Positive Psychology has been developed as a complement to traditional Psychology, in order to cater for positive components of personality which can lead to sustainable satisfaction, happiness and well-being. PositiveTechnologies have been developed to build technological tools to support Positive Psychology. Artificial Intelligence research has focused on the development of artefacts to relieve humans from undesired tasks, with lesser focus on artefacts to promote positive components towards well-being and happiness. In this article, the concept oPfositive Artificial Intelligence is proposed as a counterpart in Artificial Intelligence in general to the role Positive Psychology has played to Psychology.
multifaceted endeavour. Several initiatives target the development of methods and techniques to reconstruct intelligent behaviour in artefacts which could, ultimately, relieve humans from unwanted activities which can be repetitive, strenuous and/or potentially harmful.
This brief review of the work of these founders of important areas in computer science and technology has the purpose of highlighting two complementary approaches to design technologies that support human actions and interactionsp:osaitive approach, focusing on desired human capabilities to be enhanceind and for the benefit of humans , and anegative approach, focusing on undesired human tasks and traits torebteracted from humans, alsofor the benefit of humans . It resonates with – and furthers – arguments presented by Winogr2a]d, [ suggesting that the design of intelligent systems can be refined by accounting for both negative and positive approaches.
The appropriateness of adoption of methodologies that encompass negative as well as positive approaches to design has been considered in diferent domains. PInositive Psychology, similar arguments ground the proposition thanteagative approach – focusing on treatment of undesired psychological traits – should be complemented bpyosaitive approach – focusing on strengthening of desired potentialitiePso. sitive Technologies are an ofspring of Positive Psychology and encompass research about the development of artefacts to support the flourishing of desired potentialities.
In the present article the concept oPofsitive Artificial Intelligence is proposed as a complement to existing practices to design intelligent systems. The article also brings forward the proposition that regulatory bodies concerned with safety and ethical issues related to Artificial Intelligence should include strong requirements related to Positive Artificial Intelligence in regulations, in order to ensure that intelligent systems are not only harmless, but also useful.
Section2 contains a brief rendition oPofsitive Psychology. Section3 contains a discussion about Positive Technologies. Section4 contains a brief presentation of intelligent agents, highlighting how they could be enhanced by a positive approach to their design. Sec5tibornings forward the proposed concept oPfositive Artificial Intelligence , including illustrative scenarios to characterise our proposition. Finally, sectio6ncontains some conclusions, discussion and proposed future work.
Positive Psychology has been developed since the 90s to contrast with “conventional psychology”
– i.e. psychological methods based on psychoanalysis and treatment of hallmarks – by focusing
on well being, happiness and positivity3][. It has been characterised as a branch of Psychology
since the late 90s, although with roots in Humanistic Psycholo4g]y, L[ogotherapy 5[], Jungian
analysis and ancient Greek and Eastern philosoph6ie,s7,[
Very briefly, Positive Psychology is grounded on thPeERMA Theory proposed by Seligman and theFlow Theory proposed by CsikszentmihalyPi.ERMA is an acronym forP ositive emotions, Engagement, Relationships,Meaning andAchievement, which are identified as the five factors influencing the path towards a life of fulfilment, happiness and meaning.Flow is a state of awareness of inner and external events which leads to pleasure, satisfaction and efectiveness in goal seeking. Astate of Flow is reached and sustained via the dynamic balance of perceived challenge and skills, which must be informed to the individual through feedback about inner and external events and progress towards self determined goals.
Seligman and Csikszentmihalyi proposed a “Calculus of Well being” to clarify how “negative” and “positive” psychology should be balanced. Resorting to a simplistic analogy with Newtonian mechanics, retraction of undesired psychological traits can neutralise acceleration of a patient towards an unwanted direction in life, but actual movement in a new direction can only be achieved by adding energy to potentialities that can overcome inertia and lead to a desired direction (Figure2).
Seligman and Csikszentmihalyi, together with other scholars in Positive Psychology, have also clarified the diference betweenhedonic and eudaimonic states, in order to highlight the importance of the latter over the former: hedonic forces – which relate to momentary pleasure, contentment and satisfaction – are not self sustained and require permanent extrinsic reinforcement, in contrast with eudaimonic forces – which relate to sensations of prosperity, blessedness and happiness, and are perennial or, at least, longlasting. Resorting to the same analogy with Newtonian mechanics, hedonic forces can change direction of movement of oneself but cannot alter friction that works to stop that movement in the long run, whereas eudaimonic forces change direction of movemeanntd reduce friction towards zero.
A challenge to Positive Psychology has been the development of methods and techniques for efective interventions and assessment of results related to the strengthening of positive potentialities.Positive Technologies have been proposed as tools to enable controlled interventions and assessment methods.
Positive Technologies have been proposed to support Positive Psychology through artefact mediated, tangible and measurable actions and efects. In broad terms, Positive Technologies have addressed9[]: • Mental health: promotion of Positive Psychology components in – patients in vulnerable situations; – patients presenting symptoms such as depression, eating disorders and other observable behaviours; – patients requiring emotion regulation. • Neuro-rehabilitation: support to treatment of – patients with visuospatial partial disabilities; – patients in motor-cognitive neuro-rehabilitation; – patients with partial disabilities related to ageing. • Empathy and pro-social behaviour: support to – patients with partial disabilities in development of empathy; – patients with partial learning disabilities; – scenarios in which cross-cultural integration is required.
• Self-transcendence: promotion of experiences that generate a stataewoef.
Generally speaking, Positive Technologies must stand on three pillars:
1. Intervention: technology mediated construction of experiences that can induce states
that promote Positive Psychology components.
2. Monitoring: (possibly quantified) assessment of efectiveness of intervention, through
measurable surrogate markers such as
• fluctuations in hemodynamic parameters 1[0],
• occurrence and intensity of goosebump1s1[
An artefact whose design is based on these three pillars is capable of interventions which can be expected to help in the flourishing of Positive Psychology components, as validated by monitoring and empirical assessment.
If an artefact is comprised of technology to feature intelligent behaviour, it is called an Intelligent Agent. Intelligent Agents whose design is based on the three pillars of Positive Technology characterise Positive Artificial Intelligence. The following section contains a brief review of the concept of Intelligent Agents, as a final ingredient for the presentation of Positive Artificial Intelligence.
Following the general literature about Artificial Intellige1n8c, e19[], the field of Artificial Intelligence can be summarised as the scientific and technological development to build systems which can be organised according to two attributes: 1. Structural / Imitative intelligence: • Systems that areinherently intelligent by featuring structural organisation which aligns with explanatory theories of intelligence applicable to biologicalvaergseunsts • Systems that are capable oifmitating intelligent behaviour, regardless of their organisation; and 2. Human / Mathematically defined rationality: • Systems that adopt as reference for intelligence/intelligent behaviour human (or other biological) agenvtsersus • Systems that adopt as reference some theory of rationality that characterises “optimal” intelligence in terms of eficiency in goal seeking.
The combination of possibilities builds four alternatives: 1. Systems that are approximateliyntelligent as humans (or other biological entities); 2. Systems that provide good approximatiemitations of intelligent behaviour as observed in humans (or other biological entities); 3. Systems that are close to optimally aligned wmithathematical theories of rationality; and 4. Systems that can generate outputs that are good approximations of what is determined by mathematical theories of rationality.
From an engineering standpoint, the focus in this article is on the second and fourth alternatives. Instead of considering them as a dichotomy, however, the consideration of a spectrum of possibilities is suggested, in which these attributes can be combined in diferent ways.
Mathematical theories of rationality are based on optimisation: how to optimally reach a (possibly multi-attribute) goal, maximising reward and minimising use of resources. When dealing with complex systems, it is frequently required to deal with incomplete mathematical models, in which equations and rules are not completely known or may not be analytically solvable. In such cases, surrogate systems must be developed, which are capabilmeitoafting input-output behaviour of the complex systems being studied. Artificial Intelligence considering the fourth alternative above is about building such surrogate systems, which can be called intelligent rational agents.
Models of human behaviour are based on psychological theories and empirical data, which identify patterns in human behaviour and align these patterns with explanatory models. Artificial Intelligence considering the second alternative above is about building surrogate systems which are capable oimfitating the behaviour of human agents when facing similar scenarios and stimuli. Such systems can be calleindtelligent imitative agents.
In general, these attributes are independent, and an intelligent agent can be better or worse as either an intelligent rational agent or an intelligent imitative agent. In some rare scenarios and problems, these attributes can be conflicting (e.g. when humans whose behaviour must be imitated show pathological self-destructive tendencies). In such cases, a choice must be made about giving priority to one of the attributes.
In both cases, another dimension for analysis and design of intelligent agents can be considered, which is coined herdeegree of awareness of social interactions and relates to the extent to which a designed agent includes consideration about relations with other agents – which can be other intelligent agents or humans (or other biological entities). In order to make the presentation clear, the spectrum of possibilities of awareness of social interactions is reduced to four values: 1. Egocentric agents, which only account for their own goals and resources and consider any other entity and event in the environment as either resources to be exploited or barriers to be overcome. Such agents correspond to what was considered during the pioneering development of Artificial Intelligence. 2. Strategic agents, which are aware of the existence of other agents, which also have goals and resources of their own, but still give full priority to management of their own resources to reach their own goals. These agents assume that the other agents will behave similarly, and build strategies which, in order to optimise their own goals and use of resources, may be mutually beneficial to other agents. Such agents correspond to what is considered in Classical Economics and Mathematical Game Theo1r9y].[ 3. Social agents, which take into account collective goals and resources and act to optimise them, considering long term goals that may outlive the agents themselves. Such agents consider the benefit of the collectivity and of future generations as well as their own. 4. Empathic agents, which are capable of “wearing other agents’ shoes”, balance the importance of their own goals with respect to those of other agents and decide for actions based on social emotions2[0].
In this spectrum, each value adds to the previous one on refinement and, as a consequence, complexity of modelling of agent interactions: strategic agents are egocentric apgleunsts awareness of existence of other agents; social agents are strategic agpelunstsawareness of collective goals; and empathic agents are social agpelnutssawareness of (or “sensitivity” to) motivations based on social emotions.
Increased awareness of social interactions makes room for the design and development of intelligent agents capable of more efective human-agent interactions, which is fundamental to build intelligent agents as Positive Technologies – henPcoes,itive Artificial Intelligence.
Ultimately, intelligent agents are designed and developed to serve human needs, hence – directly or indirectly – intelligent agents exist to interact with humans.
In some cases the interactions are more evident, e.g. when agents interact with humans as artefacts designed for end users. In all cases, consideration of the widest possible spectrum of consequences of interactions with intelligent agents is advised, and has become object of attention of scholars and regulatory agencies.
The ACM and the IEEE have prepared general Codes of Ethics for professionals in
Computer Science and Engineering, catering specifically for autonomous and intelligent agents
in specialised sections2[
Adopting a terminology suggested by Peters et al2.3[], there has been an imbalance towards
“nonmaleficence” overbeneficence in the design of intelligent agents, i.e. codes and regulations
have focused on what should be avoided to prevent harmful interactions, instead of what should
be ensured to promote positive interactions (an important exception to this trend being the
work of Peters et al.1[
Priority to “nonmaleficence” seems to be prevalent for certification and quality assurance, with regulations focusing on risk mitigation to avoid undesired behaviour of systems (see e.g. the preliminary proposal developed by the FDA – U.S. Food and Drugs Administration – to regulateArtificial Intelligence and Machine Learning in Software as a Medical Device [24]). This seems necessary, although not suficient for interactive systems, and regulations must be complemented by requirements to ensure beneficence as well as “nonmaleficence”.
The fundamental proposition in this article is precisely that, in order for intelligent agents that interact directly with end users to be certified by regulatory bodies, these agents should be required not only to ensure that all measures were taken to mitigate risks and avoid undesired issues, but also that interactions were designed following strict guidelines to ensure that positive outcomes are likely to result to users.
As a common reference to characterise the practice of inclusion of attributes in intelligent agents to promote positive outcomes from interactions, the frameworkBolufe Zones is adopted [25, 26]. Blue Zones are communities around the world which share empirically observable characteristics in their citizens: longevity, high sense of satisfaction with respect to life as a whole, good health and well-being. Interestingly, these communities also share patterns in everyday habits, culture, family structure and social interactions, although disguised according to local culture and traditions. The first four Blue Zones that were studied are located in disparate locations such as Okinawa (Japan), Ogliastra (Sardinia, Italy), Loma Linda (Californa, US) and Nicoya (Costa Rica).
The common patterns in all Blue Zones produce observable hallmarks which can be assessed as markers of the “Blue Zone efect”. These hallmarks can be summarised as: • Having a physically active everyday life; • Having frequent, small and well-balanced meals; • Having a sense of community belonging such as those that you get by surrounding yourself with friends, families and neighbours; and • Finding a sense of purpose.
It is interesting to notice the correlation between these hallmarks andPEthReMA Theory of Positive Psychology: • A physically active life can bring a senseAocfhievement, promoteEngagement and fosterP ositive emotions; • Similarly, carefully managed meals can leverage on a senAsechoifevement, Engagement andP ositive emotions; • A sense of community belonging is directly relatedEtnogagement andRelationships; • Finding a sense of purpose directly relates to findinMgeaning, which directly relates to
P ositive emotions,Engagement, Relationships andAchievement.
In the following paragraphs this proposition is illustrated with concrete scenarios, taking into account designed actions of intelligent agents which are potentially capable of leveraging these hallmarks.
Two important trends can be observed globally: urbanisation and ageing (seehtet.pgs.:// ourworldindata.orgfo/r up to date statistical data). Therefore, it is important to develop technologies to improve the quality of experience of elderly citizens as pedestrians in urban environments.
Several projects have been developed to study the behaviour of pedestrians in specific contexts [27, 28], and several of these projects focus on the ageing populati2o9n, 3[0, 31]. Most projects focus on safety issues and how to improve safety through smart design and interactions with intelligent agents.
These are important issues, which nevertheless should be complemented with design practices
and features in intelligent agents to cater for positive compone3n2t,s3[
Personalisation, in this case, refers to psychological traits and values such as taste, preferences (which can be based on aesthetics, personal history, values, tradition etc.) and perceived capabilities, potentialities, scale of values and sense of community belonging. All these aspects are highly dynamic and directly related to the dynamics of interactions between individuals, individuals and society, as well as individuals and artefacts comprising intelligent agents. Hence, contrasting with intelligent agents that are built considering only safety – which can be developed based on diferent possibilities in the spectrum of awareness of social interactions – the design of agents to interact with humans accounting for positive componernetqsuires empathy, which encompasses all other possibilities in the spectrum.
Design practices such as the ones sketched in the previous paragraphs, together with design patterns that can enable them, can be clearly characterised. Our strong claim in this article is that organisations devoted to the development of rules and regulations for certification of intelligent systems should add to their already existing norms specific design rules catering for positive components, and explicitly verify and require alignment with these rules for certification.
Traditional medicine in China was centred aroundvtihlleage doctors, also referred to absarefoot doctors (see e.g. https://www.who.int/bulletin/volumes/86/12/08-021208/e n)./An interesting practice which is now withering away is to have village doctors remunerated by healthy villagers – citizens who fell sick would stop contributing to the remuneration of the doctor, and only start contributing again when they were again fit for work. This way, village doctors would focus on healthcare, instead of treatment, contrasting with Western practice.
In recent years, there has been growing interestPirnecision Medicine, which has proven to be relevant particularly for patients who may be sensitive to the side efects of medications and treatments, e.g. older adults with cancer, who may need treatment based on radiotherapy and chemotherapy, which can cause longstanding, highly debilitating side efects and be aggressive to any patient, particularly to elderly individu37a,l3s8[].
This specific scenario illustrates a situation in which conventional Western medicine (which is negative in the sense of Positive Psychology) is obviously important but, also, clearly not suficient. Novel technologies for Precision Medicinsehould include, as a requirement from regulatory agencies, functionalities to cater foprositive components of well-being for patients.
Two concrete examples which illustrate possibilities in this direction are: 1. An innovative medical device has been developed to treat specific types of cancer using electromagnetic fields to promote well-being and decelerate the growth of tumours, which are calibrated for individual patients based on a database of previous treatments and Machine Learning techniques that classify patients by similarity with previous patients with respect to specific physiological response to stimul3i9[]. Even though this initiative is still aligned with the perspective of Western medicine, the proposed measurement of success of treatment is based on Qauality of Life Index, instead of simplistic measurements such as growth ratio of tumours. 2. More directly to the point, studies have been developed to assess how technologies based on virtual reality, tele-presence and interaction with intelligent agents can be employed as Positive Technologies during treatment of patients with COVID-1490][.
Possible convergence of these two initiatives could lead to enriched devices and treatment protocols which can complement therapies with decreased harmful side efects with immediate, medium and long term interventions such as: • Incorporation of ludified and gamified experiences during treatment sessions, e.g. based on virtual reality that can promoPteositive emotions related to a sense oAfchievement and Engagement, as well as resources based on tele-presence that can proRmoelteationships and a sense of community belonging during treatments; • Incorporation of followup experiences to manage diet, physical activities and long term outpatient treatment, which can also be ludified and gamified and promotPeositive emotions, Engagement, Relationships, and a sense oAf chievement; • Incorporation of resources to promote participation in communities of support which can encourage proper management of treatment, reintroduction in social and professional activities – including support for self reinvention – and induction to transition from passive to active participation in such communities, therefore promoting a renovated sense of Meaning for recovered patients.
Functionalities to ensure positive components as the ones suggested in the previous paragraphs can be characterised based on design principles, which should be included by regulatory bodies as requirements for certification of novel devices and protocols for treatment of patients. This would bring Western medicine closer to a holistic and hence more efective approach to treatment, moving fromPrecision Medicine to Precision Health and Care [41, 42].
In this article the concept oPfositive Artificial Intelligence has been introduced, based on a particular view about the evolution of Artificial Intelligence and how the design of intelligent agents can be improved to include features that can promote Positive Psychology components.
This concept is illustrated with sketches of agent design whichfaearseible, viable anddesirable [41]. The article brings forward the proposition that regulatory bodies in charge of certification of systems that embed intelligent agents should include explicit requirements catering for Positive Psychology components for certification.
Planned future work shall follow at least two lines: 1. Experimental design of intelligent agents focusing on Positive Psychology components, to build evidence of feasibility of what is proposed here; and 2. Development of concrete propositions of requirements that could be incorporated in existing standards and norms for certification of systems.
This work has been developed with partial support from FAPEMSCPTIC/CGI – Future Internet for Smart Cities and the INCTInterSCity – Enabling the Future Internet for Smart Cities. [19] Y. Shoham, K. Leyton-Brown, Multiagent Systems: Algorithmic, Game-Theoretic, and
Logical Foundations, Cambridge University Press, 2009. [20] S. Hareli, B. Parkinson, What’s social about social emotions?, Journal for the Theory of
Social Behaviour 38 (2008) 131–156. [21] R. E. Anderson, Acm code of ethics and professional conduct, Communications of the
ACM 35 (1992) 94–99. [22] S. Spiekermann, Ieee p7000—the first global standard process for addressing ethical concerns in system design, Multidisciplinary Digital Publishing Institute Proceedings 1 (2017) 159. [23] D. Peters, K. Vold, D. Robinson, R. A. Calvo, Responsible ai—two frameworks for ethical design practice, IEEE Transactions on Technology and Society 1 (2020) 34–47. [24] U. Food, D. Administration, et al., Proposed regulatory framework for modifications to artificial intelligence, Machine Learning (AI/ML)-based Software as a Medical Device (SaMD). Available via https://www. fda. gov/media/122535/download. Accessed (2019). [25] D. Buettner, The secrets of long life, National Geographic 208 (2005) 2–27. [26] P. K. Hitchcott, M. C. Fastame, M. P. Penna, More to blue zones than long life: positive psychological characteristics, Health, Risk & Society 20 (2018) 163–181. [27] C. Feliciani, A. Gorrini, L. Crociani, G. Vizzari, K. Nishinari, S. Bandini, Calibration and validation of a simulation model for predicting pedestrian fatalities at unsignalized crosswalks by means of statistical trafic data, Journal of trafic and transportation engineering (English edition) 7 (2020) 1–18. [28] J. C. C. Silva, F. S. Correa da Silva, Contextual analysis of pedestrian mobility in transport terminals, submitted (2020). [29] A. Gorrini, G. Vizzari, S. Bandini, Age and group-driven pedestrian behaviour: from observations to simulations, Collective Dynamics 1 (2016) 1–16. [30] S. Bandini, L. Crociani, A. Gorrini, G. Vizzari, Crossing behaviour of the elderly: Road safety assessment through simulations., in: AI* AAL@ AI* IA, 2017, pp. 4–16. [31] J. Choi, R. Tay, S. Kim, S. Jeong, Behaviors of older pedestrians at crosswalks in south korea, Accident Analysis & Prevention 127 (2019) 231–235. [32] N. Avento, INDEPENDENT LIVING IN AGE-FRIENDLY CITIES: STUDY ON DYADS OF ELDERLY PEDESTRIANS WALKING DYNAMICS, Ph.D. thesis, Università degli Studi di Milano-Bicocca, 2015. [33] C. Musselwhite, Creating a convivial public realm for an ageing population. being a pedestrian and the built environment, Transport, Travel and Later Life (2017) 129. [34] L. V. Costa, A. Veloso, M. Loizou, S. Arnab, R. Tomlins, A. P. Sukumar, “what a mobilitylimited world”: Design requirements of an age-friendly playable city, Preprints (2018). [35] J. Fietkau, The case for including senior citizens in the playable city, in: Proceedings of the International Conference on Web Intelligence, 2017, pp. 1072–1075. [36] N. Distefano, G. Pulvirenti, S. Leonardi, Neighbourhood walkability: Elderly’s priorities,
Research in Transportation Business & Management (2020) 100547. [37] C. Fumagalli, C. Catania, A. Ranghiero, C. Bosi, G. Viale, F. de Marinis, M. Barberis, E. Guerini-Rocco, Molecular profile of advanced non-small cell lung cancers in octogenarians: The door to precision medicine in elderly patients, Journal of clinical medicine 8 (2019) 112. [38] G. Novelli, M. Biancolella, A. Latini, A. Spallone, P. Borgiani, M. Papaluca, Precision medicine in non-communicable diseases, High-throughput 9 (2020) 3. [39] E. Santana, F. Costa, Y. Setogute, L. Carvalho, P. Lima, M. Sousa, B. Gunz, F. Azevedo, J. Sabbaga, M. Santos, et al., 1604p exposure to low energy amplitude modulated radiofrequency electromagnetic fields (emf) is associated with rapid improvement in quality of life (qol) status in patients with advanced hepatocellular carcinoma (hcc), using various analyses of eortc-c30, Annals of Oncology 30 (2019) mdz261–006. [40] G. Riva, F. Mantovani, B. K. Wiederhold, Positive technology and covid-19, Cyberpsychology, Behavior, and Social Networking (2020). [41] A. Starkweather, C. S. Jacelon, S. Bakken, D. L. Barton, A. DeVito Dabbs, S. G. Dorsey, B. J. Guthrie, M. M. Heitkemper, K. T. Hickey, T. J. Kelechi, et al., The use of technology to support precision health in nursing science, Journal of Nursing Scholarship 51 (2019) 614–623. [42] T. J. VanderWeele, E. McNeely, H. K. Koh, Reimagining health—flourishing, JAMA 321 (2019) 1667–1668.