having a conversation. Being asked a question, answering it to the best of your ability, and then moving on.
The reactions are obviously the most diverse and vary from individual to individual. Observe Tengai's robot trainer.
Some people are skeptical at first, but are surprised by how natural it is I held handles the conversation. Others maintain their skepticism even after the interview as they have high expectations regarding the robot's ability to mimic human attitudes and speak naturally, which I held he is not yet able to do it perfectly, even if he is on his way. He's social, but not in the same way as a human being.
Then there are others who are in favor of the concept, but who get nervous before the interview. After all, they are being asked to 'perform' in front of a machine that tracks their movements and records their responses.
Strbac continues
There's nothing strange about that, really; humans do the same thing. But their [the candidates'] feeling is that the robot is much more than a human. It literally “records my responses!” It's the comment they make."
Acceptance
As in the legendary Mechanical Turk, the interview is a sort of chess game with the machine.
Indeed, there is some difference. It resides, perhaps, both in human psychology and in the profound and multifaceted ability of analysis that AI has now achieved. The automaton is perceived, rightly or wrongly, as a much more difficult subject to deceive or, if one wishes, to be able to persuade and "govern", compared to a human being. When faced with the machine, many certainties are missing.
Last but not least is the trust that some individuals place in their own ability to be right about the interlocutor they are faced with, or at least in the possibility of being able to ensnare him. The machine is emotionless and the emotional component is a powerful lever to use in interpersonal relationships. With the robot the game is played on a completely different level.
Furthermore, artificial intelligence now possesses analytical qualities which, taken individually, far exceed the capabilities of a human being. Furthermore, these apply to so many aspects and with such detail that a single individual would hardly be able to embrace and evaluate them in the field as a whole.
Not to mention, then, the recognized efficiency of the machines, where everything is meticulously and inexorably, in fact, recorded ... These prerogatives are frightening: «some people can't help but feel a bit as if they were under surveillance and therefore find themselves excessively uncomfortable".
Although the feedback that TNG says received during the development of Tengai appears overall more than positive, it must be said that it is one thing to be a guinea pig in an experiment, perhaps even a singular one, it is quite another to stake one's future. Playing the role of the candidate and carrying the burden could turn out to be something else entirely: "At the end of the day, they're sitting there having a complex conversation with a computer that will ultimately have real-life implications."
Strbac is optimistic and trusts Tengai's qualities:
On the other hand, most people are surprised by the naturalness of the discussion; the longer the interview lasts, the less they think about the fact that they are talking to a robot. It becomes just another component of the interview.
uncanny Valley
According to the Japanese scientist Masahiro Mori, trust in robots grows as they take on more and more human features.
Ultimately, the automaton must be accepted by one's interlocutor. And not only by some, but by all, whatever the psychology, sensitivity and ideas of the subject in question. Joe Mendelson, Head of interaction at Furhat, gets straight to the point: «People will accept robots in these new types of roles in society (interviewer in a job interview), which go beyond entertainment or even information , within the somewhat serious territory of job hunting?”
Then he continues:
This will be quite a leap for some people — so I hope to learn to understand how different types of people feel about this type of interaction.
Reactions to Furhat's robot have not been unanimous. Some time ago Joel Cheesman, American podcaster, defined it "creepy” (creepy, creepy). The latter, in fact, has then met Tengai at Tatech Europe 2019 and the perception has partially changed, moving slightly towards the "cool" (cool). According to what was declared on that occasion, after a few hundred interviews in the test phase, 3/4 of the candidates say they are in favor of meeting Tengai again. A decidedly high percentage!
When you interact with a robot, and in particular with an (anthropomorphic) robot that imitates human attitudes and behaviors, the question becomes considerably more complex. At the moment, scientists have not yet found the magic formula that allows automatons not to be perceived as foreign bodies by their flesh and blood counterparts. It is a phenomenon, still in part controversial, which leads the machine to generate a feeling of annoyance and unease in man, which can then lead to a firm refusal.
In such cases we speak of uncanny valley. It is a hypothesis formulated by the Japanese scientist Masahiro Mori in 1970, still not universally recognized. According to Mori, the acceptance and feeling of comfort felt towards anthropomorphic devices and automatons grows all the more the more they are similar to humans. Up to a certain threshold though. Beyond this threshold, excessive realism produces unpleasant sensations, causes discomfort and can give rise to rejection.
The mistake of I held
As an acceptance, Tengai tends to fall within the level of IBM Project Debater, the AI system capable of holding a debate on an impressive number of topics
This is probably what happened in Edinburgh with the robot Fabio, hired by the (very Italian) supermarket chain Margiotta Food & Wine. Fabio represented an experiment promoted, in collaboration with the Heriot-Watt University of Edinburgh, from the series BBC's Six Robots & US.
"Unfortunately, Fabio didn't work as we hoped," Luisa Margiotta clarified. «The performance was not what was expected, people avoided it, it scared the customers». The automaton was therefore fired.
Similar sensations seems to have raised in his debut also English AI Anchor, the hologram resulting from the collaboration between Xinhua, China's state news agency e Sogou. To most it seemed "robotic", but there are also those who called it "a little horrible", while someone considered "really scary".
And perhaps the discomfort experienced by is also to be traced back to the same phenomenon Dieter Bohn towards theIBM Project Debater, on the occasion of the presentation of the latest IBM marvel, held in San Francisco in the summer of 2018:
Frankly it made me feel a little uneasy, but not because of the usual worries like “robots will become self-aware and take over” or “AI is coming [to steal] our jobs”. It was something more subtle and difficult to grasp. Maybe for the first time, I felt like an AI was trying to disguise. I have not seen her lie, nor do I think she attempted to deceive us, but she did engage in a debating tactic that, if you had seen her attempted by a human, you would have placed a little less faith in that human being . It was a somewhat dissonant moment from a cognitive point of view. As mentioned, creepy.
Acceptance will be just one of the many tests that Furhat's robot will be called to overcome. The fact that it represents an innovative experiment in the field of artificial intelligence, as well as the particular vocation of the automaton for interaction, indeed pose multiple challenges with which Tengai and the members of the project will have to measure themselves.
Some of the most significant comparisons are called algorithmic bias; black box and explainable AI; accountability and audit as well as, last but not least, that the much praised and sought-after fairness really proves to be fair and benefits everyone, rather than a few. These will be the topics we will address in the near future.
