Search results for: explainability of aI

Authors: David Pham, Yongfeng Zhang

Abstract:

Graphs are an important structure for data storage and computation. Recent years have seen the success of deep learning on graphs such as Graph Neural Networks (GNN) on various data mining and machine learning tasks. However, most of the deep learning models on graphs cannot easily explain their predictions and are thus often labelled as “black boxes.” For example, Graph Attention Network (GAT) is a frequently used GNN architecture, which adopts an attention mechanism to carefully select the neighborhood nodes for message passing and aggregation. However, it is difficult to explain why certain neighbors are selected while others are not and how the selected neighbors contribute to the final classification result. In this paper, we present a graph learning model called Explainable Graph Attention Network (XGAT), which integrates graph attention modeling and explainability. We use a single model to target both the accuracy and explainability of problem spaces and show that in the context of graph attention modeling, we can design a unified neighborhood selection strategy that selects appropriate neighbor nodes for both better accuracy and enhanced explainability. To justify this, we conduct extensive experiments to better understand the behavior of our model under different conditions and show an increase in both accuracy and explainability.

Keywords: explainable AI, graph attention network, graph neural network, node classification

Procedia PDF Downloads 203

Authors: Manik Sharma, Ganapathy Krishnamurthi

Abstract:

The cost of some drugs and medical treatments has risen in recent years, that many patients are having to go without. A classification project could make researchers more efficient. One of the more surprising reasons behind the cost is how long it takes to bring new treatments to market. Despite improvements in technology and science, research and development continues to lag. In fact, finding new treatment takes, on average, more than 10 years and costs hundreds of millions of dollars. If successful, we could dramatically improve the industry's ability to model cellular images according to their relevant biology. In turn, greatly decreasing the cost of treatments and ensure these treatments get to patients faster. This work aims at solving a part of this problem by creating a cellular image classification model which can decipher the genetic perturbations in cell (occurring naturally or artificially). Another interesting question addressed is what makes the deep-learning model decide in a particular fashion, which can further help in demystifying the mechanism of action of certain perturbations and paves a way towards the explainability of the deep-learning model.

Keywords: cellular images, genetic perturbations, deep-learning, explainability

Procedia PDF Downloads 113

Authors: Debarati Bhaumik, Diptish Dey

Abstract:

Multilevel models, increasingly deployed in industries such as insurance, food production, and entertainment within functions such as marketing and supply chain management, need to be transparent and ethical. Applications usually result in binary classification within groups or hierarchies based on a set of input features. Using open-source datasets, we demonstrate that popular explainability methods, such as SHAP and LIME, consistently underperform inaccuracy when interpreting these models. They fail to predict the order of feature importance, the magnitudes, and occasionally even the nature of the feature contribution (negative versus positive contribution to the outcome). Besides accuracy, the computational intractability of SHAP for binomial classification is a cause of concern. For transparent and ethical applications of these hierarchical statistical models, sound audit frameworks need to be developed. In this paper, we propose an audit framework for technical assessment of multilevel regression models focusing on three aspects: (i) model assumptions & statistical properties, (ii) model transparency using different explainability methods, and (iii) discrimination assessment. To this end, we undertake a quantitative approach and compare intrinsic model methods with SHAP and LIME. The framework comprises a shortlist of KPIs, such as PoCE (Percentage of Correct Explanations) and MDG (Mean Discriminatory Gap) per feature, for each of these three aspects. A traffic light risk assessment method is furthermore coupled to these KPIs. The audit framework will assist regulatory bodies in performing conformity assessments of AI systems using multilevel binomial classification models at businesses. It will also benefit businesses deploying multilevel models to be future-proof and aligned with the European Commission’s proposed Regulation on Artificial Intelligence.

Keywords: audit, multilevel model, model transparency, model explainability, discrimination, ethics

Procedia PDF Downloads 95

Authors: Gaurav Shinde, Sai Charan Gongiguntla, Prajwal Shirur, Ahmed Hambaba

Abstract:

Health issues are significantly increasing, putting a substantial strain on healthcare services. This has accelerated the integration of machine learning in healthcare, particularly following the COVID-19 pandemic. The utilization of machine learning in healthcare has grown significantly. We introduce DeClEx, a pipeline that ensures that data mirrors real-world settings by incorporating Gaussian noise and blur and employing autoencoders to learn intermediate feature representations. Subsequently, our convolutional neural network, paired with spatial attention, provides comparable accuracy to state-of-the-art pre-trained models while achieving a threefold improvement in training speed. Furthermore, we provide interpretable results using explainable AI techniques. We integrate denoising and deblurring, classification, and explainability in a single pipeline called DeClEx.

Keywords: machine learning, healthcare, classification, explainability

Procedia PDF Downloads 58

Authors: Anastasis Petrou

Abstract:

Black Swan Theories (BSTs) have existed since the 2nd Century BC. However, problematisation in the interdisciplinary field of Public Administration and Informatics (PA&I) about the impact of Black Swans as rare events in Society is a more recent phenomenon but with a growing, although dispersed, body of research literature. This paper offers a synopsis of core issues and questions raised in PA&I literature about the impacts of rare events in Society, the need for knowledge accumulation and explainability processes about rare events and asks what could help explain the occurrence, severity, heterogeneity, overall impact of Black Swans and the challenges they represent to established scientific methods. The second part of the paper considers how the use of Artificial Intelligence (AI) could assist researchers in better explaining rare events in PA&I. However, the research shows that whilst AI use at the start of knowledge accumulation and explainability processes about rare events is beneficial it is also fraught with challenges discussed herein. The paper concludes with recommendations for future research.

Keywords: black swans, public administration, AI, informatics

Procedia PDF Downloads 23

Authors: Paloma Mendes Saldanha

Abstract:

Access to information is a prerequisite for democracy while also guiding the material construction of fundamental rights. The exercise of citizenship requires knowing, understanding, questioning, advocating for, and securing rights and responsibilities. In other words, it goes beyond mere active electoral participation and materializes through awareness and the struggle for rights and responsibilities in the various spaces occupied by the population in their daily lives. In times of hyper-cultural connectivity, active citizenship is shaped through ethical trust processes, most often established between humans and algorithms. Automated decisions, so prevalent in various everyday situations, such as purchase preference predictions, virtual voice assistants, reduction of accidents in autonomous vehicles, content removal, resume selection, etc., have already found their place as a normalized discourse that sometimes does not reveal or make clear what violations of fundamental rights may occur when algorithmic explainability is lacking. In other words, technological and market development promotes a normalization for the use of automated decisions while silencing possible restrictions and/or breaches of rights through a culturally modeled, unethical, and unexplained trust process, which hinders the possibility of the right to a healthy, transparent, and complete exercise of citizenship. In this context, the article aims to identify the violations caused by the absence of algorithmic explainability in the exercise of citizenship through the construction of an unethical and silent trust process between humans and algorithms in automated decisions. As a result, it is expected to find violations of constitutionally protected rights such as privacy, data protection, and transparency, as well as the stipulation of algorithmic explainability as a fundamental right in the exercise of Brazilian citizenship in the era of virtualization, facing a threefold foundation called trust: culture, rules, and systems. To do so, the author will use a bibliographic review in the legal and information technology fields, as well as the analysis of legal and official documents, including national documents such as the Brazilian Federal Constitution, as well as international guidelines and resolutions that address the topic in a specific and necessary manner for appropriate regulation based on a sustainable trust process for a hyperconnected world.

Keywords: artificial intelligence, ethics, citizenship, trust

Procedia PDF Downloads 65

Authors: Belkacem Laimouche

Abstract:

With the field of artificial intelligence (AI) experiencing exponential growth, fueled by technological advancements that pave the way for increasingly innovative and promising applications, there is an escalating need to develop rigorous methods for assessing their performance in pursuit of transparency and equity. This article proposes a metrology-inspired statistical framework for evaluating bias and explainability in AI systems. Drawing from the principles of metrology, we propose a pioneering approach, using a concrete example, to evaluate the accuracy and precision of AI models, as well as to quantify the sources of measurement uncertainty that can lead to bias in their predictions. Furthermore, we explore a statistical approach for evaluating the explainability of AI systems based on their ability to provide interpretable and transparent explanations of their predictions.

Keywords: artificial intelligence, metrology, measurement uncertainty, prediction error, bias, machine learning algorithms, probabilistic models, interlaboratory comparison, data analysis, data reliability, measurement of bias impact on predictions, improvement of model accuracy and reliability

Procedia PDF Downloads 105

Authors: Kitty Kioskli, Theofanis Fotis, Nineta Polemi

Abstract:

The European Union (EU) directive Artificial Intelligence (AI) Act in Article 9 requires that risk management of AI systems includes both technical and human oversight, while according to NIST_AI_RFM (Appendix C) and ENISA AI Framework recommendations, claim that further research is needed to understand the current limitations of social threats and human-AI interaction. AI threats within social contexts significantly affect the security and trustworthiness of the AI systems; they are interrelated and trigger technical threats as well. For example, lack of explainability (e.g. the complexity of models can be challenging for stakeholders to grasp) leads to misunderstandings, biases, and erroneous decisions. Which in turn impact the privacy, security, accountability of the AI systems. Based on the NIST four fundamental criteria for explainability it can also classify the explainability threats into four (4) sub-categories: a) Lack of supporting evidence: AI systems must provide supporting evidence or reasons for all their outputs. b) Lack of Understandability: Explanations offered by systems should be comprehensible to individual users. c) Lack of Accuracy: The provided explanation should accurately represent the system's process of generating outputs. d) Out of scope: The system should only function within its designated conditions or when it possesses sufficient confidence in its outputs. Biases may also stem from historical data reflecting undesired behaviors. When present in the data, biases can permeate the models trained on them, thereby influencing the security and trustworthiness of the of AI systems. Social related AI threats are recognized by various initiatives (e.g., EU Ethics Guidelines for Trustworthy AI), standards (e.g. ISO/IEC TR 24368:2022 on AI ethical concerns, ISO/IEC AWI 42105 on guidance for human oversight of AI systems) and EU legislation (e.g. the General Data Protection Regulation 2016/679, the NIS 2 Directive 2022/2555, the Directive on the Resilience of Critical Entities 2022/2557, the EU AI Act, the Cyber Resilience Act). Measuring social threats, estimating the risks to AI systems associated to these threats and mitigating them is a research challenge. In this paper it will present the efforts of two European Commission Projects (FAITH and THEMIS) from the HorizonEurope programme that analyse the social threats by building cyber-social exercises in order to study human behaviour, traits, cognitive ability, personality, attitudes, interests, and other socio-technical profile characteristics. The research in these projects also include the development of measurements and scales (psychometrics) for human-related vulnerabilities that can be used in estimating more realistically the vulnerability severity, enhancing the CVSS4.0 measurement.

Keywords: social threats, artificial Intelligence, mitigation, social experiment

Procedia PDF Downloads 66

Authors: Michelangelo Sofo, Giuseppe Labianca

Abstract:

In recent years, the scientific community's interest in the exploratory analysis of biomedical data has increased exponentially. Considering the field of research of nutritional biologists, the curative process, based on the analysis of clinical data, is a very delicate operation due to the fact that there are multiple solutions for the management of pathologies in the food sector (for example can recall intolerances and allergies, management of cholesterol metabolism, diabetic pathologies, arterial hypertension, up to obesity and breathing and sleep problems). In this regard, in this research work a system was created capable of evaluating various dietary regimes for specific patient pathologies. The system is founded on a mathematical-numerical model and has been created tailored for the real working needs of an expert in human nutrition using the human-centered design (ISO 9241-210), therefore it is in step with continuous scientific progress in the field and evolves through the experience of managed clinical cases (machine learning process). DietAdhoc® is a decision support system nutrition specialists for patients of both sexes (from 18 years of age) developed with an agile methodology. Its task consists in drawing up the biomedical and clinical profile of the specific patient by applying two algorithmic optimization approaches on nutritional data and a symbolic solution, obtained by transforming the relational database underlying the system into a deductive database. For all three solution approaches, particular emphasis has been given to the explainability of the suggested clinical decisions through flexible and customizable user interfaces. Furthermore, the system has multiple software modules based on time series and visual analytics techniques that allow to evaluate the complete picture of the situation and the evolution of the diet assigned for specific pathologies.

Keywords: medical decision support, physiological data extraction, data driven diagnosis, human centered AI, symbiotic AI paradigm

Procedia PDF Downloads 26

Authors: Ola Hall, Ibrahim Wahab, Thorsteinn Rognvaldsson, Mattias Ohlsson

Abstract:

The subfield of poverty and welfare estimation that applies machine learning tools and methods on satellite imagery is a nascent but rapidly growing one. This is in part driven by the sustainable development goal, whose overarching principle is that no region is left behind. Among other things, this requires that welfare levels can be accurately and rapidly estimated at different spatial scales and resolutions. Conventional tools of household surveys and interviews do not suffice in this regard. While they are useful for gaining a longitudinal understanding of the welfare levels of populations, they do not offer adequate spatial coverage for the accuracy that is needed, nor are their implementation sufficiently swift to gain an accurate insight into people and places. It is this void that satellite imagery fills. Previously, this was near-impossible to implement due to the sheer volume of data that needed processing. Recent advances in machine learning, especially the deep learning subtype, such as deep neural networks, have made this a rapidly growing area of scholarship. Despite their unprecedented levels of performance, such models lack transparency and explainability and thus have seen limited downstream applications as humans generally are apprehensive of techniques that are not inherently interpretable and trustworthy. While several studies have demonstrated the superhuman performance of AI models, none has directly compared the performance of such models and human readers in the domain of poverty studies. In the present study, we directly compare the performance of human readers and a DL model using different resolutions of satellite imagery to estimate the welfare levels of demographic and health survey clusters in Tanzania, using the wealth quintile ratings from the same survey as the ground truth data. The cluster-level imagery covers all 608 cluster locations, of which 428 were classified as rural. The imagery for the human readers was sourced from the Google Maps Platform at an ultra-high resolution of 0.6m per pixel at zoom level 18, while that of the machine learning model was sourced from the comparatively lower resolution Sentinel-2 10m per pixel data for the same cluster locations. Rank correlation coefficients of between 0.31 and 0.32 achieved by the human readers were much lower when compared to those attained by the machine learning model – 0.69-0.79. This superhuman performance by the model is even more significant given that it was trained on the relatively lower 10-meter resolution satellite data while the human readers estimated welfare levels from the higher 0.6m spatial resolution data from which key markers of poverty and slums – roofing and road quality – are discernible. It is important to note, however, that the human readers did not receive any training before ratings, and had this been done, their performance might have improved. The stellar performance of the model also comes with the inevitable shortfall relating to limited transparency and explainability. The findings have significant implications for attaining the objective of the current frontier of deep learning models in this domain of scholarship – eXplainable Artificial Intelligence through a collaborative rather than a comparative framework.

Keywords: poverty prediction, satellite imagery, human readers, machine learning, Tanzania

Procedia PDF Downloads 107

Authors: Raphael Tuor, Denis Lalanne

Abstract:

The ATM research community is missing suitable tools to design, test, and validate new UI prototypes. Important stakes underline the implementation of both DSS and XAI methods into current systems. ML-based DSS are gaining in relevance as ATFM becomes increasingly complex. However, these systems only prove useful if a human can understand them, and thus new XAI methods are needed. The human-machine dyad should work as a team and should understand each other. We present xSky, a configurable benchmark tool that allows us to compare different versions of an ATC interface in conflict detection tasks. Our main contributions to the ATC research community are (1) a conflict detection task simulator (xSky) that allows to test the applicability of visual prototypes on scenarios of varying difficulty and outputting relevant operational metrics (2) a theoretical approach to the explanations of AI-driven trajectory predictions. xSky addresses several issues that were identified within available research tools. Researchers can configure the dimensions affecting scenario difficulty with a simple CSV file. Both the content and appearance of the XAI elements can be customized in a few steps. As a proof-of-concept, we implemented an XAI prototype inspired by the maritime field.

Keywords: air traffic control, air traffic simulation, conflict detection, explainable artificial intelligence, explainability, human-automation collaboration, human factors, information visualization, interpretability, trajectory prediction

Procedia PDF Downloads 160

Authors: Jaiden Xuan Schraut, Leon Liu, Yiqiao Yin

Abstract:

Computer vision in medical diagnosis has achieved a high level of success in diagnosing diseases with high accuracy. However, conventional classifiers that produce an image to-label result provides insufficient information for medical professionals to judge and raise concerns over the trust and reliability of a model with results that cannot be explained. In order to gain local insight into cancerous regions, separate tasks such as imaging segmentation need to be implemented to aid the doctors in treating patients, which doubles the training time and costs which renders the diagnosis system inefficient and difficult to be accepted by the public. To tackle this issue and drive AI-first medical solutions further, this paper proposes a multi-output network that follows a U-Net architecture for image segmentation output and features an additional convolutional neural networks (CNN) module for auxiliary classification output. Class activation maps are a method of providing insight into a convolutional neural network’s feature maps that leads to its classification but in the case of lung diseases, the region of interest is enhanced by U-net-assisted Class Activation Map (CAM) visualization. Therefore, our proposed model combines image segmentation models and classifiers to crop out only the lung region of a chest X-ray’s class activation map to provide a visualization that improves the explainability and is able to generate classification results simultaneously which builds trust for AI-led diagnosis systems. The proposed U-Net model achieves 97.61% accuracy and a dice coefficient of 0.97 on testing data from the COVID-QU-Ex Dataset which includes both diseased and healthy lungs.

Keywords: multi-output network model, U-net, class activation map, image classification, medical imaging analysis

Procedia PDF Downloads 205

Authors: Mpho Mokoatle, Darlington Mapiye, James Mashiyane, Stephanie Muller, Gciniwe Dlamini

Abstract:

Great advances in high-throughput sequencing technologies have resulted in availability of huge amounts of sequencing data in public and private repositories, enabling a holistic understanding of complex biological phenomena. Sequence data are used for a wide range of applications such as gene annotations, expression studies, personalized treatment and precision medicine. However, this rapid growth in sequence data poses a great challenge which calls for novel data processing and analytic methods, as well as huge computing resources. In this work, a machine and statistical learning approach for DNA sequence classification based on $k$-mer representation of sequence data is proposed. The approach is tested using whole genome sequences of Mycobacterium tuberculosis (MTB) isolates to (i) reduce the size of genomic sequence data, (ii) identify an optimum size of k-mers and utilize it to build classification models, (iii) predict the phenotype from whole genome sequence data of a given bacterial isolate, and (iv) demonstrate computing challenges associated with the analysis of whole genome sequence data in producing interpretable and explainable insights. The classification models were trained on 104 whole genome sequences of MTB isoloates. Cluster analysis showed that k-mers maybe used to discriminate phenotypes and the discrimination becomes more concise as the size of k-mers increase. The best performing classification model had a k-mer size of 10 (longest k-mer) an accuracy, recall, precision, specificity, and Matthews Correlation coeffient of 72.0%, 80.5%, 80.5%, 63.6%, and 0.4 respectively. This study provides a comprehensive approach for resampling whole genome sequencing data, objectively selecting a k-mer size, and performing classification for phenotype prediction. The analysis also highlights the importance of increasing the k-mer size to produce more biological explainable results, which brings to the fore the interplay that exists amongst accuracy, computing resources and explainability of classification results. However, the analysis provides a new way to elucidate genetic information from genomic data, and identify phenotype relationships which are important especially in explaining complex biological mechanisms.

Keywords: AWD-LSTM, bootstrapping, k-mers, next generation sequencing

Procedia PDF Downloads 169

Authors: Darlington Mapiye, Mpho Mokoatle, James Mashiyane, Stephanie Muller, Gciniwe Dlamini

Abstract:

Great advances in high-throughput sequencing technologies have resulted in availability of huge amounts of sequencing data in public and private repositories, enabling a holistic understanding of complex biological phenomena. Sequence data are used for a wide range of applications such as gene annotations, expression studies, personalized treatment and precision medicine. However, this rapid growth in sequence data poses a great challenge which calls for novel data processing and analytic methods, as well as huge computing resources. In this work, a machine and statistical learning approach for DNA sequence classification based on k-mer representation of sequence data is proposed. The approach is tested using whole genome sequences of Mycobacterium tuberculosis (MTB) isolates to (i) reduce the size of genomic sequence data, (ii) identify an optimum size of k-mers and utilize it to build classification models, (iii) predict the phenotype from whole genome sequence data of a given bacterial isolate, and (iv) demonstrate computing challenges associated with the analysis of whole genome sequence data in producing interpretable and explainable insights. The classification models were trained on 104 whole genome sequences of MTB isoloates. Cluster analysis showed that k-mers maybe used to discriminate phenotypes and the discrimination becomes more concise as the size of k-mers increase. The best performing classification model had a k-mer size of 10 (longest k-mer) an accuracy, recall, precision, specificity, and Matthews Correlation coeffient of 72.0 %, 80.5 %, 80.5 %, 63.6 %, and 0.4 respectively. This study provides a comprehensive approach for resampling whole genome sequencing data, objectively selecting a k-mer size, and performing classification for phenotype prediction. The analysis also highlights the importance of increasing the k-mer size to produce more biological explainable results, which brings to the fore the interplay that exists amongst accuracy, computing resources and explainability of classification results. However, the analysis provides a new way to elucidate genetic information from genomic data, and identify phenotype relationships which are important especially in explaining complex biological mechanisms

Keywords: AWD-LSTM, bootstrapping, k-mers, next generation sequencing

Procedia PDF Downloads 160

Authors: Michael Lognoul

Abstract:

In April 2021, the European Commission released its AI Act Proposal, which is the first policy proposal at the European Union level to target AI systems comprehensively, in a horizontal manner. This Proposal notably aims to achieve an ecosystem of trust in the European Union, based on the respect of fundamental rights, regarding AI. Among many other requirements, the AI Act Proposal aims to impose several generic transparency obligationson all AI systems to the benefit of natural persons facing those systems (e.g. information on the AI nature of systems, in case of an interaction with a human). The Proposal also provides for more stringent transparency obligations, specific to AI systems that qualify as high-risk, to the benefit of their users, notably on the characteristics, capabilities, and limitations of the AI systems they use. Against that background, this research firstly presents all such transparency requirements in turn, as well as related obligations, such asthe proposed obligations on record keeping. Secondly, it focuses on a legal analysis of their scope of application, of the content of the obligations, and on their practical implications. On the scope of transparency obligations tailored for high-risk AI systems, the research notably notes that it seems relatively narrow, given the proposed legal definition of the notion of users of AI systems. Hence, where end-users do not qualify as users, they may only receive very limited information. This element might potentially raise concern regarding the objective of the Proposal. On the content of the transparency obligations, the research highlights that the information that should benefit users of high-risk AI systems is both very broad and specific, from a technical perspective. Therefore, the information required under those obligations seems to create, prima facie, an adequate framework to ensure trust for users of high-risk AI systems. However, on the practical implications of these transparency obligations, the research notes that concern arises due to potential illiteracy of high-risk AI systems users. They might not benefit from sufficient technical expertise to fully understand the information provided to them, despite the wording of the Proposal, which requires that information should be comprehensible to its recipients (i.e. users).On this matter, the research points that there could be, more broadly, an important divergence between the level of detail of the information required by the Proposal and the level of expertise of users of high-risk AI systems. As a conclusion, the research provides policy recommendations to tackle (part of) the issues highlighted. It notably recommends to broaden the scope of transparency requirements for high-risk AI systems to encompass end-users. It also suggests that principles of explanation, as they were put forward in the Guidelines for Trustworthy AI of the High Level Expert Group, should be included in the Proposal in addition to transparency obligations.

Keywords: aI act proposal, explainability of aI, high-risk aI systems, transparency requirements

Procedia PDF Downloads 326

Authors: Amarendar Reddy Addula

Abstract:

Definition of Artificial Intelligence: Artificial intelligence is the simulation of mortal intelligence processes by machines, especially computer systems. Explicit operations of AI comprise expert systems, natural language processing, and speech recognition, and machine vision. Artificial Intelligence (AI) is an original medium for digital business, according to a new report by Gartner. The last 10 times represent an advance period in AI’s development, prodded by the confluence of factors, including the rise of big data, advancements in cipher structure, new machine literacy ways, the materialization of pall computing, and the vibrant open- source ecosystem. Influence of AI to a broader set of use cases and druggies and its gaining fashionability because it improves AI’s versatility, effectiveness, and rigidity. Edge AI will enable digital moments by employing AI for real- time analytics closer to data sources. Gartner predicts that by 2025, further than 50 of all data analysis by deep neural networks will do at the edge, over from lower than 10 in 2021. Responsible AI is a marquee term for making suitable business and ethical choices when espousing AI. It requires considering business and societal value, threat, trust, translucency, fairness, bias mitigation, explainability, responsibility, safety, sequestration, and nonsupervisory compliance. Responsible AI is ever more significant amidst growing nonsupervisory oversight, consumer prospects, and rising sustainability pretensions. Generative AI is the use of AI to induce new vestiges and produce innovative products. To date, generative AI sweats have concentrated on creating media content similar as photorealistic images of people and effects, but it can also be used for law generation, creating synthetic irregular data, and designing medicinals and accoutrements with specific parcels. AI is the subject of a wide- ranging debate in which there's a growing concern about its ethical and legal aspects. Constantly, the two are varied and nonplussed despite being different issues and areas of knowledge. The ethical debate raises two main problems the first, abstract, relates to the idea and content of ethics; the alternate, functional, and concerns its relationship with the law. Both set up models of social geste, but they're different in compass and nature. The juridical analysis is grounded on anon-formalistic scientific methodology. This means that it's essential to consider the nature and characteristics of the AI as a primary step to the description of its legal paradigm. In this regard, there are two main issues the relationship between artificial and mortal intelligence and the question of the unitary or different nature of the AI. From that theoretical and practical base, the study of the legal system is carried out by examining its foundations, the governance model, and the nonsupervisory bases. According to this analysis, throughout the work and in the conclusions, International Law is linked as the top legal frame for the regulation of AI.

Keywords: artificial intelligence, ethics & human rights issues, laws, international laws

Procedia PDF Downloads 96

Authors: Sylvie Michel, Sylvie Gerbaix, Marc Bidan

Abstract:

Artificial intelligence (AI) can be held accountable for its detrimental impacts. This question gains heightened relevance given AI's pervasive reach across various domains, magnifying its power and potential. The expanding influence of AI raises fundamental ethical inquiries, primarily centering on biases, responsibility, and transparency. This encompasses discriminatory biases arising from algorithmic criteria or data, accidents attributed to autonomous vehicles or other systems, and the imperative of transparent decision-making. This article aims to stimulate reflection on AI accountability, denoting the necessity to elucidate the effects it generates. Accountability comprises two integral aspects: adherence to legal and ethical standards and the imperative to elucidate the underlying operational rationale. The objective is to initiate a reflection on the obstacles to this "accountability," facing the challenges of the complexity of artificial intelligence's system and its effects. Then, this article proposes to mobilize Edgar Morin's complex thought to encompass and face the challenges of this complexity. The first contribution is to point out the challenges posed by the complexity of A.I., with fractional accountability between a myriad of human and non-human actors, such as software and equipment, which ultimately contribute to the decisions taken and are multiplied in the case of AI. Accountability faces three challenges resulting from the complexity of the ethical issues combined with the complexity of AI. The challenge of the non-neutrality of algorithmic systems as fully ethically non-neutral actors is put forward by a revealing ethics approach that calls for assigning responsibilities to these systems. The challenge of the dilution of responsibility is induced by the multiplicity and distancing between the actors. Thus, a dilution of responsibility is induced by a split in decision-making between developers, who feel they fulfill their duty by strictly respecting the requests they receive, and management, which does not consider itself responsible for technology-related flaws. Accountability is confronted with the challenge of transparency of complex and scalable algorithmic systems, non-human actors self-learning via big data. A second contribution involves leveraging E. Morin's principles, providing a framework to grasp the multifaceted ethical dilemmas and subsequently paving the way for establishing accountability in AI. When addressing the ethical challenge of biases, the "hologrammatic" principle underscores the imperative of acknowledging the non-ethical neutrality of algorithmic systems inherently imbued with the values and biases of their creators and society. The "dialogic" principle advocates for the responsible consideration of ethical dilemmas, encouraging the integration of complementary and contradictory elements in solutions from the very inception of the design phase. Aligning with the principle of organizing recursiveness, akin to the "transparency" of the system, it promotes a systemic analysis to account for the induced effects and guides the incorporation of modifications into the system to rectify deviations and reintroduce modifications into the system to rectify its drifts. In conclusion, this contribution serves as an inception for contemplating the accountability of "artificial intelligence" systems despite the evident ethical implications and potential deviations. Edgar Morin's principles, providing a lens to contemplate this complexity, offer valuable perspectives to address these challenges concerning accountability.

Keywords: accountability, artificial intelligence, complexity, ethics, explainability, transparency, Edgar Morin

Procedia PDF Downloads 63

Authors: Diogo José Morgado Rebelo, Francisco António Carneiro Pacheco de Andrade, Paulo Jorge Freitas de Oliveira Novais

Abstract:

Not everything in AI-power consumer credit scoring turns out to be a wonder. When using AI in Creditworthiness Assessment (CWA), opacity and unfairness ‘sins’ must be considered to the task be deemed Responsible. AI software is not always 100% accurate, which can lead to misclassification. Discrimination of some groups can be exponentiated. A hetero personalized identity can be imposed on the individual(s) affected. Also, autonomous CWA sometimes lacks transparency when using black box models. However, for this intended purpose, human analysts ‘on-the-loop’ might not be the best remedy consumers are looking for in credit. This study seeks to explore the legality of implementing a Multi-Agent System (MAS) framework in consumer CWA to ensure compliance with the regulation outlined in Article 14(4) of the Proposal for an Artificial Intelligence Act (AIA), dated 21 April 2021 (as per the last corrigendum by the European Parliament on 19 April 2024), Especially with the adoption of Art. 18(8)(9) of the EU Directive 2023/2225, of 18 October, which will go into effect on 20 November 2026, there should be more emphasis on the need for hybrid oversight in AI-driven scoring to ensure fairness and transparency. In fact, the range of EU regulations on AI-based consumer credit will soon impact the AI lending industry locally and globally, as shown by the broad territorial scope of AIA’s Art. 2. Consequently, engineering the law of consumer’s CWA is imperative. Generally, the proposed MAS framework consists of several layers arranged in a specific sequence, as follows: firstly, the Data Layer gathers legitimate predictor sets from traditional sources; then, the Decision Support System Layer, whose Neural Network model is trained using k-fold Cross Validation, provides recommendations based on the feeder data; the eXplainability (XAI) multi-structure comprises Three-Step-Agents; and, lastly, the Oversight Layer has a 'Bottom Stop' for analysts to intervene in a timely manner. From the analysis, one can assure a vital component of this software is the XAY layer. It appears as a transparent curtain covering the AI’s decision-making process, enabling comprehension, reflection, and further feasible oversight. Local Interpretable Model-agnostic Explanations (LIME) might act as a pillar by offering counterfactual insights. SHapley Additive exPlanation (SHAP), another agent in the XAI layer, could address potential discrimination issues, identifying the contribution of each feature to the prediction. Alternatively, for thin or no file consumers, the Suggestion Agent can promote financial inclusion. It uses lawful alternative sources such as the share of wallet, among others, to search for more advantageous solutions to incomplete evaluation appraisals based on genetic programming. Overall, this research aspires to bring the concept of Machine-Centered Anthropocentrism to the table of EU policymaking. It acknowledges that, when put into service, credit analysts no longer exert full control over the data-driven entities programmers have given ‘birth’ to. With similar explanatory agents under supervision, AI itself can become self-accountable, prioritizing human concerns and values. AI decisions should not be vilified inherently. The issue lies in how they are integrated into decision-making and whether they align with non-discrimination principles and transparency rules.

Keywords: creditworthiness assessment, hybrid oversight, machine-centered anthropocentrism, EU policymaking

Procedia PDF Downloads 36

Authors: Faruk Ozdemir, Roy Kalawsky, Peter Hubbard

Abstract:

Accurate estimation of RUL provides a basis for effective predictive maintenance, reducing unexpected downtime for industrial equipment. However, while models such as the Random Forest have effective predictive capabilities, they are the so-called ‘black box’ models, where interpretability is at a threshold to make critical diagnostic decisions involved in industries related to aviation. The purpose of this work is to present a prognostic framework that embeds Explainable Artificial Intelligence (XAI) techniques in order to provide essential transparency in Machine Learning methods' decision-making mechanisms based on sensor data, with the objective of procuring actionable insights for the aviation industry. Sensor readings have been gathered from critical equipment such as turbofan jet engine and landing gear, and the prediction of the RUL is done by a Random Forest model. It involves steps such as data gathering, feature engineering, model training, and evaluation. These critical components’ datasets are independently trained and evaluated by the models. While suitable predictions are served, their performance metrics are reasonably good; such complex models, however obscure reasoning for the predictions made by them and may even undermine the confidence of the decision-maker or the maintenance teams. This is followed by global explanations using SHAP and local explanations using LIME in the second phase to bridge the gap in reliability within industrial contexts. These tools analyze model decisions, highlighting feature importance and explaining how each input variable affects the output. This dual approach offers a general comprehension of the overall model behavior and detailed insight into specific predictions. The proposed framework, in its third component, incorporates the techniques of causal analysis in the form of Granger causality tests in order to move beyond correlation toward causation. This will not only allow the model to predict failures but also present reasons, from the key sensor features linked to possible failure mechanisms to relevant personnel. The causality between sensor behaviors and equipment failures creates much value for maintenance teams due to better root cause identification and effective preventive measures. This step contributes to the system being more explainable. Surrogate Several simple models, including Decision Trees and Linear Models, can be used in yet another stage to approximately represent the complex Random Forest model. These simpler models act as backups, replicating important jobs of the original model's behavior. If the feature explanations obtained from the surrogate model are cross-validated with the primary model, the insights derived would be more reliable and provide an intuitive sense of how the input variables affect the predictions. We then create an iterative explainable feedback loop, where the knowledge learned from the explainability methods feeds back into the training of the models. This feeds into a cycle of continuous improvement both in model accuracy and interpretability over time. By systematically integrating new findings, the model is expected to adapt to changed conditions and further develop its prognosis capability. These components are then presented to the decision-makers through the development of a fully transparent condition monitoring and alert system. The system provides a holistic tool for maintenance operations by leveraging RUL predictions, feature importance scores, persistent sensor threshold values, and autonomous alert mechanisms. Since the system will provide explanations for the predictions given, along with active alerts, the maintenance personnel can make informed decisions on their end regarding correct interventions to extend the life of the critical machinery.

Keywords: predictive maintenance, explainable artificial intelligence, prognostic, RUL, machine learning, turbofan engines, C-MAPSS dataset

Procedia PDF Downloads 9