In a comprehensive position paper published in AI Perspectives, computer scientist Malik Ghallab evaluates the foundational requirements and systemic challenges associated with responsible artificial intelligence development, structuring the domain around two interdependent objectives: directing research and development toward socially beneficial applications and proactively mitigating human and social risks. The analysis observes that artificial intelligence technologies have contributed to advancements reflected in global metrics such as the Human Development Index, enhancing medical prevention, diagnostic, and care procedures, assisting in environmental monitoring and preservation, advancing agricultural projects, optimizing urban and industrial infrastructures, improving transportation safety, and providing multilingual and multimodal communication interfaces. However, the author emphasizes that the rapid acceleration of technology development generates legitimate concerns regarding disruptive social strains, ethical dilemmas, and complex technical vulnerabilities that necessitate systematic investigation rather than passive reliance on market forces.
International bodies and professional organizations have increasingly mobilized to address these challenges through formal policy initiatives and collaborative frameworks. The position paper highlights efforts led by the United Nations and specialized agencies such as the International Telecommunication Union, the European Union through its High-Level Expert Group on AI, and the Organisation for Economic Co-operation and Development through the AI Policy Observatory and AIGO. Political leaders from the Group of Seven announced plans to establish an International Panel on AI, modeled after the Intergovernmental Panel on Climate Change, while technical societies including the IEEE through its Ethically Aligned Design initiative, academic centers such as AI Now, HUMANe AI, and AI4People, and non-governmental foundations have initiated targeted investigations. The paper stresses that while these frameworks establish critical visibility, translating high-level ethical commitments into tangible technical practices requires addressing deep structural barriers within both research funding mechanisms and scientific methodologies.
Mobilizing artificial intelligence for social benefit demands concentrated alignment with broader global priorities, such as the United Nations Sustainable Development Goals. The paper notes that forums like the ITU AI for Global Good Summit have focused on specific application domains, including interpreting satellite imagery for agronomy under SDG 2 and environmental preservation under SDGs 6, 13, and 17, systematically collecting and disseminating open medical data regarding epidemics under SDG 3, and simulating urban dynamics for smart city management under SDG 11, alongside applications in education, water management, and industrial efficiency. Nevertheless, conventional market incentives favor rapid and high returns on investment, frequently neglecting the substantial capital requirements of early-stage, socially oriented projects. Although all OECD nations and multiple developing states maintain national AI strategies, public investments remain modest relative to commercial research budgets, underscoring the necessity of scaled-up public funding and targeted international cooperation to support vital non-profit initiatives such as the Thorn Spotlight project against human trafficking and ecological conservation efforts.
Beyond economic incentives, fostering beneficial applications requires overcoming disciplinary silos through extensive integrative research within artificial intelligence and across external disciplines. Ghallab observes that academic research traditionally favors specialized analytical methods and narrowly bounded community assumptions, which, while vital for formal scientific progress, remain insufficient for multifaceted societal challenges. Within the discipline, integrative research must bridge heterogeneous cognitive functions, unifying sensory data association with ontological extraction and domain reasoning. This agenda requires reconciling bottom-up machine learning correlations with top-down causal modeling, integrating mathematically diverse knowledge representations, and fusing disparate information sources. Externally, transdisciplinary collaboration must connect artificial intelligence specialists with domain experts, public stakeholders, and social actors, prioritizing empirical field efficacy over abstract computational benchmarks while addressing academic evaluation systems that disincentivize long-term, high-overhead collaborative ventures.
The second primary objective examined in the position paper concerns the mitigation of multifaceted risks, categorized into safety-critical system failures, user security and privacy vulnerabilities, and broad societal disruptions. In safety-critical domains where artificial intelligence governs physical systems endowed with sensory-motor capabilities and autonomy—such as surgical robotics, medical prostheses, automated transit, power grids, logistics networks, and surveillance mechanisms—the potential costs in human lives and environmental stability are severe. Unlike commercial aviation or invasive medical devices, relatively few artificial intelligence applications face rigorous certification mandates. The author argues that existing compliance declarations and informal descriptive standards are inadequate due to the opacity and computational complexity of advanced models, coupled with the intricate traceability of interacting hardware and software modules within rapidly scaling systems.
Addressing these safety-critical vulnerabilities requires expanding verification and validation methods from closed, well-modeled computational environments into open, partially observed, and dynamic operational spaces. Formal methods, deterministic analysis, stochastic techniques, and testing regimes must evolve to formally articulate operational assumptions, establish explicit behavioral boundaries, and determine core operating parameters including error probabilities, false positive rates, and parameter sensitivities. Critical unresolved technical questions identified in the paper include formally quantifying uncertainty within complex models such as medical diagnostics, enabling autonomous systems to perform online self-monitoring of environmental assumptions and adapt when conditions diverge, establishing compositional verification techniques for assemblies containing opaque black-box components, and developing robust verification protocols for systems designed to learn and update parameters continuously during operational deployment.
For individual users, artificial intelligence increasingly acts as an pervasive mediator between personal behavior and the digital sphere, interpreting natural language vocal inputs, indexing sensory data, and constructing continuous behavioral profiles. The position paper identifies widespread public ambivalence characterized by uncritical overconfidence in computational rationality alongside legitimate concerns regarding surveillance and unfounded anxieties surrounding artificial general consciousness. Technical challenges in this sphere encompass insufficient deployment of existing security standards in consumer devices and robots, the persistent vulnerability of neural networks to adversarial attacks, and the lack of algorithmic intelligibility. A decision-support architecture must be capable of elucidating its decision criteria in comprehensible terms, particularly since optimizing a mathematically defined objective function may conflict with personal human values or individual priorities.
The paper argues that the widespread lag in deploying robust confidentiality and security protections stems primarily from misaligned commercial incentives and weak regulatory enforcement. Dominant corporate platforms benefit from an asymmetric contractual dynamic, aggregating user interactions to harvest valuable predictive behavioral models that consumers cannot govern, even when formal consent policies exist for individual data points. The analysis notes that while regulatory frameworks such as the European Union General Data Protection Regulation advance data rights, formal ethics guidelines and corporate pledges remain fundamentally insufficient to resolve structural commercial imbalances. The author advocates for empirical legal research, regulatory intervention, and institutional innovations such as accountable data trusts that can serve as independent intermediaries between individual users and commercial platforms to protect collective privacy interests.
Examining systemic social hazards, the paper details how decision-support algorithms frequently incorporate detrimental biases, ranging from intentional distortion in commercially driven recommender systems to unexamined demographic prejudices embedded in historical training datasets across health, banking, insurance, hiring, and judicial sentencing. Additionally, advanced behavioral modeling and targeted optimization tools facilitate unprecedented capabilities for public manipulation, commercial steering, and political disruption, as demonstrated in high-profile democratic controversies. In economic systems, automated mechanisms like algorithmic pricing can generate artificial price increases akin to anti-competitive cartels without explicit collusion, directly contradicting traditional assumptions of self-regulating free markets, while high-frequency trading introduces destabilizing dynamics that remain incompletely understood and urgently require rigorous oversight.
The economic impact of automation represents another acute societal risk, with cited OECD estimates indicating that nine percent of jobs across twenty-one member countries face a high risk of automation, while an additional twenty to twenty-five percent exhibit medium vulnerability, potentially intensifying socio-economic inequality. In defense contexts, the dual-use nature and relative affordability of autonomous systems present severe risks of proliferation to non-state entities, yet digital weapons remain excluded from international treaties such as the Arms Trade Treaty. Furthermore, the paper critiques the theoretical viability of delegating moral choices to computational systems, rejecting the concept of autonomous moral agents. Ghallab concludes that because formal regulatory mechanisms require decades to deliberate and establish while technological adoption accelerates rapidly, society cannot rely on reactive governance; instead, researchers and policymakers must engage in proactive social experiments, computational simulation, and sustained interdisciplinary collaboration within science, technology, and society studies to steer technological deployment responsibly.
Subject of Research: Requirements and challenges for responsible artificial intelligence development and governance
Article Title: Responsible AI: requirements and challenges
Article References: Ghallab, M. (2019). Responsible AI: requirements and challenges. AI Perspectives, 1(1), Article 3. https://doi.org/10.1186/s42467-019-0003-z
Image Credits: AI Generated
DOI: 10.1186/s42467-019-0003-z
Keywords: Artificial Intelligence, Responsible AI, Verification and Validation, Algorithmic Bias, AI Safety, Public Policy, Social Good, Technology Regulation, Responsible, requirements, challenges, scientific research
News Source: Blake Davidson. (November 7, 2026). Position Paper Details Technical and Social Challenges for Responsible Artificial Intelligence. Scienmag.
