In this work we report the highlights of the work done at the University of Bari within the FAIR project and concerning the ∗Corresponding author.
CEUR ceur-ws.org
The notion of symbiosis originated in the 19th century to indicate a relationship between two taxonomically separate life forms that nevertheless give rise to a single organism. Life forms in a symbiotic relationship are not isolated but coexist in ways that are more or less essential to their survival and development. The first to advocate a symbiosis between humans and machines was J.C.R Licklider in 1960 [1]. In his view, this kind of symbiosis would allow the computer to become an active part of the thinking process that leads to resolving technical problems and not just an executor of solutions thought up beforehand. Licklider was mainly thinking of human-computer interfaces that would allow greater real-time collaboration and shorten the distance between human and machine language. He was pointing to a road that has since been successfully travelled, bringing Human-AI symbiosis promises to boost human-machine collaboration and socio-technical teaming, with mutually beneficial relationships, by augmenting (and valuing) human cognitive abilities rather than replacing them [2]. In CEUR htp:/ceur-ws.org ISN1613-073
CEUR
Workshop Proceedings (CEUR-WS.org) tive partnership between humans and machines within a broader social and technological context, where the focus is not on a substantial peer-to-peer relationship but on integrating technology into human-centric processes and systems. In this context, symbiosis involves humans and machines working together as a cohesive unit, each playing a specific role and contributing to the team’s overall performance. On one hand, humans provide the cognitive and emotional capabilities necessary for creativity, empathy, ethical decision-making, and adaptability. On the other hand, machines ofer computational power, data processing, and automation capabilities that can handle repetitive and data-intensive tasks eficiently.
When applied to AI, the concept of symbiosis becomes more complex, posing a whole series of foundational questions. Addressing these questions is one of the goals of the research done by the University of Bari (together with INFN) within the project Future AI Research (FAIR). research for our investigation within a dedicated work package (WP 6.5) of FAIR. Acceptability involves value alignment between AI and humans. It is related, e.g., to understanding AI decisions, the algorithmic bias, the rethe struggle between security ensured by AI systems and fundamental freedoms, the mitigation of possible safety and health risks. In FAIR, studies on the acceptability of SAI adopt an interdisciplinary approach involving researchers in AI, Law, and Philosophy.
In this paper, we briefly report the main achievements of our research on ethical and legal acceptability of SAI in the 1st year of the project (Sections 2-3) and outline the steps needed to go from general principles to operational definitions for ethical acceptability (Section 4). Section 5 us to the so-called Symbiotic Artificial Intelligence (SAI). In particular, the acceptability of SAI is the subject of particular, socio-technical teaming refers to the collabora- spect of privacy policies for data collected by AI systems, © 2024 Copyright for this paper by its authors. Use permitted under Creative Commons License concludes the paper with final remarks.
Attribution 4.0 International (CC BY 4.0).
but are an integral part of the same evolutionary process and are responsible for it. We think that this approach is The philosophical approach to AI is contributing to the keeping with the provisions, ex multis, of memorandum debate on the identification and analysis of the ethical no. 38 of the Proposal for a EU Reg. on artificial intelliimplications of algorithms. We have continued the inves- gence. A procedural condition ensures the fairness and tigation aiming to build the proposal of a methodological transparency of decision-making and it allows recipients framework grounded in process-oriented evaluations to to understand and respect the decision itself. Indeed, in assess the human-centricity and acceptability of SAIs law, it is not suficient the content of the decision, but together with their societal benefit. also its enforcement. Thus, the efectiveness remains a
The research carried out concerned two diferent sci- constitutive element of legality [9]. entific lines: Furthermore, some legal issues raised by the interac
Questioning the notion of “symbiosis” in SAI sys- tion between humans and AI were addressed in some tems. The research focused mainly on the meaning of areas of law (such as those that most require judgments “symbiosis” and its applicability to AI [3]. To this end, of predictive type, like the assessment of dangerousness preliminary research has been carried out on the trans- aimed, for example, at commensurate punishment and/or formation of the concept of intelligence in the history granting alternative measures). It has been so possible of ideas [4]. In several internal meetings, the notion to observe and identify some essential conditions that of symbiosis was explored both from a biological and should be taken into account in designing AI systems in phenomenological point of view, with reference to the this field, necessary to promote the symbiosis between key recent AI-driven technological developments (AI and humans and AI as well as to improve the trustworthiness, drones, AI and robotics, LLM, ML, etc.). fairness and eficiency of the interaction (for example,
Assessing the ethical impact of SAI in terms of enriching the methods of responding to the crime in acceptability and human-centricity. Defining the fun- compliance with the fundamental principles of propordamental conceptual stages of a methodology for eval- tionality and dignity of the person, realizing the requests uating AI systems involves comparing and studying a for individualization of the punishment) [10]. series of international regulatory frameworks – inter alia Finally, we would like to mention that, the European AI HLEG, Ethics Guidelines for Trustworthy AI (2018- legal framework for AI gives minimal consideration to 19). We have outlined a model with diferent fundamen- regulating AI based technologies where there is a reciprotal steps: (a) onto-epistemic foundation of the method; cate relationship between human and machine (symbio(b) screening; (c) risk evaluation; (d) impact assessment. sis). The research field of symbiotic AI is technologically Now, we need to work within each step to refine proce- challenging. In [11], we have undertaken a foundational dures and metrics further. study with the aim of conceptualizing and designing a
The eforts in this direction have led to a joint paper comprehensive symbiotic approach to AI, with the goal of presented at the BEWARE workshop organized in Rome producing fair, legitimate, and efective outcomes while within the 22nd International Conference of the Italian ensuring their ethical and legal acceptability. This theoAssociation for Artificial Intelligence (AI*IA 2023) [ 5], an retical research is expected to influence the development article accepted for publication in the journal Intelligenza of Symbiotic AI systems and technological governance Artificiale [6], and diferent book chapters in the final through model assessment. stages of publication [7], [8].
In line with the ethical and philosophical considerations on symbiosis, moving from the perspective of humanmachine interaction to a procedural model of construc- The ethical implications of Human-AI symbiosis are multion and assessment of SAI decision, within a legal tifaceted and complex. Thus, it has become increasingly methodology theory we have identified the first legal paramount to take in consideration the ethical issues surpragmatic conditions of algorithmic decision-making, rounding SAI development, deployment, and impact. The such as that of the significant human control , a notion concept of ‘SAI Ethics’ ofers a nuanced perspective that borrowed from the international debate within the UN emphasizes the harmonious coexistence and collaboraon autonomous weapons. In this way, symbiosis trans- tion between humans and AI systems. Operationalizing lates also a techno-procedural legal principle capable of SAI Ethics involves translating abstract ethical princiformalizing a human-centric value where persons do not ples and values into concrete guidelines and practices remain behind technological development and society that govern every stage of the AI lifecycle, including data collection, algorithm design, model training, evalua- the integration of ethical principles into the design and tion, and deployment [12]. It requires a multidisciplinary development of AI algorithms and models. This means approach, involving collaboration between computer sci- translating ethical principles, values, and guidelines into entists, ethicists, policymakers, and other stakeholders to actionable and measurable practices or procedures. We ensure their alignment with societal values and human need to define specific rules, standards, or protocols that well-being, and to foster harmony and mutual benefit guide the behavior and decision-making in ethical dilembetween humans and machines. mas or concrete situations [16, 17]. Moreover, SAI Ethics emphasizes the importance of continuous learning and 4.1. Operationalizing SAI Ethics adaptation. As AI technologies evolve and their societal impact unfolds, ethical standards and norms must evolve From a practical perspective, operationalizing SAI Ethics in tandem [18, 19]. This requires interdisciplinary rerequires the establishment of governance frameworks, search, ethical reflection, and stakeholder engagement standards, and regulations to govern the responsible de- to address emerging challenges and dilemmas. velopment, deployment, and use of AI technologies. This includes the development of ethical guidelines, codes 4.2. Building a Computational Model of of conduct, and best practices to guide AI practitioners and organizations in navigating ethical dilemmas and SAI Ethics decision-making processes [13]. These tools should be Ethical Principles are abstract rules intended for guiding domain specific. Moreover, fostering interdisciplinary ethical decision making and judgement. There are a varicollaboration and stakeholder engagement is essential ety of techniques used for technical implementation of to ensure that ethical considerations are adequately ad- ethical principles. In the previous literature of machine dressed and that AI technologies serve the broader soci- ethics, ethical principles are integrated into machines etal interest. in a top-down, bottom-up, or hybrid architectures (see
One key aspect of operationalizing SAI Ethics is the [20] for a survey). However, so far, no model seems to development of robust frameworks and methodologies satisfy ethical judgement and decision making needs for for ethical risk assessment and mitigation. This involves an acceptable and responsible AI system. Approaches identifying potential ethical risks associated with AI sys- to encode principles into a format that computers can tems, such as bias, discrimination, privacy violations, and understand include logical reasoning, probabilistic reaunintended consequences, and implementing strategies soning, learning, optimisation, and case-based reasoning to address these risks proactively [14]. Thus, it is im- [21]. portant to design algorithms and systems that are trans- We argue that it is impossible to build a ’general ethical parent, interpretable, and accountable, enabling stake- AI’, i.e., a machine that is generally ethical, a machine that holders to understand how AI decisions are made and to can reason and take ethical decisions in any domain and detect and rectify ethical issues when they arise. Here in every context. We believe that we need to concentrate we would like to highlight the role of logic program- on building domain-based ethical machines, i.e., machines ming for designing such models [15]. Additionally, op- that are able of ethical reasoning and decision making in erationalizing SAI Ethics requires ongoing monitoring any context and situation in a specific domain, which is, and evaluation of AI systems in real-world contexts to any way, still a very challenging task. Considering the ensure that they continue to operate ethically and re- purpose and the specific domain for which the AI system sponsibly throughout their lifecycle. From a technical is developed, developers should consider codes of ethics perspective, operationalization should focus on human- and conduct of the domain (domain ethics, e.g. medical centricity through the development of AI systems that ethics) as a guiding framework. Furthermore, the key are transparent, interpretable, and accountable. This en- aspects of SAI, such as the collaborative and cooperative tails implementing mechanisms for explainability and nature between human and machine, the human-centric interpretability, allowing users to understand how AI approach, the mutual benefit, the adaptability and responalgorithms make decisions and providing insights into siveness of SAI, and the interdisciplinary perspective, their underlying processes. Techniques such as model in- should be taken in consideration in the design decisions terpretability, transparency tools, and algorithmic audits to be taken by the developers. enable stakeholders to scrutinize AI systems and iden- To build a computational model of domain ethics to tify potential biases, errors, or unintended consequences. be integrated into the AI system; the ethical principles of Additionally, ensuring the robustness and reliability of the domain should be operationalized. The operationalAI systems through rigorous testing, validation, and ver- ization task should be carried out involving all stakeification processes is essential to minimize the risk of holders and domain ethical experts. Developers should harmful outcomes and instil confidence in their use. also decide on the architecture to adopt for integrating
Furthermore, operationalizing SAI Ethics necessitates the ethical principles. Being clear about which principle is being used will help designers to further specify revise a previously learned rule and present it to the huwhat inputs are necessary for their application, which in man. Through a collaborative dialogue, The human can turn will improve the ethical reasoning capabilities and correct the ethical behavior of the machine, but also the explainability of how decisions have been made [22]. machine can automatically demonstrate to the humans
However, defining principles in an intentional manner their errors in reasoning. In this way both will learn so that they may be applied in a deductive manner, is and improve their reasoning capabilities (mutual benefit). often challenging and, in many cases, appears to be an This adaptability aspect will be tested and evaluated in impossible task. The issue lies in the gap between ab- our experiments. stract, open-textured principles and tangible, concrete facts. The abstract principles should be operationalized by linking them to the facts. When ethical experts jus- 5. Conclusions and Future Work tify their conclusions in particular cases, they frequently connect ethical principles directly to the specific facts of In this work, we reported on ongoing work in the Workthose cases. Essentially, these established connections Package 6.5 of the project FAIR. A model of ethical acbetween ethical principles and relevant facts serve as ceptability of SAI was outlined. Many legal issues raised operational (concrete) definitions of the principles. The by SAI systems were addressed. Currently, we are conexperts operationalize the abstract principles by tying centrating on SAI ethics operationalization. Next, we them directly to the factual context. will work on the operationalization of legal aspects in
We are going to investigate, computationally, the pos- SAI by the development of a framework for embedding sibility of operationalizing abstract ethical principles by the considerations of legal issues in SAI, then on realinducing practical rules for ethical judgement and deci- izing a computational model of legal reasoning for our sion making in SAI systems from real-life interactions be- SAI system to be ultimately integrated in the SAI system tween human and machine in diferent domains [ 19, 23]. together with the ethical model.
These rules evolve overtime through the interaction be- By operationalizing SAI Ethics and legal issues, we tween human and machine which is an important aspect can foster a collaborative and mutually beneficial relato SAI ethics. SAI recognizes the dynamic nature of tionship between humans and AI systems, promoting human-AI interactions and the need for AI systems to responsible and trustworthy AI development for the benadapt and respond to human preferences, values, and efit of the society. This requires a multifaceted approach feedback overtime. To achieve this, we are going to con- that integrates technical, organizational, regulatory, and sider diferent domains as case studies, collect and ana- societal perspectives. lyze a large set of domain ethics cases and build a com- A socio-technical approach to SAI systems developputational model employing diferent operationalization ment will be adopted which leads to an increased actechniques. Then, we are planning to carry out exper- ceptability of these systems [24]. To capture the socioiments to test our hypothesis that the computational technical complexity we are planning to adopt Multimodel will accurately classify actions as ethical or uneth- Agent Systems (MAS) for modelling the SAI system at ical. The model will be developed using a foundational hand [25]. The ethical and legal components in the sysset of cases that will be collected for this purpose. The tem will be implemented as a MAS, which will act as system performance will be evaluated using quantitative an ethical and legal over-layer in the overall decision measures like precision and recall. making process. A starting point might be the MAS pro
An important aspect, mentioned above, is the model totype presented in [26, 27] for the ethical evaluation and adaptability overtime. In the context of SAI systems, hu- monitoring of dialogue systems. man and machine (as agents) work as a team, collaborate Finally, since a human-centric approach is central and learn from each other, evolve together. The machine to SAI, transparency and explainability are key require(as well as the human) will learn concrete ethical rules ments for establishing trust in SAI systems which leads to from interaction with humans, the machine will apply acceptability. We would like to emphasize the the promithe previously learned ethical rules on concrete cases, nent role of computational logic in the development of will also revise and update the previously learned rules the computational model of ethical and legal acceptabilif needed. Here, it is important to emphasize the col- ity of SAI. Logic Programming (LP) has a great potential laborative aspect of SAI in revising and correcting the for developing such perspective ethical and legal SAI ethical behavior overtime by both the human and the systems, as in fact logic rules are easily comprehensible machine. In fact, this task is, in reality, a collaborative by humans. Furthermore, LP is able to model causality, task, the machine will extract the case facts (the facts of which is crucial for ethical and legal decision making the real-life case at hand), present them to the human, [15]. the human will provide an ethical judgment of the case at hand. Then the machine will learn a new rule and/or ing, Cham, 2022, pp. 1–19. doi:10.1007/978- 3- 3 19- 31739- 7_142- 1.
This work was partially supported by the project FAIR - [10] L. Pulito, Algoritmi predittivi e valutazione della Future AI Research (PE00000013), under the NRRP MUR pericolosità, L’Ircocervo (2024). Invited essay, subprogram funded by the NextGenerationEU. mitted. [11] P. Marra, L. Pulito, A. Carnevale, F. Lisi, A. LomReferences bardi, A. Dyoub, A procedural idea of decisionmaking in the context of symbiotic ai, in: Pro[1] J. C. R. Licklider, Man-computer symbiosis, IRE ceedings of the 1st International Workshop on DeTransactions on Human Factors in Electronics HFE- signing and Building Hybrid Human-AI Systems, 1 (1960) 4–11. doi:10.1109/THFE2.1960.4503259. co-located with 17th International Conference on [2] S. S. Grigsby, Artificial intelligence for advanced Advanced Visual Interfaces (AVI 2024), Arenzano human-machine symbiosis, in: D. Schmorrow, (Genoa), Italy, June 3rd, 2024., CEUR Workshop ProC. Fidopiastis (Eds.), Augmented Cognition: In- ceedings, 2024. URL: https://synergy.trx.li/ceur-w telligent Technologies, volume 10915 of Lecture s/paper9.pdf.
Notes in Computer Science, Springer, Cham, 2018. [12] J. Morley, L. Kinsey, A. Elhalal, F. Garcia, M. Ziosi, doi:10.1007/978- 3- 319- 91470- 1_22. L. Floridi, Operationalising AI ethics: barriers, en[3] A. Carnevale, Condizione e struttura del nos- ablers and next steps, AI Soc. 38 (2023) 411–423. tro rapporto con le macchine. dieci proposizioni URL: https://doi.org/10.1007/s00146-021-01308-8. per una filosofia critica dell’intelligenza artificiale doi:10.1007/S00146- 021- 01308- 8. antropocentrica, in: S. Barone, et al. (Eds.), L’uomo [13] J. Mökander, L. Floridi, Operationalising AI govanimale tecnologico, Sciascia Editore, Caltanissetta- ernance through ethics-based auditing: an indusRome, 2024. Invited chapter, accepted, in publica- try case study, AI Ethics 3 (2023) 451–468. URL: tion. https://doi.org/10.1007/s43681- 022- 00171-7. [4] A. Lombardi, L’origine dell’io. il “mistero” dell’intel- doi:10.1007/S43681- 022- 00171- 7. ligenza da Darwin al riduzionismo contemporaneo, [14] C. Novelli, F. Casolari, A. Rotolo, M. Taddeo, Studium/Ricerca 119 (2023) 651–688. L. Floridi, AI risk assessment: A scenario-based, [5] A. Carnevale, A. Lombardi, F. A. Lisi, Exploring eth- proportional methodology for the AI act, Digit. Soc. ical and conceptual foundations of human-centred 3 (2024) 13. URL: https://doi.org/10.1007/s44206-0 symbiosis with artificial intelligence, in: G. Boella, 24-00095-1. doi:10.1007/S44206- 024- 00095- 1. et al. (Eds.), Proceedings of the 2nd Workshop on [15] A. Dyoub, S. Costantini, F. A. Lisi, Logic proBias, Ethical AI, Explainability and the role of Logic gramming and machine ethics, in: Proceedings and Logic Programming co-located with the 22nd 36th International Conference on Logic ProgramInternational Conference of the Italian Association ming (Technical Communications), ICLP Technifor Artificial Intelligence (AI*IA 2023), volume 3615 cal Communications 2020, (Technical Communicaof CEUR Workshop Proceedings, 2023, pp. 30–43. tions) UNICAL, Rende (CS), Italy, 18-24th SeptemURL: https://ceur-ws.org/Vol-3615/paper3.pdf. ber 2020, volume 325 of EPTCS, 2020, pp. 6–17. [6] A. Carnevale, A. Lombardi, F. A. Lisi, A human- doi:10.4204/EPTCS.325.6.
centred approach to symbiotic AI: Questioning the [16] A. Dyoub, S. Costantini, F. A. Lisi, Learning answer ethical and conceptual foundation, Intelligenza set programming rules for ethical machines, in: Artificiale (2024). Invited paper, in publication. A. Casagrande, E. G. Omodeo (Eds.), Proceedings [7] A. Carnevale, Assessing the impacts of symbiotic of the 34th Italian Conference on Computational AI (SAI) on individual and societal well-being, in: Logic, Trieste, Italy, June 19-21, 2019, volume 2396 H. Webb, et al. (Eds.), AI Impact Assessment: meth- of CEUR Workshop Proceedings, CEUR-WS.org, 2019, ods and practices, Oxford University Press, 2024. pp. 300–315. URL: http://ceur-ws.org/Vol-2396/pa Invited chapter, accepted, in publication. per14.pdf. [8] C. Falchi Delgado, M. T. Ferretti, A. Carnevale, Be- [17] A. Dyoub, S. Costantini, F. A. Lisi, Towards an yond one-size-fits-all: Precision medicine and novel ILP application in machine ethics, in: Inductive technologies for sex and gender-inclusive covid-19 Logic Programming - 29th International Conferpandemic management, in: D. Cirillo, et al. (Eds.), ence, ILP 2019, Plovdiv, Bulgaria, September 3-5, Innovating Health against Future Pandemics, Else- 2019, Proceedings, volume 11770 of Lecture Notes vier, 2024. Invited chapter, accepted, in publication. in Computer Science, Springer, Netherlands, 2019, [9] P. Marra, I. Galatola, Efectiveness as Threat to Con- pp. 26–35. doi:10.1007/978- 3- 030- 49210- 6. stitutional Systems, Springer International Publish- [18] A. Dyoub, S. Costantini, I. Letteri, Care robots learning rules of ethical behavior under the supervision