共找到 20 条结果
To evaluate the societal impacts of GenAI requires a model of how social harms emerge from interactions between GenAI, people, and societal structures. Yet a model is rarely explicitly defined in societal impact evaluations, or in the taxonomies of societal impacts that support them. In this provocation, we argue that societal impacts should be conceptualised as application- and context-specific, incommensurable, and shaped by questions of social power. Doing so leads us to conclude that societal impact evaluations using existing taxonomies are inherently limited, in terms of their potential to reveal how GenAI systems may interact with people when introduced into specific social contexts. We therefore propose a governance-first approach to managing societal harms attended by GenAI technologies.
In natural language understanding (NLU) production systems, users' evolving needs necessitate the addition of new features over time, indexed by new symbols added to the meaning representation space. This requires additional training data and results in ever-growing datasets. We present the first systematic investigation of this incremental symbol learning scenario. Our analysis reveals a troubling quirk in building broad-coverage NLU systems: as the training dataset grows, performance on the new symbol often decreases if we do not accordingly increase its training data. This suggests that it becomes more difficult to learn new symbols with a larger training dataset. We show that this trend holds for multiple mainstream models on two common NLU tasks: intent recognition and semantic parsing. Rejecting class imbalance as the sole culprit, we reveal that the trend is closely associated with an effect we call source signal dilution, where strong lexical cues for the new symbol become diluted as the training dataset grows. Selectively dropping training examples to prevent dilution often reverses the trend, showing the over-reliance of mainstream neural NLU models on simple lexical cues
Multicollinearity is relevant to many different fields where linear regression models are applied, and its existence may affect the analysis of ordinary least squares (OLS) estimators from both the numerical and statistical points of views. Thus, multicollinearity can lead to incoherence in the statistical significance of the independent variables and the global significance of the model. The variance inflation factor (VIF) is traditionally applied to diagnose the possible existence of multicollinearity, but it is not always the case that detection by VIF of a troubling degree of multicollinearity corresponds to negative effects on the statistical analysis. The reason for the lack of specificity of VIF is that there are other factors, such as the size of the sample and the variance of the random disturbance, that can lead to high values of the VIF but not to problematic variance in the OLS estimators (see O'Brien 2007). This paper presents a new variance inflation factor (TVIF) that consider all these additional factors. Thresholds for this new measure and from the index provided by Stewart (1987) are also provided. These thresholds are reinterpreted and presented as a new statisti
The design of algorithms that generate personalized ranked item lists is a central topic of research in the field of recommender systems. In the past few years, in particular, approaches based on deep learning (neural) techniques have become dominant in the literature. For all of them, substantial progress over the state-of-the-art is claimed. However, indications exist of certain problems in today's research practice, e.g., with respect to the choice and optimization of the baselines used for comparison, raising questions about the published claims. In order to obtain a better understanding of the actual progress, we have tried to reproduce recent results in the area of neural recommendation approaches based on collaborative filtering. The worrying outcome of the analysis of these recent works-all were published at prestigious scientific conferences between 2015 and 2018-is that 11 out of the 12 reproducible neural approaches can be outperformed by conceptually simple methods, e.g., based on the nearest-neighbor heuristics. None of the computationally complex neural methods was actually consistently better than already existing learning-based techniques, e.g., using matrix factori
Collectively, machine learning (ML) researchers are engaged in the creation and dissemination of knowledge about data-driven algorithms. In a given paper, researchers might aspire to any subset of the following goals, among others: to theoretically characterize what is learnable, to obtain understanding through empirically rigorous experiments, or to build a working system that has high predictive accuracy. While determining which knowledge warrants inquiry may be subjective, once the topic is fixed, papers are most valuable to the community when they act in service of the reader, creating foundational knowledge and communicating as clearly as possible. Recent progress in machine learning comes despite frequent departures from these ideals. In this paper, we focus on the following four patterns that appear to us to be trending in ML scholarship: (i) failure to distinguish between explanation and speculation; (ii) failure to identify the sources of empirical gains, e.g., emphasizing unnecessary modifications to neural architectures when gains actually stem from hyper-parameter tuning; (iii) mathiness: the use of mathematics that obfuscates or impresses rather than clarifies, e.g., b
We investigate users' perspectives on an online reflective question activity (RQA) that prompts people to externalize their underlying emotions on a troubling situation. Inspired by principles of cognitive behavioral therapy, our 15-minute activity encourages self-reflection without a human or automated conversational partner. A deployment of our RQA on Amazon Mechanical Turk suggests that people perceive several benefits from our RQA, including structured awareness of their thoughts and problem-solving around managing their emotions. Quantitative evidence from a randomized experiment suggests people find that our RQA makes them feel less worried by their selected situation and worth the minimal time investment. A further two-week technology probe deployment with 11 participants indicates that people see benefits to doing this activity repeatedly, although the activity may get monotonous over time. In summary, this work demonstrates the promise of online reflection activities that carefully leverage principles of psychology in their design.
Transformers encode structure in sequences via an expanding contextual history. However, their purely feedforward architecture fundamentally limits dynamic state tracking. State tracking -- the iterative updating of latent variables reflecting an evolving environment -- involves inherently sequential dependencies that feedforward networks struggle to maintain. Consequently, feedforward models push evolving state representations deeper into their layer stack with each new input step, rendering information inaccessible in shallow layers and ultimately exhausting the model's depth. While this depth limit can be bypassed by dynamic depth models and by explicit or latent thinking that externalizes state representations, these solutions are computationally and memory inefficient. In this article, we argue that temporally extended cognition requires refocusing from explicit thought traces to implicit activation dynamics via recurrent architectures. We introduce a taxonomy of recurrent and continuous-thought transformer architectures, categorizing them by their recurrence axis (depth versus step) and their ratio of input tokens to recurrence steps. Finally, we outline promising research di
We identify troubling cases of so-called `permanent underdetermination' in both dark energy and inflationary cosmology. We bring to bear (a) a taxonomy of possible responses to underdetermination, and (b) an understanding of both dark energy and inflationary cosmology from an effective field theory point of view. We argue that, under certain conditions, there are viable responses which can arguably alleviate at least some of the concerns about underdetermination in the dark energy and inflationary sectors. However, the epistemic threat of permanent underdetermination remains a significant challenge.
We adapt a troubled-cell indicator developed for discontinuous Galerkin (DG) methods to the finite volume method (FVM) framework for solving hyperbolic conservation laws. This indicator depends solely on the cell-average data of the target cell and its immediate neighbours. Once the troubled-cells are identified, we apply the limiter only in these cells instead of applying in all computational cells. We introduce a novel technique to quantify the quality of the solution in the neighbourhood of the shock by defining a monotonicity parameter $μ$. Numerical results from various two-dimensional simulations on the hyperbolic systems of Euler equations using a finite volume solver employing MUSCL reconstruction validate the performance of the troubled-cell indicator and the approach of limiting only in the troubled-cells. These results show that limiting only in the troubled-cells is preferable to limiting everywhere as it improves convergence without compromising on the solution accuracy.
The rapid evolution of the telecommunication industry necessitates efficient troubleshooting processes to maintain network reliability, software maintainability, and service quality. Trouble Reports (TRs), which document issues in Ericsson's production system, play a critical role in facilitating the timely resolution of software faults. However, the complexity and volume of TR data, along with the presence of diverse criteria that reflect different aspects of each fault, present challenges for retrieval systems. Building on prior work at Ericsson, which utilized a two-stage workflow, comprising Initial Retrieval (IR) and Re-Ranking (RR) stages, this study investigates different TR observation criteria and their impact on the performance of retrieval models. We propose \textbf{CREST} (\textbf{C}riteria-specific \textbf{R}etrieval via \textbf{E}nsemble of \textbf{S}pecialized \textbf{T}R models), a criterion-driven retrieval approach that leverages specialized models for different TR fields to improve both effectiveness and interpretability, thereby enabling quicker fault resolution and supporting software maintenance. CREST utilizes specialized models trained on specific TR criteri
We adapt a troubled-cell indicator from discontinuous Galerkin (DG) methods to finite volume methods (FVM) with MUSCL reconstruction and using a novel monotonicity parameter show there is a trade-off between convergence and quality of the solution. Employing two dimensional compressible Euler equations for flows with oblique shocks, this trade-off is studied by varying the number of troubled-cells systematically. An oblique shock is characterized primarily by the upstream Mach number, the shock angle $β$, and the deflection angle $θ$. We study these factors and their combinations and find that the degree of the shock misalignment with the grid determines the optimal number of troubled-cells. On each side of the shock, the optimal set consists of three troubled-cells for aligned shocks, and the troubled-cells identified by tracing the shock and four lines parallel to it, separated by the grid spacing, for nonaligned shocks. We show that the adapted troubled-cell indicator identifies a set of cells that is close to and contains the optimal set of cells for a threshold constant $K = 0.05$, and consequently, produces a solution close to that obtained by limiting everywhere, but with im
As large language models (LLMs) advance to produce human-like arguments in some contexts, the number of settings applicable for human-AI collaboration broadens. Specifically, we focus on subjective decision-making, where a decision is contextual, open to interpretation, and based on one's beliefs and values. In such cases, having multiple arguments and perspectives might be particularly useful for the decision-maker. Using subtle sexism online as an understudied application of subjective decision-making, we suggest that LLM output could effectively provide diverse argumentation to enrich subjective human decision-making. To evaluate the applicability of this case, we conducted an interview study (N=20) where participants evaluated the perceived authorship, relevance, convincingness, and trustworthiness of human and AI-generated explanation-text, generated in response to instances of subtle sexism from the internet. In this workshop paper, we focus on one troubling trend in our results related to opinions and experiences displayed in LLM argumentation. We found that participants rated explanations that contained these characteristics as more convincing and trustworthy, particularly
Text-to-image models are increasingly popular and impactful, yet concerns regarding their safety and fairness remain. This study investigates the ability of ten popular Stable Diffusion models to generate harmful images, including NSFW, violent, and personally sensitive material. We demonstrate that these models respond to harmful prompts by generating inappropriate content, which frequently displays troubling biases, such as the disproportionate portrayal of Black individuals in violent contexts. Our findings demonstrate a complete lack of any refusal behavior or safety measures in the models observed. We emphasize the importance of addressing this issue as image generation technologies continue to become more accessible and incorporated into everyday applications.
Against a backdrop of widespread interest in how publics can participate in the design of AI, I argue for a research agenda focused on AI incidents - examples of AI going wrong and sparking controversy - and how they are constructed in online environments. I take up the example of an AI incident from September 2020, when a Twitter user created a 'horrible experiment' to demonstrate the racist bias of Twitter's algorithm for cropping images. This resulted in Twitter not only abandoning its use of that algorithm, but also disavowing its decision to use any algorithm for the task. I argue that AI incidents like this are a significant means for participating in AI systems that require further research. That research agenda, I argue, should focus on how incidents are constructed through networked online behaviours that I refer to as 'networked trouble', where formats for participation enable individuals and algorithms to interact in ways that others - including technology companies - come to know and come to care about. At stake, I argue, is an important mechanism for participating in the design and deployment of AI.
Biological data in digital form has become a, if not the, driving force behind innovations in biology, medicine, and the environment. No study and no model would be complete without access to digital data (including text) collected by others and available in public repositories. With this ascent in the fundamental importance of data for reproducible scientific progress has come a troubling paradox.
In this paper, eight different troubled cell indicators (shock detectors) are reviewed for the solution of nonlinear hyperbolic conservation laws using discontinuous Galerkin (DG) method and a WENO limiter on both structured and unstructured meshes. Extensive simulations using one-dimensional and two-dimensional problems (2D Riemann problem and the double Mach reflection) for various orders on the hyperbolic system of Euler equations are used to compare these troubled cell indicators. They are evaluated based on the percentage of cells flagged as troubled cells for various orders and various grid sizes. CPU time taken to test a single cell for discontinuity is also compared. For one-dimensional problems, the performance of Fu and Shu indicator and the modified KXRCF indicator is better than other indicators. For two-dimensional problems, the performance of the artificial neural network (ANN) indicator of Ray and Hesthaven is quite good and the Fu and Shu and the modified KXRCF indicators are also good. These three indicators are suitable candidates for applications of DGM using WENO limiters though it should be noted that the ANN indicator is quite expensive and requires a lot of t
The ethics of AI as both material and medium for interaction remains in murky waters within the context of musical and artistic practice. The interdisciplinarity of the field is revealing matters of concern and care, which necessitate interdisciplinary methodologies for evaluation to trouble and critique the inheritance of "residue-laden" AI-tools in musical applications. Seeking to unsettle these murky waters, this paper critically examines the example of Holly+, a deep neural network that generates raw audio in the likeness of its creator Holly Herndon. Drawing from theoretical concerns and considerations from speculative feminism and care ethics, we care-fully trouble the structures, frameworks and assumptions that oscillate within and around Holly+. We contribute with several considerations and contemplate future directions for integrating speculative feminism and care into musical-AI agent and system design, derived from our critical feminist examination.
This paper examines some common problems in Human-Robot Interaction (HRI) causing failures and troubles in Chat. A given use case's design decisions start with the suitable robot, the suitable chatting model, identifying common problems that cause failures, identifying potential solutions, and planning continuous improvement. In conclusion, it is recommended to use a closed-loop control algorithm that guides the use of trained Artificial Intelligence (AI) pre-trained models and provides vocabulary filtering, re-train batched models on new datasets, learn online from data streams, and/or use reinforcement learning models to self-update the trained models and reduce errors.
In this position paper, we argue that the classical evaluation on Natural Language Processing (NLP) tasks using annotated benchmarks is in trouble. The worst kind of data contamination happens when a Large Language Model (LLM) is trained on the test split of a benchmark, and then evaluated in the same benchmark. The extent of the problem is unknown, as it is not straightforward to measure. Contamination causes an overestimation of the performance of a contaminated model in a target benchmark and associated task with respect to their non-contaminated counterparts. The consequences can be very harmful, with wrong scientific conclusions being published while other correct ones are discarded. This position paper defines different levels of data contamination and argues for a community effort, including the development of automatic and semi-automatic measures to detect when data from a benchmark was exposed to a model, and suggestions for flagging papers with conclusions that are compromised by data contamination.