Artificial Empathy - An Artificial Intelligence Challenge

Introduction

Artificial intelligence (Al) allows machines to analyse and solve problems utilizing heuristic, stochastic, fuzzy and other computational paradigms including biological principles (Xu et al., 2019). Al can learn from experiential data to automatically model and solve complex problems that may exceed the capacity of humans. Al-based deep learning algorithms have been found in certain instances to be superior to human clinicians in diagnosis, for example, pathologists in detecting the spread of breast cancer (Ehteshami et al., 2017). The current generation of Al systems is widely used in applications that enhance daily human life and those that further sophisticated research. Al and AE researchers could make significant strides in healthcare if Al can generate empathy at appropriate levels to optimize the delivery and effectiveness of healthcare services. Care provision robotics, for instance (as opposed to surgical robots), could ease the current care crisis due to longer life expectancy, the burden of multi-morbidity and shortage of skilled careprovision workforce. But a significant challenge for Al is the possession and deliverance of the human attribute of empathy. In this chapter, we strive to describe the trait of human empathy, analyse the possibility of implementing it using Al, explore the limitations in doing so and briefly discuss the perceivable wide-ranging repercussions of ARTIFICIAL EMPATHY (AE) specifically in the healthcare context while projecting future directions.

Empathy

While we acknowledge empathy as a variably defined term in different domains and contexts, we opine empathy is the quintessential^ subjective ability of an individual’s consciousness to experience the world of the other person, albeit briefly, and extend compassion. Notwithstanding the difficulty in defining this trait consistently, it is nevertheless, definitely discernible and deliverable by most humans. The deep analysis of this trait invariably leads to philosophy and phenomenology and has a myriad of other complexities -including cultural and religious ones. It is also not clear to us if this attribute is a learned one or is given at its full intensity at birth.

In theatrical and performance art, for example, empathy may mean an analogizing and/or a dialogical endeavour that makes an individual human subject feel the performer’s emotions/feelings and be able to reciprocate the same (Rousseau, 1754). In the philosopher Jean Jacques Rousseau’s words,

I perceive two principles antecedent to reason, of which one interests us intensely in our well-being and self-preservation, and the other inspires in us a natural repugnance at seeing any sentient being, particularly a fellow human, perish or suffer. It is from the collaboration and combination of the two principles of which our minds are capable...all the rules of natural law flow.

(Madumal et al., 2019)

The trait he eludes to as “antecedent to reason” is perhaps empathy or its primordial source. We take that as the best available description of the concept of empathy.

Enabling Empathy with Al

In our view, empathy is a superior manifestation of human intelligence and in that way could be accommodated by Al. Human intelligence is portrayed, in most circumstances, in an empathetic context. This being the case, it is easily deducible to the reader that such “context-sensitive” Al would be more human and, at least to a significant degree of plausibility, renders itself to be designed using the current Artificial Al tools and techniques.

Al systems could work like automatons, interacting with humans constantly or between themselves. Explainable artificial intelligence (XAI) models could incorporate dynamic interactions to obtain data, respond appropriately and learn to model their future behaviour automatically and transparently (Howarth, 2017). For example, in the healthcare domain, sig-nals/data are constantly generated by patients, and a reasoning system for modelling can draw data for insight analysis. Such systems can provide the fundamental framework to implement AE given the availability of a large amount of electronic medical data from real-time monitoring. XAI models could be designed to discern cognitive, emotional and compassionate parameters (pain, gestures, verbal/non-verbal cues, historical data, specific diagnoses, etc.), in addition to vital physiological data and laboratory data of patients to determine the empathy needed for in a particular care provision situation and deliver it appropriately with verbal and non-verbal methods.

A pioneering study done in Australia by one of the authors of this chapter and a team of researchers, where the concept of acceptability of humanoids in direct clinical care was explored, had garnered significant interest (ANZCTR, 2007; Schwartz, 2019). In this particular study, a commercially available programmable humanoid robot was trained to perform a limited clinical interaction with healthcare professionals in an acute hospital setting. The emphasis being on the nature of clinical communication, the study explored the feasibility of delivering empathy, a clinically pertinent quality, to see if the health professionals (acting as patients) could indeed place trust in such a machine in the position of a healthcare professional (Howarth, 2017; Madumal et al., 2019). While the results of the study are yet to be made publicly available, studies such as this attest to the feasibility of exploring this area further.