AI Problems Research Articles

Attribution problem of generative AI: a view from US copyright law

Attribution problem of generative AI: a view from US copyright law

Counterfactual explanations for misclassified images: How human and machine explanations differ

Counterfactual explanations have emerged as a popular solution for the eXplainable AI (XAI) problem of elucidating the predictions of black-box deep-learning systems because people easily understand them, they apply across different problem domains and seem to be legally compliant. Although over 100 counterfactual methods exist in the XAI literature, each claiming to generate plausible explanations akin to those preferred by people, few of these methods have actually been tested on users (∼7%). Even fewer studies adopt a user-centered perspective; for instance, asking people for their counterfactual explanations to determine their perspective on a “good explanation”. This gap in the literature is addressed here using a novel methodology that (i) gathers human-generated counterfactual explanations for misclassified images, in two user studies and, then, (ii) compares these human-generated explanations to computationally-generated explanations for the same misclassifications. Results indicate that humans do not “minimally edit” images when generating counterfactual explanations. Instead, they make larger, “meaningful” edits that better approximate prototypes in the counterfactual class. An analysis based on “explanation goals” is proposed to account for this divergence between human and machine explanations. The implications of these proposals for future work are discussed.

Maintain a Better Balance between Performance and Cost for Image Captioning by a Size-Adjustable Convolutional Module

Image captioning is a challenging AI problem that connects computer vision and natural language processing. Many deep learning (DL) models have been proposed in the literature for solving this problem. So far, the primary concern of image captioning has been focused on increasing the accuracy of generating human-style sentences for describing given images. As a result, state-of-the-art (SOTA) models are often too expensive to be implemented in computationally weak devices. In contrast, the primary concern of this paper is to maintain a balance between performance and cost. For this purpose, we propose using a DL model pre-trained for object detection to encode the given image so that features of various objects can be extracted simultaneously. We also propose adding a size-adjustable convolutional module (SACM) before decoding the features into sentences. The experimental results show that the model with the properly adjusted SACM could reach a BLEU-1 score of 82.3 and a BLEU-4 score of 43.9 on the Flickr 8K dataset, and a BLEU-1 score of 83.1 and a BLEU-4 score of 44.3 on the MS COCO dataset. With the SACM, the number of parameters is decreased to 108M, which is about 1/4 of the original YOLOv3-LSTM model with 430M parameters. Specifically, compared with mPLUG with 510M parameters, which is one of the SOTA methods, the proposed method can achieve almost the same BLEU-4 scores, but the number of parameters is 78% less than the mPLUG.

Theory languages in designing artificial intelligence

The foundations of AI design discourse are worth analyzing. Here, attention is paid to the nature of theory languages used in designing new AI technologies because the limits of these languages can clarify some fundamental questions in the development of AI. We discuss three types of theory language used in designing AI products: formal, computational, and natural. Formal languages, such as mathematics, logic, and programming languages, have fixed meanings and no actual-world semantics. They are context- and practically content-free. Computational languages use terms referring to the actual world, i.e., to entities, events, and thoughts. Thus, computational languages have actual-world references and semantics. They are thus no longer context- or content-free. However, computational languages always have fixed meanings and, for this reason, limited domains of reference. Finally, unlike formal and computational languages, natural languages are creative, dynamic, and productive. Consequently, they can refer to an unlimited number of objects and their attributes in an unlimited number of domains. The differences between the three theory languages enable us to reflect on the traditional problems of strong and weak AI.

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

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

Semiring Reasoning Frameworks in AI and Their Computational Complexity

Many important problems in AI, among them #SAT, parameter learning and probabilistic inference go beyond the classical satisfiability problem. Here, instead of finding a solution we are interested in a quantity associated with the set of solutions, such as the number of solutions, the optimal solution or the probability that a query holds in a solution. To model such quantitative problems in a uniform manner, a number of frameworks, e.g. Algebraic Model Counting and Semiring-based Constraint Satisfaction Problems, employ what we call the semiring paradigm. In the latter the abstract algebraic structure of the semiring serves as a means of parameterizing the problem definition, thus allowing for different modes of quantitative computations by choosing different semirings. While efficiently solvable cases have been widely studied, a systematic study of the computational complexity of such problems depending on the semiring parameter is missing. In this work, we characterize the latter by NP(R), a novel generalization of NP over semiring R, and obtain NP(R)-completeness results for a selection of semiring frameworks. To obtain more tangible insights into the hardness of NP(R), we link it to well-known complexity classes from the literature. Interestingly, we manage to connect the computational hardness to properties of the semiring. Using this insight, we see that, on the one hand, NP(R) is always at least as hard as NP or ModpP depending on the semiring R and in general unlikely to be in FPSPACEpoly. On the other hand, for broad subclasses of semirings relevant in practice we can employ reductions to NP, ModpP and #P. These results show that in many cases solutions are only mildly harder to compute than functions in NP, ModpP and #P, give us new insights into how problems that involve counting on semirings can be approached, and provide a means of assessing whether an algorithm is appropriate for a given class of problems.

Explainable artificial intelligence and its practical applications

With the continuous development of the times, the artificial intelligence industry is also booming, and its presence in various fields has a huge role in promoting social progress and advancing industrial development. Research on it is also in full swing. People are eager to understand the cause-and-effect relationship between the actions performed or the strategies decided based on the black-box model, so that they can learn or judge from another perspective. Thus the Explainable AI is proposed, it is a new generation of AI that allows humans to understand the cause and give them a decision solution, so that every outcome has its own basis for decision. Although some considerable results have been achieved in the application of explainable AI, it is still at the beginning stage and there are still some challenges to be solved. From the transportation industry, which facilitates people's access to autonomous driving, to the medical industry, which saves people's lives, to the financial industry, which is a huge industry, and even in the education industry, which is accessible to all people, it has a presence. This paper talks about the current situation and problems of explainable AI through its application in various aspects. Explainable AI can serve not only developers but also users by satisfying their interest-related needs. The transparency of explainable AI is important when it is used in socially relevant applications, which is why we have conducted extensive research on explainable AI.

Imperfect-Information Game AI Agent Based on Reinforcement Learning Using Tree Search and a Deep Neural Network

In the field of computer intelligence, it has always been a challenge to construct an agent model that can be adapted to various complex tasks. In recent years, based on the planning algorithm of Monte Carlo tree search (MCTS), a new idea has been proposed to solve the AI problems of two-player zero-sum games such as chess and Go. However, most of the games in the real environment rely on imperfect information, so it is impossible to directly use the normal tree search planning algorithm to construct a decision-making model. Mahjong, which is a popular multiplayer game with a long history in China, attracts great attention from AI researchers because it contains a large game state space and a lot of hidden information. In this paper, we utilize an agent learning approach that leverages deep learning, reinforcement learning, and dropout learning techniques to implement a Mahjong AI game agent. First, we improve the state transition of the tree search based on the learned MDP model, the player position variable and transition information are introduced into the tree search algorithm to construct a multiplayer search tree. Then, the model training based on a deep reinforcement learning method ensures the stable and sustainable training process of the learned MDP model. Finally, we utilize the strategy data generated by the tree search and use the dropout learning method to train the normal decision-making agent. The experimental results demonstrate the efficiency and stability performance of the agent trained by our proposed method compared with existing agents in terms of test data accuracy, tournament ranking performance, and online match performance. The agent plays against human players and acts like real humans.

ASP and subset minimality: Enumeration, cautious reasoning and MUSes

Answer Set Programming (ASP) is a well-known logic-based formalism that has been used to model and solve a variety of AI problems. For several years, ASP implementations primarily focused on the main computational task: the computation of one answer set of a (logic) program. Nonetheless, several AI problems, that can be conveniently modelled in ASP, require to enumerate solutions characterized by an optimality property that can be expressed in terms of subset-minimality with respect to some objective atoms. In this context, solutions are often either (i) answer sets that are subset-minimal w.r.t. the objective atoms or (ii) atoms that are contained in all subset-minimal answer sets, or (iii) sets of atoms that enforce the absence of answer sets on the ASP program at hand — such sets are referred to as minimal unsatisfiable subsets (MUSes). In all the above-mentioned cases, the corresponding computational task is currently not supported by plain state-of-the-art ASP solvers. In this paper, we study formally these tasks and fill the gap in current implementations by proposing several algorithms to enumerate MUSes and subset-minimal answer sets, as well as perform cautious reasoning on subset-minimal answer sets. We implement our algorithms on top of wasp and perform an experimental analysis on several hard benchmarks showing the good performance of our implementation.

Holding AI to Account: Challenges for the Delivery of Trustworthy AI in Healthcare

The need for AI systems to provide explanations for their behaviour is now widely recognised as key to their adoption. In this article, we examine the problem of trustworthy AI and explore what delivering this means in practice, with a focus on healthcare applications. Work in this area typically treats trustworthy AI as a problem of Human–Computer Interaction involving the individual user and an AI system. However, we argue here that this overlooks the important part played by organisational accountability in how people reason about and trust AI in socio-technical settings. To illustrate the importance of organisational accountability, we present findings from ethnographic studies of breast cancer screening and cancer treatment planning in multidisciplinary team meetings to show how participants made themselves accountable both to each other and to the organisations of which they are members. We use these findings to enrich existing understandings of the requirements for trustworthy AI and to outline some candidate solutions to the problems of making AI accountable both to individual users and organisationally. We conclude by outlining the implications of this for future work on the development of trustworthy AI, including ways in which our proposed solutions may be re-used in different application settings.

Introduction, Summary, Questions for the Future

Introduction, Summary, Questions for the Future

О философии искусственного интеллекта и научной революции

The article is devoted to the philosophical analysis of artificial intelligence and its role in understanding the ideas of the scientific revolution. Over the past decades, the problem of AI has not lost its relevance due to the rapid development of cognitive sciences, on the one hand, and the human desire to achieve tangible results in a short period, on the other. The work is an analysis of some modern problems in Azerbaijani science and an analysis of how it can be modernized with the help of AI. According to the authors, Azerbaijani science is at the stage of transition from a pre-problem to a problem stage, the essence of which is detailed in the article. Such significant questions are revealed as what is AI and a revolution in the minds of scientists, what prevents science from being modern, how the scientific revolution can affect the reform processes in other areas, etc. The scientific novelty of the work lies in the philosophical analysis of AI in the context of the scientific revolution. The authors strive to show that the scientific revolution is possible only through a significant change in individual consciousness (consciousness of scientists) and the influence of science on the collective consciousness (consciousness of society). At the same time, the scientific revolution should be understood as a problem that can be solved with the help of AI. The authors also propose to introduce the construction of interdisciplinary and inter-academic relations into the scientific community on the basis of a “horizontal dialogue”.

Leverage zones in Responsible AI: towards a systems thinking conceptualization

There is a growing debate amongst academics and practitioners on whether interventions made, thus far, towards Responsible AI have been enough to engage with the root causes of AI problems. Failure to effect meaningful changes in this system could see these initiatives not reach their potential and lead to the concept becoming another buzzword for companies to use in their marketing campaigns. Systems thinking is often touted as a methodology to manage and effect change; however, there is little practical advice available for decision-makers to include systems thinking insights to work towards Responsible AI. Using the notion of ‘leverage zones’ adapted from the systems thinking literature, we suggest a novel approach to plan for and experiment with potential initiatives and interventions. This paper presents a conceptual framework called the Five Ps to help practitioners construct and identify holistic interventions that may work towards Responsible AI, from lower-order interventions such as short-term fixes, tweaking algorithms and updating parameters, through to higher-order interventions such as redefining the system’s foundational structures that govern those parameters, or challenging the underlying purpose upon which those structures are built and developed in the first place. Finally, we reflect on the framework as a scaffold for transdisciplinary question-asking to improve outcomes towards Responsible AI.

Which neural networks can be computed by an algorithm? – Generalised hardness of approximation meets Deep Learning

AbstractClassical hardness of approximation (HA) is the phenomenon that, assuming P ≠ NP, one can easily compute an ϵ‐approximation to the solution of a discrete computational problem for ϵ > ϵ0 > 0, but for ϵ < ϵ0 – where ϵ0 is the approximation threshold – it becomes intractable. Recently, a similar yet more general phenomenon has been documented in AI: Generalised hardness of approximation (GHA). This phenomenon includes the following occurrence: For any approximation threshold ϵ1 > 0, there are AI problems for which provably there exist stable neural networks (NNs) that solve the problem, but no algorithm can compute any NN that approximates the AI problem to ϵ1‐accuracy. Moreover, this issue is independent of the P vs NP question and thus is a rather different mathematical phenomenon than HA. GHA implies that the universal approximation theorem for NNs only provides a partial understanding of the power of NNs in AI. Thus, a classification theory describing which NNs can be computed by algorithms to particular accuracies is needed to fill this gap. We initiate such a theory by showing the correspondence between the functions that can be computed to ϵ‐accuracy by an algorithm and those functions that can be approximated by NNs which can be computed to ϵ̂‐accuracy by an algorithm. In particular, the approximation thresholds ϵ and ϵ̂ cannot differ by more than a factor of 12. This means that computing function approximations through NNs will be optimal – in the sense of best approximation accuracy achievable by an algorithm – up to a small constant, compared to any other computational technique.

Where does ChatGPT fit into the Framework for Information Literacy? The possibilities and problems of AI in library instruction

Where does ChatGPT fit into the Framework for Information Literacy? The possibilities and problems of AI in library instruction

Robotizácia, umelá inteligencia a budúcnosť učiteľskej profesie

Robots are used in many different fields today, including education. For example, students are introduced to technology through their programming. In non-technical education, robots are used as teaching aids. In assistive robotics, they are used for students' cognitive development. However, robots can also be used in the role of teachers. Experiments have already been conducted in this regard. However, with existing technologies, robots cannot yet be fully autonomous. General (strong) AI cannot replace human teachers today. Nevertheless, it is interesting and important to consider such possibilities. Robots may be able to teach the curriculum more effectively and have much more information at their disposal than human teachers, but can and should we leave the moral education of students to them? After all, the educational process is more than just imparting knowledge. For human being, as a biological species, it is very important to be in contact with other humans, so it is important that the interpersonal teacher-student relationship be part of the educational process. The aim of the study is to draw attention to the topic of the use of artificial intelligence in education with an emphasis on its ethical and axiological dimension. Rather than giving a complete overview of the problem of AI in education, it is intended to stimulate discussion in our region.

일본 지방자치단체에서 AI/RPA 도입과 행정 업무 자동화 분석

이 연구는 일본의 지방자치단체에서 AI/Robotics 기술 도입이 진행되고 있는 배경과 필요성, AI/RPA 도입 과정, AI/RPA 기술을 활용한 행정업무 자동화 현황을 살펴본다. 그리고 AI/Robotics 도입의 효과, 일본의 지방행정, 지역경영에 주는 함의를 분석하였다. 일본 지방자치단체에서 도입 분야를 살펴보면, 매칭, 최적 방안제시, 화상인식, 미래예측 분야에서 도입이 증가하고 있다. 지방자치단체에서 AI는 행정업무의 효율화, 주민 서비스 향상 그리고 미래의 지속가능성을 확보하는 수단으로 활용하고 있다는 점이 특징적이다. 지방자치단체에서 AI/RPA 도입 효과는 정량적, 정성적 효과로 측정되고 있다. 지방자치단체에서 AI/RPA 도입은 실제로 고참 공무원과 신임 공무원 간 업무처리 시간을 2.6배에서 1.2배로 감소시켰다. 행정업무 도입에 따른 시간 절약 효과도 주민 28.3만 시간, 공무원 1.78만 시간 감소시키는 것으로 나타났다. 결국, AI/RPA 도입으로 지역주민에게 생기는 시간 절감 효과가 크다는 것을 알 수 있다. 일본에서는 이러한 문제를 해결하기 위하여 「지방자치단체에서 AI/RPA 도입 가이드 라인」을 중앙정부가 책정하였다. 가이드라인에는 도입 배경, 목적, 사업 개요, 활용 방침, 데이터, 보안대책, 기대효과, 추진 체계, 추진 일정, 예산 계획 등과 관련된 내용을 담고 있다. 지방자치단체에서 AI/RPA 도입을 위해서는 전문 인력 충원, 자치단체 간 협력 시스템 구축이 필요하다. AI/RPA 도입에는 전문 인력이 필요하다. 전문 인력 충원은 연수, 외부 전문과정 참여 등 다양한 방법을 모색할 수 있다. 그리고 민간 기업과 협업체계를 구축하여 대응할 필요가 있다.

Artificial intelligence for oral and dental healthcare: Core education curriculum

ObjectivesArtificial intelligence (AI) is swiftly entering oral health services and dentistry, while most providers show limited knowledge and skills to appraise dental AI applications. We aimed to define a core curriculum for both undergraduate and postgraduate education, establishing a minimum set of outcomes learners should acquire when taught about oral and dental AI. MethodsExisting curricula and other documents focusing on literacy of medical professionals around AI were screened and relevant items extracted. Items were scoped and adapted using expert interviews with members of the IADR's e-oral health group, the ITU/WHO's Focus Group AI for Health and the Association for Dental Education in Europe. Learning outcome levels were defined and each item assigned to a level. Items were systematized into domains and a curricular structure defined. The resulting curriculum was consented using an online Delphi process. ResultsFour domains of learning outcomes emerged, with most outcomes being on the “knowledge” level: (1) Basic definitions and terms, the reasoning behind AI and the principle of machine learning, the idea of training, validating and testing models, the definition of reference tests, the contrast between dynamic and static AI, and the problem of AI being a black box and requiring explainability should be known. (2) Use cases, the required types of AI to address them, and the typical setup of AI software for dental purposes should be taught. (3) Evaluation metrics, their interpretation, the relevant impact of AI on patient or societal health outcomes and associated examples should be considered. (4) Issues around generalizability and representativeness, explainability, autonomy and accountability and the need for governance should be highlighted. ConclusionBoth educators and learners should consider this core curriculum during planning, conducting and evaluating oral and dental AI education. Clinical significanceA core curriculum on oral and dental AI may help to increase oral and dental healthcare providers’ literacy around AI, allowing them to critically appraise AI applications and to use them consciously and on an informed basis.

CrowdOptim: A Crowd-driven Neural Network Hyperparameter Optimization Approach to AI-based Smart Urban Sensing

AI-based smart urban sensing (ASUS) has emerged as a scalable and pervasive application paradigm in smart city planning and management that aims to automatically assess the physical status of the urban environments by leveraging AI techniques and massive urban sensing data. In this paper, we focus on a crowd-driven neural network (NN) hyperparameter optimization problem in ASUS applications. Our goal is to utilize the human intelligence from crowdsourcing systems to identify the optimal NN hyperparameter configuration for an ASUS model. Our work is motivated by the observation that the hyperparameters of current ASUS models are often manually configured by the AI specialists, which is known to be both error-prone and suboptimal. Two key technical challenges exist in solving our problem: i) it is challenging to effectively translate the highly complex NN hyperparameter optimization problem in AI to a simplified problem that can be solved by crowd workers without extensive AI expertise; ii) it is difficult to identify the optimal hyperparameter configuration in the large hyperparameter search space given the blackbox nature of the AI model. To address the above challenges, we develop CrowdOptim, a crowd-AI collaborative learning framework that integrates the techniques from crowdsourcing, hyperparameter optimization, and estimation theory to address the crowd-driven NN hyperparameter optimization problem in ASUS applications. The evaluation results from two real-world ASUS applications (i.e., smart city infrastructure monitoring (SCIM) and urban environment cleanliness assessment (UECA)) show that CrowdOptim consistently outperforms the state-of-the-art deep convolutional networks, crowd-AI, and hyperparameter optimization baselines in achieving the ASUS application objectives under various evaluation scenarios.

Conversational AI for multi-agent communication in Natural Language

Research at the Interaction Lab focuses on human-agent communication using conversational Natural Language. The ultimate goal is to create systems where humans and AI agents (including embodied robots) can spontaneously form teams and coordinate shared tasks through the use of Natural Language conversation as a universal communication interface. This paper first introduces machine learning approaches to problems in conversational AI in general, where computational agents must coordinate with humans to solve tasks using conversational Natural Language. It also covers some of the practical systems developed in the Interaction Lab, ranging from speech interfaces on smart speakers to embodied robots interacting using visually grounded language. In several cases communication between multiple agents is addressed. The paper surveys the central research problems addressed here, the approaches developed, and our main results. Some key open research questions and directions are then discussed, leading towards a future vision of conversational, collaborative multi-agent systems.