Tag: Artificial Intelligence

Posted on

(Featured) On Artificial Intelligence and Manipulation

On Artificial Intelligence and Manipulation

On the ethics of emerging technologies, Marcello Ienca critically examines the role of digital technologies, particularly artificial intelligence, in facilitating manipulation. This research involves a comprehensive analysis of the nature of manipulation, its manifestation in the digital realm, impacts on human agency, and the ethical ramifications thereof. The findings illuminate the nuanced interplay between technology, manipulation, and ethics, situating the discussion about technology within the broader philosophical discourse.

Ienca distinguishes between concepts of persuasion and manipulation, underscoring the role of rational defenses in bypassing effective manipulation. Furthermore, they unpack how artificial intelligence and other digital technologies contribute to manipulation, with a detailed exploration of tactics such as personalization, emotional appeal, social influence, repetition, trustworthiness, user awareness, and time constraints. Finally, they propose a set of mitigation strategies, including regulatory, technical, and ethical approaches, that aim to protect users from manipulation.

The Nature of Manipulation

Within the discourse on digital ethics, the issue of manipulation has garnered notable attention. Ienca begins with an account of manipulation, revealing its layered complexity. They distinguish manipulation from persuasion, contending that while both aim to alter behavior or attitudes, manipulation uniquely bypasses the rational defenses of the subject. They posit that manipulation’s unethical nature emerges from this bypassing, as it subverts the individual’s autonomy. While persuasion is predicated on providing reasons, manipulation strategically leverages non-rational influence to shape behavior or attitudes. The author, thus, highlights the ethical chasm between these two forms of influence.

Building on this, the author contends that manipulation becomes especially potent in digital environments, given the technological means at disposal. Digital technologies, such as AI, facilitate an unprecedented capacity to bypass rational defenses by harnessing a broad repertoire of tactics, including personalized messaging, emotional appeal, and repetition. These tactics, which exploit the cognitive vulnerabilities of individuals, are coupled with the broad reach and immediate feedback afforded by digital platforms, magnifying the scope and impact of manipulation. As such, Ienca’s research contributes to a deeper understanding of the nature of digital manipulation and its divergence from the concept of persuasion.

Digital Technologies and the Unraveling of Manipulation

Ienca critically engages with the symbiotic relationship between digital technologies and manipulation. They elucidate that contemporary platforms, such as social media and search engines, employ personalized algorithms to curate user experiences. While such personalization is often marketed as enhancing user satisfaction, the author contends it serves as a conduit for manipulation. These algorithms invisibly mould user preferences and beliefs, thereby posing a potent threat to personal autonomy. The authors extend this analysis to AI technologies as well. A key dimension of their argument is the delineation of “black-box” AI systems, which make decisions inexplicably, leaving users susceptible to undisclosed manipulative tactics. The inability to scrutinize the processes underpinning these decisions amplifies their potential to manipulate users. The author’s analysis thus illuminates the subversive role digital technologies play in exacerbating the risk of manipulation, informing a nuanced understanding of the ethical complexities inherent to digital environments.

Ienca posits that such manipulation essentially thrives on two key elements – informational asymmetry and cognitive bias exploitation. Informational asymmetry is established when the algorithms controlling digital environments wield extensive knowledge about the user, engendering a power imbalance. This understanding is used to shape user experience subtly, enhancing the susceptibility to manipulation. The exploitation of cognitive biases further solidifies this manipulation by capitalizing on inherent human tendencies, thus subtly directing user choices. An example provided is the use of default settings, which exploit the status quo bias and contribute to passive consent, a potent form of manipulation. The author’s exploration of these elements illustrates the insidious mechanisms by which digital manipulation functions, enriching our understanding of the dynamics at play within digital landscapes.

Mitigation Strategies for Digital Manipulation and the Broader Philosophical Discourse

Ienca proposes a multi-pronged strategy to curb the pervasiveness of digital manipulation, relying significantly on user education and digital literacy, contending that informed users can better identify and resist manipulation attempts. Transparency, particularly around the use of algorithms and data processing practices, is also stressed, facilitating users’ understanding of their data’s utilization. From a regulatory standpoint, the authors discuss the role of governing bodies in enforcing laws that protect user privacy and promote transparency and accountability. The EU AI Act (2021) is highlighted as a significant stride in this direction. The authors also advocate for ethical design, suggesting that prioritizing user cognitive liberty, privacy, transparency, and control in digital technology can reduce manipulation potential. They also highlight the potential of policy proposals aimed at enshrining a neuroright to cognitive liberty and mental integrity. In their collective approach, Ienca and Vayena synthesize technical, regulatory, and ethical strategies, underscoring the necessity of cooperation among multiple stakeholders to cultivate a safer digital environment.

This study on digital manipulation connects to a broader philosophical discourse surrounding the ethics of technology and information dissemination, particularly in the age of proliferating artificial intelligence. It is situated at the intersection of moral philosophy, moral psychology, and the philosophy of technology, inquiring into the agency and autonomy of users within digital spaces and the ethical responsibility of technology designers. The discussion on ‘neurorights’ brings to the fore the philosophical debate on personal freedom and cognitive liberty, reinforcing the question of how these rights ought to be defined and protected in a digitized world. The author’s consideration of manipulation, not as an anomaly, but as an inherent characteristic of pre-designed digital environments challenges traditional understanding of free will and consent in these spaces. This work contributes to the broader discourse on the power dynamics between technology users and creators, a topic of increasing relevance as AI and digital technologies become ubiquitous.

Abstract

The increasing diffusion of novel digital and online sociotechnical systems for arational behavioral influence based on Artificial Intelligence (AI), such as social media, microtargeting advertising, and personalized search algorithms, has brought about new ways of engaging with users, collecting their data and potentially influencing their behavior. However, these technologies and techniques have also raised concerns about the potential for manipulation, as they offer unprecedented capabilities for targeting and influencing individuals on a large scale and in a more subtle, automated and pervasive manner than ever before. This paper, provides a narrative review of the existing literature on manipulation, with a particular focus on the role of AI and associated digital technologies. Furthermore, it outlines an account of manipulation based of four key requirements: intentionality, asymmetry of outcome, non-transparency and violation of autonomy. I argue that while manipulation is not a new phenomenon, the pervasiveness, automaticity, and opacity of certain digital technologies may raise a new type of manipulation, called “digital manipulation”. I call “digital manipulation” any influence exerted through the use of digital technology that is intentionally designed to bypass reason and to produce an asymmetry of outcome between the data processor (or a third party that benefits thereof) and the data subject. Drawing on insights from psychology, sociology, and computer science, I identify key factors that can make manipulation more or less effective, and highlight the potential risks and benefits of these technologies for individuals and society. I conclude that manipulation through AI and associated digital technologies is not qualitatively different from manipulation through human–human interaction in the physical world. However, some functional characteristics make it potentially more likely of evading the subject’s cognitive defenses. This could increase the probability and severity of manipulation. Furthermore, it could violate some fundamental principles of freedom or entitlement related to a person’s brain and mind domain, hence called neurorights. To this end, an account of digital manipulation as a violation of the neuroright to cognitive liberty is presented.

On Artificial Intelligence and Manipulation
Posted on

(Featured) Examining the Differential Risk from High-level Artificial Intelligence and the Question of Control

Examining the Differential Risk from High-level Artificial Intelligence and the Question of Control

Using scenario forecasting, Kyle A. Kilian, Christopher J. Ventura, and Mark M.Bailey propose a diverse range of future trajectories for Artificial Intelligence (AI) development. Rooted in futures studies, a multidisciplinary field that seeks to understand the uncertainties and complexities of the future, they methodically delineate a quartet of scenarios — namely, Balancing Act, Accelerating Change, Shadow Intelligent Networks, and Emergence — and contribute not only to our understanding of the prospective courses of AI technology, but also underline its broader social and philosophical implications.

The crux of the authors scenario development process resides in an interdisciplinary and philosophically informed approach, scrutinizing both the plausibility and the consequences of each potential future. This approach positions AI as more than a purely technological phenomenon; it recognizes AI as an influential force capable of reshaping the fundamental structures of human experience and society. Thus, study sets the stage for an extensive analysis of the philosophical implications of these AI futures, catalyzing dialogues at the intersection of AI, philosophy, ethics, and futures studies.

Scenario Development

The authors advance the philosophy of futures studies by conceptualizing and detailing four distinct scenarios for AI development. These forecasts are constructions predicated on an extensive array of plausible scientific, sociological, and ethical variables. Each scenario encapsulates a unique balance of these variables, and thus, portrays an alternative trajectory for AI’s evolution and its impact on society. The four scenarios—Balancing Act, Accelerating Change, Shadow Intelligent Networks, and Emergence—offer a vivid spectrum of potential AI futures, and by extension, futures for humanity itself.

In “Balancing Act”, AI progresses within established societal structures and ethical frameworks, presenting a future where regulation and development maintain an equilibrium. The “Accelerating Change” scenario envisages an exponential increase in AI capabilities, radically transforming societal norms and structures. “Shadow Intelligent Networks” constructs a future where AI’s growth happens covertly, leading to concealed, inaccessible power centers. Lastly, in “Emergence”, AI takes an organic evolutionary path, exhibiting unforeseen characteristics and capacities. These diverse scenarios are constructed with a keen understanding of AI’s potential, reflecting the depth of the authors’ interdisciplinary approach.

The Spectrum of AI Risks and Their Broader Philosophical Context

These four scenarios for AI development furnish a fertile ground for philosophical contemplation. Each scenario implicates distinct ethical, existential, and societal dimensions, demanding a versatile philosophical framework for analysis. “Balancing Act”, exemplifying a regulated progression of AI, broaches the age-old philosophical debate on freedom versus control and the moral conundrums associated with regulatory practices. “Accelerating Change” nudges us to consider the very concept of human identity and purpose in a future dominated by superintelligent entities. “Shadow Intelligent Networks” brings to light a potential future where power structures are concealed and unregulated, echoing elements of Foucault’s panopticism and revisiting concepts of power, knowledge, and their confluence. “Emergence”, with its focus on organic evolution of AI, prompts a dialogue on philosophical naturalism, while also raising queries about unpredictability and the inherent limitations of human foresight. These scenarios, collectively, invite profound introspection about our existing philosophical frameworks and their adequacy in the face of an AI-pervaded future.

This exposition on AI risks situates the potential hazards within an extensive spectrum. The spectrum ranges from tangible, immediate concerns such as privacy violations and job displacement, to the existential risks linked with superintelligent AI, including the relinquishment of human autonomy. The spectrum of AI risks engages with wider socio-political and ethical landscapes, prompting us to grapple with the potential for asymmetries in power distribution, accountability dilemmas, and ethical quandaries tied to autonomy and human rights. By placing these risks in a broader context, the authors effectively extends the discourse beyond the technical realm, highlighting the multidimensionality of the issues at hand and emphasizing the need for an integrated, cross-disciplinary approach. This lens encourages a reevaluation of established philosophical premises to comprehend and address the emerging realities of our future with AI.

And while this research is an illuminating exploration into the possible futures of AI, it simultaneously highlights a myriad of avenues for further research. The task of elucidating the connections between AI, society, and philosophical thought remains an ongoing process, requiring more nuanced perspectives. Areas that warrant further investigation include deeper dives into specific societal changes predicated by AI, such as shifts in economic structures, political systems, or bioethical norms. The potential impacts of AI on human consciousness and the conception of ‘self’ also offer fertile ground for research. Furthermore, the study of mitigation strategies for AI risks, including the development of robust ethical frameworks for AI usage, needs to be brought to the forefront. Such an examination may entail both an expansion of traditional philosophical discourses and an exploration of innovative, AI-informed paradigms.

Abstract

Artificial Intelligence (AI) is one of the most transformative technologies of the 21st century. The extent and scope of future AI capabilities remain a key uncertainty, with widespread disagreement on timelines and potential impacts. As nations and technology companies race toward greater complexity and autonomy in AI systems, there are concerns over the extent of integration and oversight of opaque AI decision processes. This is especially true in the subfield of machine learning (ML), where systems learn to optimize objectives without human assistance. Objectives can be imperfectly specified or executed in an unexpected or potentially harmful way. This becomes more concerning as systems increase in power and autonomy, where an abrupt capability jump could result in unexpected shifts in power dynamics or even catastrophic failures. This study presents a hierarchical complex systems framework to model AI risk and provide a template for alternative futures analysis. Survey data were collected from domain experts in the public and private sectors to classify AI impact and likelihood. The results show increased uncertainty over the powerful AI agent scenario, confidence in multiagent environments, and increased concern over AI alignment failures and influence-seeking behavior.

Examining the differential risk from high-level artificial intelligence and the question of control
Posted on

(Work in Progress) Resilient Failure Modes of AI Alignment

Resilient Failure Modes of AI Alignment

This project is housed at the Institute of Futures Research and relates to understanding and regularizing challenges to goal and value alignment in artificial intelligent (AI) systems when those systems exhibit nontrivial degrees of behavioral freedom and flexibility, and agency. Of particular concern are resilient failure modes, that is, failure modes that are intractable to methodological or technological resolution, owing to e.g. fundamental conflicts in the underlying ethical theory, or epistemic issues such as persistent ambiguity between the ethical theory, empirical facts, and any world models and policies held by the AI.

I will also be characterizing a resilient failure mode which has not apparently been addressed in the extant literature: misalignment incurred when reasoning and acting from shifting levels of abstraction. An intelligence apparently aligned in its outputs via some mechanism to a state space is not guaranteed to be aligned in the event that state space expands, for instance, through in-context learning or reasoning upon metastatements. This project will motivate, clarify, and formalize this failure mode as it pertains to artificial intelligence systems.

Within the scope of this research project, I am conducting a review of the literature pertaining to artificial intelligence alignment methods and failure modes, epistemological challenges to goal and value alignment, impossibility theorems in population and utilitarian ethics, and the nature of agency as it pertains to artifacts. A nonexhaustive bibliography follows.

I am greatly interested in potential feedback on this project, and suggestions for further reading.

References

Dario Amodei, Chris Olah, Jacob Steinhardt, Paul Christiano, John Schulman, & Dan Mané. (2016). Concrete Problems in AI Safety. https://doi.org/10.48550/arXiv.1606.06565

Peter Eckersley. (2019). Impossibility and Uncertainty Theorems in AI Value Alignment (or why your AGI should not have a utility function). https://doi.org/10.48550/arXiv.1901.00064

Dylan Hadfield-Menell, Anca Dragan, Pieter Abbeel, & Stuart Russell. (2016). Cooperative Inverse Reinforcement Learning. https://doi.org/10.48550/arXiv.1606.03137

Jan Leike, Miljan Martic, Victoria Krakovna, Pedro A. Ortega, Tom Everitt, Andrew Lefrancq, Laurent Orseau, & Shane Legg. (2017). AI Safety Gridworlds. https://doi.org/10.48550/arXiv.1711.09883

Scott McLean, Gemma J. M. Read, Jason Thompson, Chris Baber, Neville A. Stanton & Paul M. Salmon(2023)The risks associated with Artificial General Intelligence: A systematic review,Journal of Experimental & Theoretical Artificial Intelligence,35:5,649-663,DOI: 10.1080/0952813X.2021.1964003

Richard Ngo, Lawrence Chan, & Sören Mindermann. (2023). The alignment problem from a deep learning perspective. https://doi.org/10.48550/arXiv.2209.00626

Petersen, S. (2017). Superintelligence as Superethical. In P. Lin, K. Abney, & R. Jenkins (Eds.), Robot Ethics 2. 0: New Challenges in Philosophy, Law, and Society (pp. 322–337). New York, USA: Oxford University Press.

Max Reuter, & William Schulze. (2023). I’m Afraid I Can’t Do That: Predicting Prompt Refusal in Black-Box Generative Language Models. https://doi.org/10.48550/arXiv.2306.03423

Jonas Schuett, Noemi Dreksler, Markus Anderljung, David McCaffary, Lennart Heim, Emma Bluemke, & Ben Garfinkel. (2023). Towards best practices in AGI safety and governance: A survey of expert opinion. https://doi.org/10.48550/arXiv.2305.07153

Open Ended Learning Team, Adam Stooke, Anuj Mahajan, Catarina Barros, Charlie Deck, Jakob Bauer, Jakub Sygnowski, Maja Trebacz, Max Jaderberg, Michael Mathieu, Nat McAleese, Nathalie Bradley-Schmieg, Nathaniel Wong, Nicolas Porcel, Roberta Raileanu, Steph Hughes-Fitt, Valentin Dalibard, & Wojciech Marian Czarnecki. (2021). Open-Ended Learning Leads to Generally Capable Agents. https://doi.org/10.48550/arXiv.2107.12808

Roman V. Yampolskiy(2014)Utility function security in artificially intelligent agents,Journal of Experimental & Theoretical Artificial Intelligence,26:3,373-389. https://doi.org/10.1080/0952813X.2014.895114

Posted on

(Work in Progress) Scientific Theory and the Epistemology of Neural Networks

Scientific Theory and the Epistemology of Neural Networks

This project is housed at the Institute of Futures Research and seeks to address some challenges associated with the interpretability, explainability, and comprehensibility of neural networks—often termed ‘black boxes’ due to alleged epistemic opacity. Despite these limitations, I propose that neural network-generated knowledge can be epistemically licensed when they align with the theoretical requirements of scientific theories. Specifically, I focus on scientific theories that can be effectively represented through structures, or formal systems of symbols, statements, and inference rules. My goal is to establish a framework that positions neural networks as a plausible intermediary between terms and empirical statements in the formal apparatus of scientific theory, as satisfied traditionally by theoretical statements. This approach would bridge the gap between the computations and statistical results of neural networks and the epistemic objectives of science, and address concerns associated with the epistemic opacity of these models. By advancing a newly probabilistic account of scientific theories centered on neural networks, I hope to contribute new perspectives to the discourse on the role and interpretation of AI in scientific inquiry and the philosophy of science.

Objectives of this project include:

  • Clearly defining and operationalizing in a philosophical context such notions as artificial intelligence, artificial neural network, and the structure of scientific theory
  • Given an account of scientific theory, critically examining the available theoretical apparatus and understanding the role of ANNs in the production of science knowledge
  • Exploring the senses of epistemic opacity implicated by ANNs, and identifying those most relevant to the project
  • Understanding the scope of epistemic concerns surrounding the use of ANNs in the production of science knowledge
  • Providing a framework for translating the function and properties of ANNs to the structure, both syntactic and semantic, of theoretical statements in scientific theory
  • Demonstrating that ANNs satisfy the requirements of theoretical statements of scientific theory, epistemic and formal
  • Providing additional informal motivation towards the epistemic license of ANNs

This approach is not without complications. In particular, the recourse to ANNs in the generation of science knowledge introduces a novel source of uncertainty. If artificial neural networks are the objects of epistemic license in a scientific theory, whatever uncertainty pervades the algorithm reflexively pervades the generated science knowledge, which we would take to be (empirical) hypotheses of the theory. Furthermore, we may decide that such theories, while being adequately prescriptive, are nevertheless inadequately descriptive and transparent. They may be inadequately explanatory, and introduce a novel uncertainty when attempting to reproduce their results.

Within the scope of this research project, I am conducting a review of the literature pertaining to syntactic and semantic accounts of scientific theory, epistemological challenges to neural network methods, the formal specification of artificial neural networks, and contributions of neural network-based research to science knowledge. A nonexhaustive bibliography follows.

I am greatly interested in potential feedback on this project, and suggestions for further reading.

References

Almada, M. (2019). Human Intervention in Automated Decision-Making: Toward the Construction of Contestable Systems. Proceedings of the Seventeenth International Conference on Artificial Intelligence and Law, 2–11. Presented at the Montreal, QC, Canada. doi:10.1145/3322640.3326699

Andoni, A., Panigrahy, R., Valiant, G., & Zhang, L. (22–24 Jun 2014). Learning Polynomials with Neural Networks. In E. P. Xing & T. Jebara (Eds.), Proceedings of the 31st International Conference on Machine Learning (pp. 1908–1916). Retrieved from https://proceedings.mlr.press/v32/andoni14.html

Ananny, M., & Crawford, K. (12 2016). Seeing without knowing: Limitations of the transparency ideal and its application to algorithmic accountability. New Media & Society, 0, 146144481667664. doi:10.1177/1461444816676645

Barredo Arrieta, A., Díaz-Rodríguez, N., Del Ser, J., Bennetot, A., Tabik, S., Barbado, A., … Herrera, F. (2020). Explainable Artificial Intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI. Information Fusion, 58, 82–115. doi:10.1016/j.inffus.2019.12.012

Barron, A. (06 1993). Barron, A.E.: Universal approximation bounds for superpositions of a sigmoidal function. IEEE Trans. on Information Theory 39, 930-945. Information Theory, IEEE Transactions On, 39, 930–945. doi:10.1109/18.256500

Bennett, M.T., Maruyama, Y. (2022). The Artificial Scientist: Logicist, Emergentist, and Universalist Approaches to Artificial General Intelligence. In: Goertzel, B., Iklé, M., Potapov, A. (eds) Artificial General Intelligence. AGI 2021. Lecture Notes in Computer Science(), vol 13154. Springer, Cham. https://doi.org/10.1007/978-3-030-93758-4_6

Binns, R., Van Kleek, M., Veale, M., Lyngs, U., Zhao, J., & Shadbolt, N. (2018). “It”s Reducing a Human Being to a Percentage’. Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems. doi:10.1145/3173574.3173951

Boge, F.J. Two Dimensions of Opacity and the Deep Learning Predicament. Minds & Machines 32, 43–75 (2022). https://doi.org/10.1007/s11023-021-09569-4

Breen, P. G., Foley, C. N., Boekholt, T., & Zwart, S. P. (2020). Newton versus the machine: solving the chaotic three-body problem using deep neural networks. Monthly Notices of the Royal Astronomical Society, 494(2), 2465–2470. doi:10.1093/mnras/staa713

Burrell, J. (2016). How the Machine \textquoteleftThinks\textquoteright: Understanding Opacity in Machine Learning Algorithms. Big Data and Society, 3(1). doi:10.1177/2053951715622512

Carabantes, M. (06 2020). Black-box artificial intelligence: an epistemological and critical analysis. AI & SOCIETY, 35. doi:10.1007/s00146-019-00888-w

Carnap, R., & Gardner, M. (1966). Philosophical Foundations of Physics: An Introduction to the Philosophy of Science. Retrieved from https://books.google.com/books?id=rP3YAAAAIAAJ

Carvalho, D., Pereira, E., & Cardoso, J. (07 2019). Machine Learning Interpretability: A Survey on Methods and Metrics. Electronics, 8, 832. doi:10.3390/electronics8080832

Cichy, R., & Kaiser, D. (02 2019). Deep Neural Networks as Scientific Models. Trends in Cognitive Sciences, 23. doi:10.1016/j.tics.2019.01.009

Doran, D., Schulz, S., & Besold, T. R. (2017). What Does Explainable AI Really Mean? A New Conceptualization of Perspectives. ArXiv, abs/1710.00794.

Dressel, J., & Farid, H. (01 2018). The accuracy, fairness, and limits of predicting recidivism. Science Advances, 4, eaao5580. doi:10.1126/sciadv.aao5580

Dorffner, G., Wiklicky, H., & Prem, E. (1994). Formal neural network specification and its implications on standardization. Computer Standards & Interfaces, 16(3), 205–219. doi:10.1016/0920-5489(94)90012-4

Durán, J., & Formanek, N. (12 2018). Grounds for Trust: Essential Epistemic Opacity and Computational Reliabilism. Minds and Machines. doi:10.1007/s11023-018-9481-6

Durán, J. M., & Jongsma, K. R. (2021). Who is afraid of black box algorithms? On the epistemological and ethical basis of trust in medical AI. Journal of Medical Ethics, 47(5), 329–335. doi:10.1136/medethics-2020-106820

Fiesler, E. (1994). Neural network classification and formalization. Computer Standards & Interfaces, 16(3), 231–239. doi:10.1016/0920-5489(94)90014-0

Gilpin, L., Bau, D., Yuan, B., Bajwa, A., Specter, M., & Kagal, L. (10 2018). Explaining Explanations: An Overview of Interpretability of Machine Learning. 80–89. doi:10.1109/DSAA.2018.00018

Girard-Satabin, J., Charpiat, G., Chihani, Z., & Schoenauer, M. (2019). CAMUS: A Framework to Build Formal Specifications for Deep Perception Systems Using Simulators. CoRR, abs/1911.10735. Retrieved from http://arxiv.org/abs/1911.10735

Hollanek, T. (11 2020). AI transparency: a matter of reconciling design with critique. AI & SOCIETY, 1–9. doi:10.1007/s00146-020-01110-y

Humphreys, P. (2009). The Philosophical Novelty of Computer Simulation Methods. Synthese, 169(3), 615–626. doi:10.1007/s11229-008-9435-2

Ismailov, V. (2020). A three layer neural network can represent any discontinuous multivariate function. arXiv [Cs.LG]. Retrieved from http://arxiv.org/abs/2012.03016

Jumper, J., Evans, R., Pritzel, A. et al. Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589 (2021). https://doi.org/10.1038/s41586-021-03819-2

Karniadakis, G.E., Kevrekidis, I.G., Lu, L. et al. Physics-informed machine learning. Nat Rev Phys 3, 422–440 (2021). https://doi.org/10.1038/s42254-021-00314-5

Kim, J. (1988). What Is ‘Naturalized Epistemology?’ Philosophical Perspectives, 2, 381–405. Retrieved from http://www.jstor.org/stable/2214082

Knuuttila, T. (2011). Modelling and representing: An artefactual approach to model-based representation. Studies in History and Philosophy of Science Part A, 42(2), 262–271. doi:10.1016/j.shpsa.2010.11.034

Marquis, P., Papini, O., & Prade, H. (2020). A Guided Tour of Artificial Intelligence Research: Volume III: Interfaces and Applications of Artificial Intelligence. Retrieved from https://books.google.com/books?id=z07iDwAAQBAJ

Mahbooba, B., Timilsina, M., Sahal, R., & Serrano, M. (01 2021). Explainable Artificial Intelligence (XAI) to Enhance Trust Management in Intrusion Detection Systems Using Decision Tree Model. Complexity, 2021, 11. doi:10.1155/2021/6634811

Martínez-Ordaz, M. del R. (2023). Scientific understanding through big data: From ignorance to insights to understanding. Possibility Studies & Society, 0(0). https://doi.org/10.1177/27538699231176523

Montavon, G., Samek, W., & Müller, K.-R. (2018). Methods for interpreting and understanding deep neural networks. Digital Signal Processing, 73, 1–15. doi:10.1016/j.dsp.2017.10.011

Pappas, G. S. (2012). Justification and Knowledge: New Studies in Epistemology. Retrieved from https://books.google.com/books?id=6FlgBgAAQBAJ

Ramsey, W. (1997). Do Connectionist Representations Earn Their Explanatory Keep? Mind and Language, 12(1), 34–66. doi:10.1111/j.1468-0017.1997.tb00061.x

Ribeiro, M. T., Singh, S., & Guestrin, C. (2016). ‘Why Should I Trust You?’: Explaining the Predictions of Any Classifier. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 1135–1144. Presented at the San Francisco, California, USA. doi:10.1145/2939672.2939778

Rudin, C. (05 2019). Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature Machine Intelligence, 1, 206–215. doi:10.1038/s42256-019-0048-x

Samek, W., Wiegand, T., & Müller, K.-R. (10 2017). Explainable Artificial Intelligence: Understanding, Visualizing and Interpreting Deep Learning Models. ITU Journal: ICT Discoveries – Special Issue 1 – The Impact of Artificial Intelligence (AI) on Communication Networks and Services, 1, 1–10.

Sanchez-Gonzalez, A., Godwin, J., Pfaff, T., Ying, R., Leskovec, J., & Battaglia, P. W. (2020). Learning to Simulate Complex Physics with Graph Networks. CoRR, abs/2002.09405. Retrieved from https://arxiv.org/abs/2002.09405

Seshia, S. A., Desai, A., Dreossi, T., Fremont, D. J., Ghosh, S., Kim, E., … Yue, X. (2018). Formal Specification for Deep Neural Networks. In S. K. Lahiri & C. Wang (Eds.), Automated Technology for Verification and Analysis (pp. 20–34). Cham: Springer International Publishing.

Srećković, S., Berber, A. & Filipović, N. The Automated Laplacean Demon: How ML Challenges Our Views on Prediction and Explanation. Minds & Machines 32, 159–183 (2022). https://doi.org/10.1007/s11023-021-09575-6

Szymanski, L., & McCane, B. (2012). Deep, super-narrow neural network is a universal classifier. The 2012 International Joint Conference on Neural Networks (IJCNN), 1–8. doi:10.1109/IJCNN.2012.6252513

Taylor, B. J., & Darrah, M. A. (2005). Rule extraction as a formal method for the verification and validation of neural networks. Proceedings. 2005 IEEE International Joint Conference on Neural Networks, 2005., 5, 2915–2920 vol. 5. doi:10.1109/IJCNN.2005.1556388

Walmsley, J. (06 2021). Artificial intelligence and the value of transparency. AI & SOCIETY, 36, 1–11. doi:10.1007/s00146-020-01066-z

Wheeler, G. R., & Pereira, L. M. (2004). Epistemology and artificial intelligence. Journal of Applied Logic, 2(4), 469–493. doi:10.1016/j.jal.2004.07.007

Xu, T., Zhan, J., Garrod, O. G. B., Torr, P. H. S., Zhu, S.-C., Ince, R. A. A., & Schyns, P. G. (2018). Deeper Interpretability of Deep Networks. arXiv [Cs.CV]. Retrieved from http://arxiv.org/abs/1811.07807

Zednik, C. (06 2021). Solving the Black Box Problem: A Normative Framework for Explainable Artificial Intelligence. Philosophy & Technology, 34. doi:10.1007/s13347-019-00382-7

Zerilli, J., Knott, A., Maclaurin, J., & Gavaghan, C. (12 2019). Transparency in Algorithmic and Human Decision-Making: Is There a Double Standard? Philosophy & Technology, 32. doi:10.1007/s13347-018-0330-6

Highly accurate protein structure prediction with AlphaFold. (08 2021). Nature, 596, 583–589. doi:10.1038/s41586-021-03819-2

Posted on

(Featured) Friendly AI will still be our master. Or, why we should not want to be the pets of super-intelligent computers

Friendly AI will still be our master. Or, why we should not want to be the pets of super-intelligent computers

As our technological capabilities advance at an accelerating pace, so too does the pertinence of the hypothetical conundrum posed by super-intelligent artificial intelligence (AI) and its implications for human freedom. Robert Sparrow examines these implications, drawing extensively from political philosophy and conceptions of agency, and provides an analysis of the societal implications of super-intelligence from a uniquely philosophical standpoint. The author adopts a nuanced perspective, proposing that even benevolent, friendly AI may threaten human freedom in its capability to dominate, consciously or not, its human counterparts. It is this paradox, situated within the broader philosophical discourse of freedom versus domination, that provides the nucleus of this analysis.

The research is grounded in the seminal work of philosopher Philip Pettit, particularly his doctrine of republican freedom. This doctrine centers on the belief that freedom is not merely the absence of interference (negative liberty) but is, critically, the absence of domination or the ability to interfere at will. Pettit famously encapsulated this concept in his metaphor of the “eyeball test,” positing that one is free only when they can look others in the eye without fear or subservience. As we explore the intersection of Pettit’s philosophy and the hypothetical reality of a super-intelligent AI, the profound significance of this test in determining the future of human freedom in a world shared with AI comes sharply into focus.

The “Friendly AI” Problem

Robert Sparrow makes an acute distinction between “friendly” AI and its potential to dominate humanity. The Friendly AI problem stems from the plausible notion that super-intelligent AI, regardless of its benevolence or adherence to human values, may still pose a significant threat to human freedom due to its inherent capacity for domination. A benevolent AI could feasibly operate in a dictatorial manner, modulating its interference in human life based on its determination of human interests. However, a critical distinction must be drawn: a benevolent dictator, even though acting in our interests, is still a dictator. As the author of the article pointedly remarks, to be “free” to act as one wishes only at the behest of another entity, even a well-meaning one, is not true freedom.

Herein lies the crux of the Friendly AI problem: the ability of an AI entity to act in accordance with human interests does not automatically guarantee human freedom. Freedom, as delineated by the republicanism of Pettit, requires resilience; it must not dissolve upon the whims of a more powerful entity. Thus, for the exercise of power by AI to be compatible with human freedom, it must be possible for humans to resist it. One might propose that a genuinely Friendly AI would solicit human input before interfering in our affairs, serving as an efficient executor of our will rather than as a prescriptive entity. Yet, this proposition does not satisfactorily resolve the core tension between AI’s power and freedom and our own. Ultimately, any freedom we might enjoy under a superintelligent AI would be contingent upon the AI’s will, a position which reveals the inherent vulnerability and potential for domination inherent in the Friendly AI concept.

Superintelligence

Bostrom’s notion of Superintelligence, as outlined by Sparrow, posits an AI entity capable of outperforming the best human brains in nearly every economically relevant field. However, the potential domination by such an entity forms the bedrock of the philosophical conflict between benevolence and domination. Drawing on Pettit’s theory of republicanism, it becomes clear that benevolence alone, even if perfectly calibrated to human interests, does not suffice to guarantee freedom. The very ability of a superintelligent AI to interfere unilaterally in human affairs, regardless of its intent, embodies the antithesis of Pettit’s non-domination principle. The analysis further draws attention to the paradox inherent in relying on an external, powerful entity for the regulation of our interests, effectively highlighting the existential risk associated with superintelligent AI. While a superintelligent AI may act in line with human interests, its potential for domination raises questions about the plausibility of achieving a truly “Friendly AI”, a challenge that resonates with the larger discourse on freedom and domination in philosophical studies.

Freedom, Status, and the ‘Eyeball Test’

The question of human freedom in the context of a superintelligent AI intersects with Pettit’s conceptualization of the ‘eyeball test’. In his philosophy, the notion of freedom pivots on the individual’s status within society – a status conferred when one can ‘look others in the eye without reason for fear or deference’. This perspective becomes especially poignant when viewed in the light of a superintelligent entity’s potential dominion. Under such circumstances, the capacity for humans to pass the ‘eyeball test’ could be seriously undermined, as the superintelligent AI, by virtue of its cognitive superiority, could induce both fear and deference. The state of being subjected to the AI’s superior will could consequently impair our ability to ‘look it in the eye’, thereby eroding the human status required for true freedom. This analysis deepens the philosophical understanding of freedom and its inextricable link with status, while simultaneously challenging the concept of a ‘Friendly AI’ from the perspective of republican theory.

The Negative Liberty Doctrine and Technocratic Framing of AI

Berlin’s bifurcation of liberty into negative and positive spheres finds particular resonance in the context of superintelligent AI, and as such, provides a useful framework for interpreting the dominance problem. From a negative liberty perspective – that is, the absence of coercion or interference – the advent of a superintelligent AI could be seen as promoting freedom. However, the technocratic framing of AI, often characterized by an overemphasis on instrumental logic and utility maximization, may inadvertently favor this negative liberty doctrine, potentially to the detriment of positive freedom. This is to say, while an AI’s superior decision-making capabilities could minimize human interference in various spheres of life, it could also inadvertently curtail positive freedom – the opportunity for self-realization and autonomy. As such, this underscores the importance of incorporating broader philosophical considerations into AI research and development, beyond the narrow confines of technocratic perspectives.

This fusion of philosophy and AI research necessitates the introduction of considerations beyond the merely technical and into the sphere of ethics and moral philosophy. The potential for domination by superintelligent AI systems underscores the need for research that specifically targets these concerns, particularly in relation to upholding principles of human dignity, autonomy, and positive freedom. However, achieving this requires a re-evaluation of our current paradigms of AI development that often valorize utility maximization and efficiency. Instead, an approach that truly appreciates the full depth of the challenge must also involve a careful examination of the philosophical underpinnings that inform the design and operation of AI systems. As such, future research in this arena ought to be a collaborative effort between philosophers, ethicists, AI researchers, and policymakers, aimed at defining a coherent set of values and ethical guidelines for the development and use of superintelligent AI.

Abstract

When asked about humanity’s future relationship with computers, Marvin Minsky famously replied “If we’re lucky, they might decide to keep us as pets”. A number of eminent authorities continue to argue that there is a real danger that “super-intelligent” machines will enslave—perhaps even destroy—humanity. One might think that it would swiftly follow that we should abandon the pursuit of AI. Instead, most of those who purport to be concerned about the existential threat posed by AI default to worrying about what they call the “Friendly AI problem”. Roughly speaking this is the question of how we might ensure that the AI that will develop from the first AI that we create will remain sympathetic to humanity and continue to serve, or at least take account of, our interests. In this paper I draw on the “neo-republican” philosophy of Philip Pettit to argue that solving the Friendly AI problem would not change the fact that the advent of super-intelligent AI would be disastrous for humanity by virtue of rendering us the slaves of machines. A key insight of the republican tradition is that freedom requires equality of a certain sort, which is clearly lacking between pets and their owners. Benevolence is not enough. As long as AI has the power to interfere in humanity’s choices, and the capacity to do so without reference to our interests, then it will dominate us and thereby render us unfree. The pets of kind owners are still pets, which is not a status which humanity should embrace. If we really think that there is a risk that research on AI will lead to the emergence of a superintelligence, then we need to think again about the wisdom of researching AI at all.

Friendly AI will still be our master. Or, why we should not want to be the pets of super-intelligent computers
Posted on

(Featured) The Ethics of Terminology: Can We Use Human Terms to Describe AI?

The Ethics of Terminology: Can We Use Human Terms to Describe AI?

The philosophical discourse on artificial intelligence (AI) often negotiates the boundary of the human-anthropocentric worldview, pivoting around the use of human attributes to describe and assess AI. In this context, the research article by Ophelia Deroy presents a compelling inquiry into our linguistic and cognitive tendency to ascribe human characteristics, particularly “trustworthiness,” to artificial entities. In an attempt to unravel the philosophical implications and ramifications of this anthropomorphism, the author explores three conceptual frameworks – new ontological category, extended human-category, and Deroy’s semi-propositional beliefs. The divergence among these perspectives underscores the complexity of the issue, highlighting how our conceptions of AI shape our interactions with and attitudes towards it.

In addition to ontological and communicative aspects, the article scrutinizes the legal dimension of AI personhood. It analyzes the merits and shortcomings of the legal argument for ascribing personhood to AI, juxtaposing it with the established notion of corporate personhood. Although this comparison offers certain pragmatic and epistemic advantages, it does not unequivocally endorse the uncritical application of human terminology to AI. Through this multi-faceted analysis, the research article integrates perspectives from philosophy, cognitive science, and law, extending the ongoing discourse about AI into uncharted territories. The examination of AI within this framework thus emerges as an indispensable part of philosophical futures studies.

Understanding Folk Concepts of AI

The exploration of folk concepts of AI is critical in understanding how people conceive and interpret artificial intelligence within their worldview. Ophelia Deroy meticulously dissects these concepts by challenging the prevalent ascription of ‘trustworthiness’ to AI. The article emphasizes the potential mismatch between our cognitive conception of trust in humans and the attributes usually associated with AI, such as reliability or predictability. The focus is not only on the logical inconsistencies of such anthropomorphic attributions but also on the potential for miscommunication they could engender, especially given the complexity and variability of the term ‘trustworthiness’ across cultures and languages.

The author employs an interesting analytical angle by exploring the notion of AI as a possible extension of the human category, or alternatively, as a distinct ontological category. The question at hand is whether people perceive AI as fundamentally different from humans or merely view them as extreme non-prototypical cases of humans. This consideration reflects the complex cognitive landscape we navigate when dealing with AI, pointing towards the potential ontological ambiguity surrounding AI. Understanding these folk concepts and the mental models they reflect not only enriches our comprehension of AI from a sociocultural perspective but also yields important insights for the development and communication strategies of AI technologies.

Human Terms and their Implications, Legal Argument

The linguistic choice of using human terms such as “trustworthiness” to describe AI, arguably entrenched in anthropocentric reasoning, poses substantial problems. The author identifies three interpretations of how people categorize AI: an extension of the human category, a distinct ontological category, or a semi-propositional belief akin to religious or spiritual constructs. This last interpretation is particularly illuminating, suggesting that people might hold inconsistent beliefs about AI without considering them irrational. This offers a crucial insight into how human language shapes our understanding and discourse about AI, potentially fostering misconceptions. Yet, the author points out, there is a lack of empirical evidence supporting the appropriateness of applying such human-centric terms to AI, raising questions about the legitimacy of this linguistic practice in both scientific and broader public contexts.

In the discussion of AI’s anthropomorphic portrayal, Deroy introduces a compelling legal perspective. Drawing parallels with the legal status granted to non-human entities like corporations, the author investigates whether AI could be treated as a “legal person,” a concept that could reconcile the use of human terms in AI discourse. However, this argument presents its own set of challenges and limitations. The text using such terms must clearly state that the analogical use of “trust” is with respect to legal persons and not actual persons, a nuance often overlooked in many texts. Moreover, the justification for using such legal fiction must weigh the potential benefits against possible costs or risks, a task best left to legal experts. Thus, despite its merits, the legal argument does not provide an unproblematic justification for humanizing AI discourse.

The Broader Philosophical Discourse and Future Directions

This study is an important contribution to the broader philosophical discourse, illuminating the intersection of linguistics, ethics, and futures studies. The argument challenges the conventional notion of language as a neutral medium, stressing the normative power of language in shaping societal perception of AI. This aligns with the poststructuralist argument that reality is socially constructed, extending it to a technological context. The insight that folk concepts, embedded in language, influence our collective vision of AI’s role echoes phenomenological philosophies which underscore the role of intersubjectivity in shaping our shared reality. The ethical implications arising from the anthropomorphic portrayal of AI resonate with moral philosophy, particularly debates on moral agency and personhood. Thus, this study reinforces the growing realization that philosophical reflections are integral to our navigation of an increasingly AI-infused future.

Furthermore, the research points towards several promising avenues for future investigation. The most apparent is an extension of this study across diverse cultures and languages to explore how varying linguistic contexts may shape differing conceptions of AI, revealing cultural variations in anthropomorphizing technology. A comparative study might yield valuable insights into the societal implications of folk concepts across the globe. Additionally, an exploration into the real-world impact of anthropomorphic language in AI discourse, such as its effects on policy-making and public sentiment towards AI, would be an enlightening sequel. Lastly, this work paves the way for developing an ethical framework to guide the linguistic portrayal of AI in public discourse, a timely topic given the accelerating integration of AI into our daily lives. Thus, this research sets a fertile ground for multidisciplinary inquiries into linguistics, sociology, ethics, and futures studies.

Abstract

Despite facing significant criticism for assigning human-like characteristics to artificial intelligence, phrases like “trustworthy AI” are still commonly used in official documents and ethical guidelines. It is essential to consider why institutions continue to use these phrases, even though they are controversial. This article critically evaluates various reasons for using these terms, including ontological, legal, communicative, and psychological arguments. All these justifications share the common feature of trying to justify the official use of terms like “trustworthy AI” by appealing to the need to reflect pre-existing facts, be it the ontological status, ways of representing AI or legal categories. The article challenges the justifications for these linguistic practices observed in the field of AI ethics and AI science communication. In particular, it takes aim at two main arguments. The first is the notion that ethical discourse can move forward without the need for philosophical clarification, bypassing existing debates. The second justification argues that it’s acceptable to use anthropomorphic terms because they are consistent with the common concepts of AI held by non-experts—exaggerating this time the existing evidence and ignoring the possibility that folk beliefs about AI are not consistent and come closer to semi-propositional beliefs. The article sounds a strong warning against the use of human-centric language when discussing AI, both in terms of principle and the potential consequences. It argues that the use of such terminology risks shaping public opinion in ways that could have negative outcomes.

The Ethics of Terminology: Can We Use Human Terms to Describe AI?
Posted on

(Featured) Predictive policing and algorithmic fairness

Predictive policing and algorithmic fairness

Tzu-Wei Hung and Chun-Ping Yen contribute to the discursive field of predictive policing algorithms (PPAs) and their intersection with structural discrimination. They examine the functioning of PPAs, and lay bare their potential for propagating existing biases in policing practices and thereby question the presumed neutrality of technological interventions in law enforcement. Their investigation underscores the technological manifestation of structural injustices, adding a critical dimension to our understanding of the relationship between modern predictive technologies and societal equity.

An essential aspect of the authors’ argument is the proposition that the root of the problem lies not in the predictive algorithms themselves, but in the biased actions and unjust social structures that shape their application. Their article places this contention within the broader philosophical context, emphasizing the often-overlooked social and political underpinnings of technological systems. Thus, it offers a pertinent contribution to futures studies, prompting a more nuanced understanding of the interplay between (hotly anticipated) advanced technologies like PPAs and the structural realities of societal injustice. The authors provide a robust challenge to deterministic narratives around technology, pointing to the integral role of societal context in determining the impact of predictive policing systems.

Conceptualizing Predictive Policing

Hung and Yen Scrutinize the correlation between data inputs, algorithmic design, and resultant predictions. Their analysis disrupts the popular conception of PPAs as inherently objective and unproblematic, instead illuminating the mechanisms by which structural biases can be inadvertently incorporated and perpetuated through these algorithmic systems. The article’s critical scrutiny of PPAs further elucidates the relational dynamics between data, predictive modeling, and the societal contexts in which they are deployed.

The authors advance the argument that the implications of PPAs extend beyond individual acts of discrimination to reinforce broader systems of structural bias and social injustice. By focusing on the role of PPAs in reproducing existing patterns of discrimination, they elevate the discussion beyond a simplistic focus on technological neutrality or objectivity, situating PPAs within a larger discourse on technological complicity in the perpetuation of social injustices. This perspective fundamentally challenges conventional thinking about PPAs, prompting a shift from an algorithm-centric view to one that acknowledges the socio-political realities that shape and are shaped by these technological systems.

Structural Discrimination, Predictive Policing, and Theoretical Frameworks

The study goes further in its analysis by arguing that discrimination perpetuated through PPAs is, in essence, a manifestation of broader structural discrimination within societal systems. This perspective illuminates the connections between predictive policing and systemic power imbalances, rendering visible the complex ways in which PPAs can reify and intensify existing social injustices. The authors critically underline the potentially negative impact of stakeholder involvement in predictive policing, postulating that equal participation may unintentionally replicate or amplify pre-existing injustices. The analysis posits that the sources of discrimination lie in biased police actions reflecting broader societal inequities rather than the algorithmic systems themselves. Hence, addressing these challenges necessitates a focus not merely on rectifying algorithmic anomalies, but on transforming the unjust structures that they echo.

The authors propose a transformative theoretical framework, referred to as the social safety net schema, which envisions PPAs as integrated within a broader social safety net. This schema reframes the purpose and functioning of PPAs, advocating for their use not to penalize but to predict social vulnerabilities and facilitate requisite assistance. This is a paradigm shift from crime-focused approaches to a welfare-oriented model that situates crime within socio-economic structures. In this schema, the role of predictive policing is reimagined, with crime predictions used as indicators of systemic inequities that necessitate targeted interventions and redistribution of resources. With this reorientation, predictive policing becomes a tool for unveiling societal disparities and assisting in welfare improvement. The implementation of this schema implies a commitment to equity rather than just equality, addressing the nuances and complexities of social realities and aiming at the underlying structures fostering discrimination.

Community and Stakeholder Involvement, and Implications for Future Research

The issue of stakeholder involvement is addressed with both depth and nuance. Acknowledging the criticality of involving diverse stakeholders in the governance and control of predictive policing technology, the authors assert that equal participation could inadvertently reproduce the extant societal disparities. In their view, a stronger representation of underrepresented groups in decision-making processes is vital. This necessitates more resources and mechanisms to ensure their voices are heard and acknowledged in shaping public policies and social structures. The role of local communities in this process is paramount; they act as informed advocates, ensuring the proper understanding and representation of disadvantaged groups. This framework, hence, pivots on a bottom-up approach to power and control over policing, ensuring democratic community control and fostering collective efficacy. The approach is envisioned to counterbalance the persisting inequality, thereby reducing the likelihood of discrimination and improving community control over policing.

The analysis brings forth notable implications for future academic inquiries and policy-making. It endorses the importance of scrutiny of social structures rather than the predictive algorithms themselves as the catalyst for discriminatory practices in predictive policing. This view drives the necessity of further research into the multifaceted intersection between social structures, law enforcement, and advanced predictive technologies. Moreover, it prompts consideration of how policies can be implemented to reflect this understanding, centering on creating a socially aware and equitable technological governance structure. The policy schema of the social safety net for predictive policing, as proposed by the authors, offers a starting point for such a discourse. Future research may focus on implementing and testing this schema, critically examining its effectiveness in mitigating discriminatory impacts of predictive policing, and identifying potential adjustments necessary for enhancing its efficiency and inclusivity. In essence, future inquiries and policy revisions should foster a context-sensitive, democratic, and community-focused approach to predictive policing.

Abstract

This paper examines racial discrimination and algorithmic bias in predictive policing algorithms (PPAs), an emerging technology designed to predict threats and suggest solutions in law enforcement. We first describe what discrimination is in a case study of Chicago’s PPA. We then explain their causes with Broadbent’s contrastive model of causation and causal diagrams. Based on the cognitive science literature, we also explain why fairness is not an objective truth discoverable in laboratories but has context-sensitive social meanings that need to be negotiated through democratic processes. With the above analysis, we next predict why some recommendations given in the bias reduction literature are not as effective as expected. Unlike the cliché highlighting equal participation for all stakeholders in predictive policing, we emphasize power structures to avoid hermeneutical lacunae. Finally, we aim to control PPA discrimination by proposing a governance solution—a framework of a social safety net.

Predictive policing and algorithmic fairness
Posted on

(Featured) Cognitive architectures for artificial intelligence ethics

Cognitive architectures for artificial intelligence ethics

The landscape of artificial intelligence (AI) is a complex and rapidly evolving field, one that increasingly intersects with ethical, philosophical, and societal considerations. The role of AI in shaping our future is now largely uncontested, with potential applications spanning an array of sectors from healthcare to education, logistics to creative industries. Of particular interest, however, is not merely the surface-level functionality of these AI systems, but the cognitive architectures underpinning them. Cognitive architectures, a theoretical blueprint for cognitive and intelligent behavior, essentially dictate how AI systems perceive, think, and act. They therefore represent a foundational aspect of AI design and hold substantial implications for how AI systems will interact with, and potentially transform, our broader societal structures.

Yet, the discourse surrounding these architectures is, to a large extent, bifurcated between two paradigms: the biological cognitive architecture and the functional cognitive architecture. The biological paradigm, primarily drawing from neuroscience and biology, emphasizes replicating the cognitive processes of the human brain. On the other hand, the functional paradigm, rooted more in computer science and engineering, is concerned with designing efficient systems capable of executing cognitive tasks, regardless of whether they emulate human cognitive processes. This fundamental divergence in design philosophy thus embodies distinct assumptions about the nature of cognition and intelligence, consequently shaping the way AI systems are created and how they might impact society. It is these paradigms, their implications, and their interplay with AI ethics principles, that form the main themes of this essay.

Frameworks for Understanding Cognitive Architectures and the Role of Mental Models in AI Design

Cognitive architectures, central to the progression of artificial intelligence, encapsulate the fundamental rules and structures that drive the operation of an intelligent agent. The research article situates its discussion within two dominant theoretical frameworks: symbolic and connectionist cognitive architectures. Symbolic cognitive architectures, rooted in the realm of logic and explicit representation, emphasize rule-based systems and algorithms. They are typified by their capacity for discrete, structured reasoning, often relating to high-level cognitive functions such as planning and problem-solving. This structured approach carries the advantage of interpretability, affording clearer insights into the decision-making processes.

On the other hand, connectionist cognitive architectures embody a divergent perspective, deriving their inspiration from biological neural networks. Connectionist models prioritize emergent behavior and learning from experience, expressed in the form of neural networks that adjust synaptic weights based on input. These architectures have exhibited exceptional performance in pattern recognition and adaptive learning scenarios. However, their opaque, ‘black-box’ nature presents challenges to understanding and predicting their behavior. The interplay between these two models, symbolizing the tension between the transparent but rigid symbolic approach and the flexible but opaque connectionist approach, forms the foundation upon which contemporary discussions of cognitive architectures in AI rest.

The incorporation of mental models in AI design represents a nexus where philosophical interpretations of cognition intersect with computational practicalities. The use of mental models, i.e., internal representations of the world and its operational mechanisms, is a significant bridge between biological and functional cognitive architectures. This highlights the philosophical significance of mental models in the study of AI design: they reflect the complex interplay between the reality we perceive and the reality we construct. The efficacy of mental models in AI system design underscores their pivotal role in knowledge acquisition and problem-solving. In the biological cognitive framework, mental models mimic human cognition’s non-linear, associative, and adaptive nature, thereby conforming to the cognitive isomorphism principle. On the other hand, the functional cognitive framework employs mental models as pragmatic tools for efficient task execution, demonstrating a utilitarian approach to cognition. Thus, the role of mental models in AI design serves as a litmus test for the philosophical assumptions underlying distinct cognitive architectures.

Philosophical Reflections and AI Ethics Principles in Relation to Cognitive Architectures

AI ethics principles, primarily those concerning autonomy, beneficence, and justice, possess substantial implications for the understanding and application of cognitive architectures. If we consider the biological framework, ethical considerations significantly arise concerning the autonomy and agency of AI systems. To what extent can, or should, an AI system with a human-like cognitive structure make independent decisions? The principle of beneficence—commitment to do good and prevent harm—profoundly impacts the design of functional cognitive architectures. Here, a tension surfaces between the utilitarian goal of optimized task execution and the prevention of potential harm resulting from such single-mindedness. Meanwhile, the principle of justice—fairness in the distribution of benefits and burdens—prompts critical scrutiny of the societal consequences of both architectures. As these models become more prevalent, we must continuously ask: Who benefits from these technologies, and who bears the potential harms? Consequently, the intricate intertwining of AI ethics principles with cognitive architectures brings philosophical discourse to the forefront of AI development, establishing its pivotal role in shaping the future of artificial cognition.

The philosophical discourse surrounding AI and cognitive architectures is deeply entwined with the ethical, ontological, and epistemological considerations inherent to AI design. On an ethical level, the discourse probes the societal implications of these technologies and the moral responsibilities of their developers. The questions of what AI is and what it could be—an ontological debate—become pressing as cognitive architectures increasingly mimic the complexities of the human mind. Furthermore, the epistemological dimension of this discourse explores the nature of AI’s knowledge acquisition and decision-making processes. This discourse, therefore, cannot be separated from the technological progression of AI, as the philosophical issues at play directly inform the design choices made. Thus, philosophical reflections are not merely theoretical musings but tangible influences on the future of AI and, by extension, society. As AI continues to evolve, the ongoing dialogue between philosophy and technology will be critical in guiding its development towards beneficial and ethical ends.

Future Directions for Research

Considering the rapid advancement of AI, cognitive architectures, and their deep-rooted philosophical implications, potential avenues for future research appear vast and multidimensional. It would be valuable to delve deeper into the empirical examination of cognitive architectures’ impact on decision-making processes in AI, quantitatively exploring their effect on AI reliability and behavior. A comparative study across different cognitive architecture models, analyzing their benefits and drawbacks in diverse real-world contexts, would further enrich the understanding of their practical applications. As ethical considerations take center stage, research exploring the development and implementation of ethical guidelines specific to cognitive architectures is essential. Notably, studies addressing the question of how to efficiently integrate philosophical perspectives into the technical development process could be transformative. Furthermore, in this era of advancing AI technologies, maintaining a dialogue between the technologists and the philosophers is crucial; thus, fostering interdisciplinary collaborations between AI research and philosophy should be a high priority in future research agendas.

Abstract

As artificial intelligence (AI) thrives and propagates through modern life, a key question to ask is how to include humans in future AI? Despite human involvement at every stage of the production process from conception and design through to implementation, modern AI is still often criticized for its “black box” characteristics. Sometimes, we do not know what really goes on inside or how and why certain conclusions are met. Future AI will face many dilemmas and ethical issues unforeseen by their creators beyond those commonly discussed (e.g., trolley problems and variants of it) and to which solutions cannot be hard-coded and are often still up for debate. Given the sensitivity of such social and ethical dilemmas and the implications of these for human society at large, when and if our AI make the “wrong” choice we need to understand how they got there in order to make corrections and prevent recurrences. This is particularly true in situations where human livelihoods are at stake (e.g., health, well-being, finance, law) or when major individual or household decisions are taken. Doing so requires opening up the “black box” of AI; especially as they act, interact, and adapt in a human world and how they interact with other AI in this world. In this article, we argue for the application of cognitive architectures for ethical AI. In particular, for their potential contributions to AI transparency, explainability, and accountability. We need to understand how our AI get to the solutions they do, and we should seek to do this on a deeper level in terms of the machine-equivalents of motivations, attitudes, values, and so on. The path to future AI is long and winding but it could arrive faster than we think. In order to harness the positive potential outcomes of AI for humans and society (and avoid the negatives), we need to understand AI more fully in the first place and we expect this will simultaneously contribute towards greater understanding of their human counterparts also.

Cognitive architectures for artificial intelligence ethics
Posted on

(Featured) Moral disagreement and artificial intelligence

Moral disagreement and artificial intelligence

Pamela Robinson proposes a robust examination of the methodological problems arising due to moral disagreement in the development and decision-making processes of artificial intelligence (AI). The central point of discussion is the formulation of ethical AI systems, in particular, the AI Decider, that needs to make decisions in cases where its decision subjects have moral disagreements. The author posits that the conundrum could potentially be managed using moral, compromise, or epistemic solutions.

The author systematically elucidates the possible solutions by presenting three categories. Moral solutions are proposed to involve choosing a moral theory and having AI align to it, like preference utilitarianism, thereby sidestepping disagreement by assuming moral consensus. Compromise solutions, on the other hand, suggest handling disagreement by aggregating moral views to arrive at a collective decision. The author introduces the Arrow’s impossibility theorem and Social Choice Theory as potential tools for AI decision-making. Lastly, epistemic solutions, arguably the most complex of the three, require the AI Decider to treat moral disagreement as evidence and adjust its decision accordingly. The author mentions several approaches within this category, such as reflective equilibrium, moral uncertainty, and moral hedging.

However, none of these solutions, the author asserts, can provide a perfect answer to the problem. Each solution is fraught with its own complexities and risks. Here, the concept of ‘moral risk,’ meaning the chance of getting things wrong morally, is introduced. The author postulates that the selection between an epistemic or compromise solution should depend on the moral risk involved. They argue that the methodological problem could be addressed by minimizing this moral risk, regardless of whether a moral, compromise, or epistemic solution is employed.

Delving into the broader philosophical themes, this paper reignites the enduring debate on the role and impact of moral relativism and objectivism within the sphere of artificial intelligence. The issues presented tie into the grand narrative of moral philosophy, particularly the discourse around meta-ethics and normative ethics, where differing moral perspectives invariably lead to dilemmas. The AI Decider, in this sense, mirrors the human condition where decision-making often requires navigating the labyrinth of moral disagreement. The author’s emphasis on moral risk provides a novel framework, bridging the gap between theoretical moral philosophy and the practical demands of AI ethics.

For future research, several intriguing pathways are suggested by this article. First, an in-depth exploration of the concept of ‘moral risk’ could illuminate new strategies for handling moral disagreement in AI decision-making. Comparative studies, analyzing the outcomes and repercussions of decisions made by an AI system utilizing moral, compromise, or epistemic solutions, could provide empirical evidence for the efficacy of these approaches. Lastly, given the dynamism of moral evolution, the impact of changes in societal moral views over time on an AI Decider’s decision-making process warrants investigation. This could include exploring how the AI system could effectively adapt to the evolution of moral consensus or disagreement within its decision subjects. Such future research could significantly enhance our understanding of ethical decision-making in AI systems, bringing us closer to the creation of more ethically aligned, responsive, and responsible artificial intelligence.

Abstract

Artificially intelligent systems will be used to make increasingly important decisions about us. Many of these decisions will have to be made without universal agreement about the relevant moral facts. For other kinds of disagreement, it is at least usually obvious what kind of solution is called for. What makes moral disagreement especially challenging is that there are three different ways of handling it. Moral solutions apply a moral theory or related principles and largely ignore the details of the disagreement. Compromise solutions apply a method of finding a compromise and taking information about the disagreement as input. Epistemic solutions apply an evidential rule that treats the details of the disagreement as evidence of moral truth. Proposals for all three kinds of solutions can be found in the AI ethics and value alignment literature, but little has been said to justify choosing one over the other. I argue that the choice is best framed in terms of moral risk.

Moral disagreement and artificial intelligence
Posted on

(Featured) We have to talk about emotional AI and crime

We have to talk about emotional AI and crime

Lena Podoletz investigates the utilization of emotional Artificial Intelligence (AI) within the context of law enforcement and criminal justice systems in a critical examination of the sociopolitical, legal, and ethical ramifications of this technology, contextualizing the analysis within the broader landscape of technological trends and potential future applications.

The opening part of the article is devoted to the intricacies of emotion recognition AI, specifically its definition, functionality, and the scientific foundations that inform its development. In dissecting these aspects, the author emphasizes the discrepancy between the common understanding of emotions and the way they are algorithmically conceptualized and processed. Key to this understanding is the recognition that emotional AI, in its current stage of development, relies heavily on theoretical constructs like the ‘basic emotions theory’ and the ‘circumplex model’, the limitations and biases of which can significantly impact its effective and ethical application in law enforcement and criminal justice contexts.

Subsequent sections of the article provide a rigorous evaluation of four areas of concern: accuracy and performance, bias, accountability, and privacy along with other rights and freedoms. The author underscores the need for distinguishing between different uses of emotional AI, stressing that the challenges presented in a law enforcement setting differ significantly from its application in other contexts, such as private homes or smart health environments. This examination extends to issues related to bias in algorithmic decision-making, where existing societal biases can be reproduced and amplified. The complex issue of accountability in emotional AI is also dissected, particularly in terms of attributing responsibility for decisions made by such systems. Finally, the author explores the intersection of emotional AI technologies with privacy and other human rights, indicating that the deployment of these systems can challenge individual autonomy and human dignity.

The thematic concerns presented in the article echo the larger philosophical discourse surrounding the role and implications of AI in society. The author’s evaluation of emotional AI is in line with post-humanist thought, which questions the Cartesian dualism of human and machine, and problematizes the reduction of complex human behaviors and emotions into codified, algorithmic processes. The exploration of bias, accountability, and privacy ties into ongoing debates around the ethics of AI, especially concerning notions of fairness, transparency, and justice in algorithmic decision-making. Moreover, the question of who holds responsibility when AI systems make mistakes or violate rights brings into focus the legal and philosophical concept of moral agency in the age of advanced AI.

Future research might delve deeper into how emotional AI, specifically within law enforcement and criminal justice systems, could be better regulated or standardized to address the highlighted concerns. It would be valuable to explore potential legislative and technical solutions to mitigate bias, improve accuracy, and establish clear lines of accountability. Moreover, further philosophical examination is needed to unpack the implications of emotional AI on our understanding of human emotions, agency, and rights in an increasingly technologized society. Finally, in line with futures studies philosophy, it would be beneficial to conceive of alternative trajectories for the development and deployment of emotional AI that are anchored in ethical foresight and participatory decision-making, thereby ensuring a future that upholds societal well-being and human dignity.

Abstract

Emotional AI is an emerging technology used to make probabilistic predictions about the emotional states of people using data sources, such as facial (micro)-movements, body language, vocal tone or the choice of words. The performance of such systems is heavily debated and so are the underlying scientific methods that serve as the basis for many such technologies. In this article I will engage with this new technology, and with the debates and literature that surround it. Working at the intersection of criminology, policing, surveillance and the study of emotional AI this paper explores and offers a framework of understanding the various issues that these technologies present particularly to liberal democracies. I argue that these technologies should not be deployed within public spaces because there is only a very weak evidence-base as to their effectiveness in a policing and security context, and even more importantly represent a major intrusion to people’s private lives and also represent a worrying extension of policing power because of the possibility that intentions and attitudes may be inferred. Further to this, the danger in the use of such invasive surveillance for the purpose of policing and crime prevention in urban spaces is that it potentially leads to a highly regulated and control-oriented society. I argue that emotion recognition has severe impacts on the right to the city by not only undertaking surveillance of existing situations but also making inferences and probabilistic predictions about future events as well as emotions and intentions.

We have to talk about emotional AI and crime