Large-scale Artificial Intelligence Research Articles

Sacred Meets Synthetic: A Multi-Method Study on the First AI Church Service

Artificial Intelligence (AI) is transforming society, including religious practices and experiences of religious communities. The first large-scale AI church service took place in 2023 in Germany as a Protestant Christian worship event within a biannual religious celebration. Studying this religious service illustrates how AI was developed as a religious agent and provides insights into the experiences of attendees. This research note presents data from a quantitative survey and two qualitative questionnaires of participants conducted at the AI worship service. Findings show diversity of opinion and a range of attendee experiences. By presenting the results of qualitative and quantitative analysis, this research note highlights the possibilities and limitations of incorporating AI into the religious sphere. While most of the responses were skeptical, participants also reported having spiritual experiences. Nevertheless, attendees also reported conflicting feelings regarding experiences of community within an artificial setting. The main concerns were technological limitations, fear of replacing humans, biases in the theology of the underlying large language model, and lack of personality and emotion. Age-based differences include younger attendees finding the AI service more attractive while older attendees found the AI service more stimulating. Implications are drawn for practical theology and AI implementation within religious settings.

When AIs become oracles: generative artificial intelligence, anticipatory urban governance, and the future of cities

Abstract Generative Artificial Intelligence (AI) is boosting anticipatory forms of governance, through which state actors seek to predict the future and strategically intervene in the present. In this context, city brains represent an emerging type of generative AI currently employed in urban governance and public policy in a growing number of cities. City brains are large-scale AIs residing in vast digital urban platforms, which manage multiple urban domains including transport, safety, health, and environmental monitoring. They use Large Language Models (LLMs) to generate visions of urban futures: visions that are in turn used by policymakers to generate new urban policies. In this paper, we advance a twofold contribution. Theoretically, we develop a critical theory of anticipatory governance in the age of generative AI. More specifically, we focus on technocratic approaches to anticipatory governance, to explain how the act of governing extends into the future by means of predictive AI technology. Our approach is critical in order to expose the dangers that the use of AI (generative AI, in particular) in urban governance poses, and to identify their causes. These dangers include the formation of a policy process that, under the influence of unintelligible LLMs, risks losing transparency and thus accountability, and the marginalization of human stakeholders (citizens, in particular) as the role of AI in the management of cities keeps growing and governance begins to turn posthuman. Empirically, we critically examine an existing city brain project under development in China and ground our critical theory in a real-life example.

Evaluation of the quality and quantity of artificial intelligence-generated responses about anesthesia and surgery: using ChatGPT 3.5 and 4.0.

The large-scale artificial intelligence (AI) language model chatbot, Chat Generative Pre-Trained Transformer (ChatGPT), is renowned for its ability to provide data quickly and efficiently. This study aimed to assess the medical responses of ChatGPT regarding anesthetic procedures. Two anesthesiologist authors selected 30 questions representing inquiries patients might have about surgery and anesthesia. These questions were inputted into two versions of ChatGPT in English. A total of 31 anesthesiologists then evaluated each response for quality, quantity, and overall assessment, using 5-point Likert scales. Descriptive statistics summarized the scores, and a paired sample t-test compared ChatGPT 3.5 and 4.0. Regarding quality, "appropriate" was the most common rating for both ChatGPT 3.5 and 4.0 (40 and 48%, respectively). For quantity, responses were deemed "insufficient" in 59% of cases for 3.5, and "adequate" in 69% for 4.0. In overall assessment, 3 points were most common for 3.5 (36%), while 4 points were predominant for 4.0 (42%). Mean quality scores were 3.40 and 3.73, and mean quantity scores were - 0.31 (between insufficient and adequate) and 0.03 (between adequate and excessive), respectively. The mean overall score was 3.21 for 3.5 and 3.67 for 4.0. Responses from 4.0 showed statistically significant improvement in three areas. ChatGPT generated responses mostly ranging from appropriate to slightly insufficient, providing an overall average amount of information. Version 4.0 outperformed 3.5, and further research is warranted to investigate the potential utility of AI chatbots in assisting patients with medical information.

A Top-Down Design Approach for Generating a Peptide PROTAC Drug Targeting Androgen Receptor for Androgenetic Alopecia Therapy.

While large-scale artificial intelligence (AI) models for protein structure prediction and design are advancing rapidly, the translation of deep learning models for practical macromolecular drug development remains limited. This investigation aims to bridge this gap by combining cutting-edge methodologies to create a novel peptide-based PROTAC drug development paradigm. Using ProteinMPNN and RFdiffusion, we identified binding peptides for androgen receptor (AR) and Von Hippel-Lindau (VHL), followed by computational modeling with Alphafold2-multimer and ZDOCK to predict spatial interrelationships. Experimental validation confirmed the designed peptide's binding ability to AR and VHL. Transdermal microneedle patching technology was seamlessly integrated for the peptide PROTAC drug delivery in androgenic alopecia treatment. In summary, our approach provides a generic method for generating peptide PROTACs and offers a practical application for designing potential therapeutic drugs for androgenetic alopecia. This showcases the potential of interdisciplinary approaches in advancing drug development and personalized medicine.

SFL-TUM: Energy efficient SFRL method for large scale AI model's task offloading in UAV-assisted MEC networks

The convergence of mobile edge computing (MEC) network with unmanned aerial vehicles (UAVs) presents an auspicious opportunity to revolutionize wireless communication and facilitate high-speed internet access in remote regions for mobile devices (MDs) as well as large scale artificial intelligence (AI) models. However, the substantial amount of data produced by the UAVs-assisted MEC network necessitates the integration of efficient distributed learning techniques in AI models. In recent times, distributed learning algorithms, including federated reinforcement learning (FRL) and split learning (SL), have been explored for the purpose of learning machine learning (ML) models that are distributed by sharing model parameters, as opposed to large raw data-sets as seen in traditional centralized learning algorithms. To implement the hybrid method, the model is first trained locally on each UAV-assisted MEC network using SL. Subsequently, the model parameters that have been encrypted are sent to a central server for federated averaging. Finally, after the model has been updated, it is distributed to each UAV-assisted MEC network for local fine-tuning. Our simulations indicate that the proposed split and federated reinforcement learning (SFRL) framework yields comparable high-test accuracy performance while consuming less energy compared to extant distributed learning algorithms. Furthermore, the SFRL algorithm efficiently realizes energy-efficient selection between the SL and FRL methods under different distributions. Numerical results shows that the proposed scheme improves the accuracy by 29.31% and reduced the energy consumption by around 67.34% and time delay by about 7.37%. as compared to the existing baseline schemes.

Exploring Types of Photonic Neural Networks for Imaging and Computing-A Review.

Photonic neural networks (PNNs), utilizing light-based technologies, show immense potential in artificial intelligence (AI) and computing. Compared to traditional electronic neural networks, they offer faster processing speeds, lower energy usage, and improved parallelism. Leveraging light's properties for information processing could revolutionize diverse applications, including complex calculations and advanced machine learning (ML). Furthermore, these networks could address scalability and efficiency challenges in large-scale AI systems, potentially reshaping the future of computing and AI research. In this comprehensive review, we provide current, cutting-edge insights into diverse types of PNNs crafted for both imaging and computing purposes. Additionally, we delve into the intricate challenges they encounter during implementation, while also illuminating the promising perspectives they introduce to the field.

국내외 인공지능 반도체에 대한 연구 동향

Recently, large-scale artificial intelligence (AI) such as ChatGPT have been developed, and as AI is used across various industrial fields, attention is focused on AI chips (semiconductors). AI chips refer to chips designed for calculations for AI algorithms, and many companies at domestic and abroad, such as NVIDIA, Tesla, and ETRI, are developing AI chips. In this paper, we survey research trends on nine types of AI chips. Currently, many attempts have been made to improve the computational performance of most AI chips, and semiconductors for specific purposes are also being designed. In order to compare various AI semiconductors, each chip is analyzed in terms of operation unit, speed, power, and energy efficiency. We introduce currently existing optimization methodologies for AI computation. Based on this, future research directions for AI semiconductors are presented in this paper.

Spatiotemporal correlation based self-adaptive pose estimation in complex scenes

In the current digital era, large-scale Artificial Intelligence (AI) models have fundamentally changed the landscape of the AI field. These models excel at extracting complex patterns and information from massive datasets, making them indispensable in a wide range of domains. With the advancement of large-scale AI models, pose estimation has achieved improved performance in managing large data sets and complex computations. However, most existing pose estimation methods suffer from the problem of inaccurate pose graph extraction in complex scenes. Moreover, the existing methods all focus on the message, ignoring the fact that it may not be optimal for the graph structure. This fixed graph structure is difficult to reflect the real dependency relationship between joint points. This study proposes a Character Redetection Model (CRM) to dynamically adjust the foreground extraction strategies to solve these problems. The CRM locates the target position based on the rich spatiotemporal information between the adjacent frames, effectively solving the problem of inaccurate foreground acquisition caused by occlusion and blur in complex scenes. In addition, an Uncertainty Graph Structure (UGS) is designed, which adaptively learns the topological structure of graphs with different Graph Convolutional Network (GCN) layers and skeleton samples in an end-to-end manner and mines the hidden space dependency between data to accurately model the different poses of the characters. The experimental results show that the proposed method achieves a superior performance compared to that of the State-Of-The-Art (SOTA) method, with a Percentage of Correct Keypoints (PCK) value of 94.6% at the Max Planck Institute for Informatics (MPII), along with the mean Average Precision (mAP) values of 83.3% and 82.1% on the PoseTrack2017 and classroom pose datasets, respectively.

Artificial Intelligence Alone Will Not Democratise Education: On Educational Inequality, Techno-Solutionism and Inclusive Tools

Artificial Intelligence (AI) in Education claims to have the potential for building personalised curricula, as well as bringing opportunities for democratising education and creating a renaissance of new ways of teaching and learning. Millions of students are starting to benefit from the use of these technologies, but millions more around the world are not, due to the digital divide and deep pre-existing social and educational inequalities. If this trend continues, the first large-scale delivery of AI in Education could lead to greater educational inequality, along with a global misallocation of educational resources motivated by the current techno-solutionist narrative, which proposes technological solutions as a quick and flawless way to solve complex real-world problems. This work focuses on posing questions about the future of AI in Education, intending to initiate the pressing conversation that could set the right foundations (e.g., inclusion and diversity) for a new generation of education that is permeated with AI technology. The main goal of our opinion piece is to conceptualise a sustainable, large-scale and inclusive AI for the education ecosystem that facilitates equitable, high-quality lifelong learning opportunities for all. The contribution starts by synthesising how AI might change how we learn and teach, focusing on the case of personalised learning companions and assistive technology for disability. Then, we move on to discuss some socio-technical features that will be crucial to avoiding the perils of these AI systems worldwide (and perhaps ensuring their success by leveraging more inclusive education). This work also discusses the potential of using AI together with free, participatory and democratic resources, such as Wikipedia, Open Educational Resources and open-source tools. We emphasise the need for collectively designing human-centred, transparent, interactive and collaborative AI-based algorithms that empower and give complete agency to stakeholders, as well as supporting new emerging pedagogies. Finally, we ask what it would take for this educational revolution to provide egalitarian and empowering access to education that transcends any political, cultural, language, geographical and learning-ability barriers, so that educational systems can be responsive to all learners’ needs.

Validation of a Deep Learning Chest X-ray Interpretation Model: Integrating Large-Scale AI and Large Language Models for Comparative Analysis with ChatGPT.

This study evaluates the diagnostic accuracy and clinical utility of two artificial intelligence (AI) techniques: Kakao Brain Artificial Neural Network for Chest X-ray Reading (KARA-CXR), an assistive technology developed using large-scale AI and large language models (LLMs), and ChatGPT, a well-known LLM. The study was conducted to validate the performance of the two technologies in chest X-ray reading and explore their potential applications in the medical imaging diagnosis domain. The study methodology consisted of randomly selecting 2000 chest X-ray images from a single institution's patient database, and two radiologists evaluated the readings provided by KARA-CXR and ChatGPT. The study used five qualitative factors to evaluate the readings generated by each model: accuracy, false findings, location inaccuracies, count inaccuracies, and hallucinations. Statistical analysis showed that KARA-CXR achieved significantly higher diagnostic accuracy compared to ChatGPT. In the 'Acceptable' accuracy category, KARA-CXR was rated at 70.50% and 68.00% by two observers, while ChatGPT achieved 40.50% and 47.00%. Interobserver agreement was moderate for both systems, with KARA at 0.74 and GPT4 at 0.73. For 'False Findings', KARA-CXR scored 68.00% and 68.50%, while ChatGPT scored 37.00% for both observers, with high interobserver agreements of 0.96 for KARA and 0.97 for GPT4. In 'Location Inaccuracy' and 'Hallucinations', KARA-CXR outperformed ChatGPT with significant margins. KARA-CXR demonstrated a non-hallucination rate of 75%, which is significantly higher than ChatGPT's 38%. The interobserver agreement was high for KARA (0.91) and moderate to high for GPT4 (0.85) in the hallucination category. In conclusion, this study demonstrates the potential of AI and large-scale language models in medical imaging and diagnostics. It also shows that in the chest X-ray domain, KARA-CXR has relatively higher accuracy than ChatGPT.

Self-Organizing Democratized Learning: Toward Large-Scale Distributed Learning Systems.

Emerging cross-device artificial intelligence (AI) applications require a transition from conventional centralized learning systems toward large-scale distributed AI systems that can collaboratively perform complex learning tasks. In this regard, democratized learning (Dem-AI) lays out a holistic philosophy with underlying principles for building large-scale distributed and democratized machine learning systems. The outlined principles are meant to study a generalization in distributed learning systems that go beyond existing mechanisms such as federated learning (FL). Moreover, such learning systems rely on hierarchical self-organization of well-connected distributed learning agents who have limited and highly personalized data and can evolve and regulate themselves based on the underlying duality of specialized and generalized processes. Inspired by Dem-AI philosophy, a novel distributed learning approach is proposed in this article. The approach consists of a self-organizing hierarchical structuring mechanism based on agglomerative clustering, hierarchical generalization, and corresponding learning mechanism. Subsequently, hierarchical generalized learning problems in recursive forms are formulated and shown to be approximately solved using the solutions of distributed personalized learning problems and hierarchical update mechanisms. To that end, a distributed learning algorithm, namely DemLearn, is proposed. Extensive experiments on benchmark MNIST, Fashion-MNIST, FE-MNIST, and CIFAR-10 datasets show that the proposed algorithm demonstrates better results in the generalization performance of learning models in agents compared to the conventional FL algorithms. The detailed analysis provides useful observations to further handle both the generalization and specialization performance of the learning models in Dem-AI systems.

Developing a Word Lexicon from Electronic Health Records for Natural Language Processing Analysis of Free-Text Reports for Patients with Venous Thromboembolism.

Patients with gliomas, particularly glioblastomas (GBM) have a higher risk of developing venous thromboembolism (VTE), including both pulmonary embolism (PE) or deep vein thrombosis (DVT), both correlating with overall survival (OS) and potential biological signaling relevance for analysis of the tumor state in conjunction with evolving biomarkers. Artificial intelligence (AI) approaches thar employ VTE as a clinical feature in brain tumor patients is understudied due to the difficulty in analyzing unstructured clinical data in electronic health records (EHR). Data expansion by creating a word lexicon for natural language processing (NLP) of free-text clinical reports will allow exposure of VTE for classification of large-scale data sets, NLP and AI. Patients with a pathologic diagnosis of GBM (2005-2021) were extracted from EHR and screened for the development of VTE based on radiology free-text reports - ultrasound (US) of extremities and Computed Topography-pulmonary angiogram (CT). Language data were collected manually and verified computationally using radiology report text filtration. Common phrases and words employed were collected to generate a lexicon for future computational analysis. Kaplan-Meier survival analyses were generated for VTE in relation OS and progression-free survival (PFS). A total of 163 patients (mean age = 56.1 ± 12.1, 65% male) were included, 48 (29.4%) were screened for VTE following clinical suspicion on history or physical exam, and 15 (9.2%) were found to have a VTE. Screening methods were US 83.3% (40) or CT 13.9% (6), or both 4.6% (2). 28.6% (12) of US and 37.5% (3) of CT resulted in a positive VTE diagnosis. Terms that were only detected in patients with VTE were for US: "complete" 44%, "thrombosis detected" (8.3%), "occlusive" (16%), "partial" (16%), "critical" (25%), "residual" (8.3%) and for CT: "critical" (67%) and "pulmonary emboli" (33%). Words used in free-text reports showed a high degree of inter-reporter variability. The words "partial", "residual", "complete", "critical" or "clotted" when used as an "or" Boolean statement applied to US and CT radiology reports identified ∼93% of the patients with VTE. Patients with VTE had worse OS (median 14 vs. 19 months, p = .0189) and PFS (median 6 vs. 9 months, p = .0239) compared to patients without VTE indicating underlying pathology associated with both prevalence of VTE and tumor burden and the need to incorporate VTE as a clinical feature in large scale data analyses. US and CT yield a similar percentage of positive VTE findings while employing different terms to characterize VTE. We confirm that patients with VTE have poorer outcomes and present a word combination that identifies patients with VTE in large-scale radiology report data. Further research will focus on validation in other datasets, NLP approaches to optimize and data aggregation with laboratory data and large-scale omics panels to expose thrombosis as a clinical feature for biomarker analysis.

CLRM-04 DEVELOPING A WORD LEXICON FOR VENOUS THROMBOEMBOLISM DATA ANNOTATION USING RADIOLOGY REPORTS IN GLIOBLASTOMA PATIENTS - LINKING LARGE-SCALE DATA SETS TO EXPAND CLINICAL FEATURES FOR AI

Abstract Patients with gliomas, particularly glioblastomas (GBM), have a higher risk of developing venous thromboembolism (VTE), correlating with overall survival (OS). Artificial intelligence (AI) approaches that employ VTE as a clinical feature in brain tumor patients is understudied due to the difficulty in analyzing electronic health records (EHR). Data expansion by creating a word lexicon for natural language processing (NLP) of free-text clinical reports will allow exposure of VTE for the classification of large-scale data sets, NLP, and AI. Patients with a pathologic diagnosis of GBM (2005-2021) were screened for the development of VTE based on radiology free-text reports (ultrasound (US) of extremities and Computed Topography-pulmonary angiogram (CT)). Kaplan-Meier survival analyses about overall survival (OS) and progression-free survival (PFS) were generated for VTE. 163 patients (mean age = 56.1 ± 12.1, 65% male) were included, 48 (29.4%) were screened for VTE following clinical suspicion on history or physical exam, and 15 (9.2%) were found to have a VTE. Screening methods were ultrasound 83.3% (40) or CT 13.9% (6), or both 4.6% (2). 28.6% (12) of US and 37.5% (3) of CT resulted in a positive VTE diagnosis. The words “partial”, “residual”, “complete”, “critical” or “clotted” when used as an “or” boolean statement applied to US and CT radiology reports identified ~93% of the patients with VTE. Patients with VTE had worse OS (median 14 vs. 19 months, p = .0189) and PFS (median 6 vs. 9 months, p = .0239) than patients without VTE, indicating underlying pathology associated with both prevalence of VTE and tumor burden. US and CT yield a similar percentage of positive VTE findings while employing different terms to characterize VTE. We confirm that patients with VTE have poorer outcomes and present a word combination that identifies patients with VTE in large-scale radiology report data.

Data infrastructures for AI in medical imaging: a report on the experiences of five EU projects

Artificial intelligence (AI) is transforming the field of medical imaging and has the potential to bring medicine from the era of ‘sick-care’ to the era of healthcare and prevention. The development of AI requires access to large, complete, and harmonized real-world datasets, representative of the population, and disease diversity. However, to date, efforts are fragmented, based on single–institution, size-limited, and annotation-limited datasets. Available public datasets (e.g., The Cancer Imaging Archive, TCIA, USA) are limited in scope, making model generalizability really difficult. In this direction, five European Union projects are currently working on the development of big data infrastructures that will enable European, ethically and General Data Protection Regulation-compliant, quality-controlled, cancer-related, medical imaging platforms, in which both large-scale data and AI algorithms will coexist. The vision is to create sustainable AI cloud-based platforms for the development, implementation, verification, and validation of trustable, usable, and reliable AI models for addressing specific unmet needs regarding cancer care provision. In this paper, we present an overview of the development efforts highlighting challenges and approaches selected providing valuable feedback to future attempts in the area.Key points• Artificial intelligence models for health imaging require access to large amounts of harmonized imaging data and metadata.• Main infrastructures adopted either collect centrally anonymized data or enable access to pseudonymized distributed data.• Developing a common data model for storing all relevant information is a challenge.• Trust of data providers in data sharing initiatives is essential.• An online European Union meta-tool-repository is a necessity minimizing effort duplication for the various projects in the area.

Exploring developments of the AI field from the perspective of methods, datasets, and metrics

The knowledge contained in academic literature is interesting to mine. Inspired by the idea of molecular markers tracing in the field of biochemistry, three named entities, namely, methods, datasets, and metrics, are extracted and used as artificial intelligence (AI) markers for AI literature. These entities can be used to trace the research process described in the bodies of papers, which opens up new perspectives for seeking and mining more valuable academic information. Firstly, the named entity recognition model is used to extract AI markers from large-scale AI literature. A multi-stage self-paced learning strategy (MSPL) is proposed to address the negative influence of hard and noisy samples on the model training. Secondly, original papers are traced for AI markers. Statistical and propagation analyses are performed based on the tracing results. Finally, the co-occurrences of AI markers are used to achieve clustering. The evolution within method clusters is explored. The above-mentioned mining based on AI markers yields many significant findings. For example, the propagation rate of the datasets gradually increases. The methods proposed by China in recent years have an increasing influence on other countries.

Harnessing AI for business development: a review of drivers and challenges in Africa

An accumulating body of research has demonstrated that artificial intelligence (AI) is an indistinguishable feature of the fourth industrial revolution. This study integrates the literature on AI and new technologies to examine the constraining and facilitating forces for developing and scaling-up AI-enabling technologies in Africa. This article proposes an integrated conceptual model to elucidate the range of external drivers encompassing global competitive drivers, and market and industry drivers. The internal drivers include the potential to enhance product development speed, improve quality, drive production cost down, and minimise errors and manual processes in organisations. The barriers identified include institutional dysfunction, poor infrastructure to support development of large-scale AI, lack of skilled AI experts, and lack of access to latest technologies in the regional population. The public policy implications for Africa as the potential next frontier market for AI development are discussed alongside contributions to theory.

Characteristic analysis of volatile avalanche diode threshold switching for bionic nerve synapse applications

The combination of biological neurology and memristive theory has greatly promoted the development of neuromorphic computing. To build a large-scale artificial intelligence alert system, the exploration of bionic synapses compatible with standard processes has become an urgent problem to be solved in the next step. In response to the above application requirements, this paper proposes a volatile avalanche diode threshold switching (VADTS) that is fully compatible with standard semiconductor technology to simulate the various functions of the synapse. Technology computer-aided design device-level simulation can verify the bionic principle of VADTS. The function of VADTS's bionic synapse was verified by the experimental test platform. The results show that under the action of the excitation signal (11.25 V), the device can continuously change from a high-resistance state to a low-resistance state. When the excitation signal is lower than the threshold, VADTS presents a “no adaptation” state of nerve synapses. When the excitation signal is higher than the threshold and changes continuously, the current changes along with the amplitude of the excitation signal, similar to the “sensitization” state of the nerve synapse.

PCJ Java library as a solution to integrate HPC, Big Data and Artificial Intelligence workloads

With the development of peta- and exascale size computational systems there is growing interest in running Big Data and Artificial Intelligence (AI) applications on them. Big Data and AI applications are implemented in Java, Scala, Python and other languages that are not widely used in High-Performance Computing (HPC) which is still dominated by C and Fortran. Moreover, they are based on dedicated environments such as Hadoop or Spark which are difficult to integrate with the traditional HPC management systems. We have developed the Parallel Computing in Java (PCJ) library, a tool for scalable high-performance computing and Big Data processing in Java. In this paper, we present the basic functionality of the PCJ library with examples of highly scalable applications running on the large resources. The performance results are presented for different classes of applications including traditional computational intensive (HPC) workloads (e.g. stencil), as well as communication-intensive algorithms such as Fast Fourier Transform (FFT). We present implementation details and performance results for Big Data type processing running on petascale size systems. The examples of large scale AI workloads parallelized using PCJ are presented.

Distributed and Democratized Learning: Philosophy and Research Challenges

Due to the availability of huge amounts of data and processing abilities, current artificial intelligence (AI) systems are effective in solving complex tasks. However, despite the success of AI in different areas, the problem of designing AI systems that can truly mimic human cognitive capabilities such as artificial general intelligence, remains largely open. Consequently, many emerging cross-device AI applications will require a transition from traditional centralized learning systems towards large-scale distributed AI systems that can collaboratively perform multiple complex learning tasks. In this paper, we propose a novel design philosophy called democratized learning (Dem-AI) whose goal is to build large-scale distributed learning systems that rely on the self-organization of distributed learning agents that are wellconnected, but limited in learning capabilities. Correspondingly, inspired by the societal groups of humans, the specialized groups of learning agents in the proposed Dem-AI system are selforganized in a hierarchical structure to collectively perform learning tasks more efficiently. As such, the Dem-AI learning system can evolve and regulate itself based on the underlying duality of two processes which we call specialized and generalized processes. In this regard, we present a reference design as a guideline to realize future Dem-AI systems, inspired by various interdisciplinary fields. Accordingly, we introduce four underlying mechanisms in the design such as plasticity-stability transition mechanism, self-organizing hierarchical structuring, specialized learning, and generalization. Finally, we establish possible extensions and new challenges for the existing learning approaches to provide better scalable, flexible, and more powerful learning systems with the new setting of Dem-AI.

AI Regulation in the European Union and Trade Law: How Can Accountability of AI and a High Level of Consumer Protection Prevail over a Trade Discipline on Source Code?

Artificial Intelligence (AI) applications can bring many benefits for consumers, as well as influence consumer behaviour and the choices they make. On a large scale AI can profoundly transform consumer markets by, for example, enabling fully personalised consumer transactions on a population-wide scale. AI-powered consumer services rapidly diffuse across the global digital ecosystem thereby connecting consumers in the European Union (EU) to business operating from outside the EU. Individuals who are at the receiving end of AI systems must be reassured that these technologies operate in a way that respects fundamental and consumer rights. In the current negotiations on electronic commerce at the World Trade Organisation (WTO), the EU supports the introduction – in the legal text – of a clause which prohibits the participating countries to introduce – in their national laws – measures that require access to, or transfer of, the source code of software, with some exceptions. This is a cause for concern for experts and rights advocates, as such a clause – if not carefully conditioned – can prevent future EU regulation of AI that may be harmful to consumers. The Federation of German Consumer Organisations (Verbraucherzentrale Bundesverband – vzbv) has commissioned this study from the Institute for Information Law (IViR) at the University of Amsterdam, in order to shed light on the cross-border supply of AI technology and its impact on EU consumer rights. This study forms a comprehensive understanding of this issue that intersects three different areas: (1) emerging EU governance of AI and (2) the application of EU consumer protection law to AI with (3) the EU’s position in the WTO electronic commerce negotiations. This study concludes that the source code clause within trade law indeed restricts the EU’s right to regulate in the field of AI governance in several important ways.