A Large Language Model (LLM) is an artificial intelligence system that has been trained on vast amounts of natural language data, enabling it to generate human-like responses to written or spoken language input. GPT-3.5 is an example of an LLM that supports a conversational agent called ChatGPT. In this work, we used a series of novel prompts to determine whether ChatGPT shows heuristics, biases, and other decision effects. We also tested the same prompts on human participants. Across four studies, we found that ChatGPT was influenced by random anchors in making estimates (Anchoring Heuristic, Study 1); it judged the likelihood of two events occurring together to be higher than the likelihood of either event occurring alone, and it was erroneously influenced by salient anecdotal information (Representativeness and Availability Heuristic, Study 2); it found an item to be more efficacious when its features were presented positively rather than negatively - even though both presentations contained identical information (Framing Effect, Study 3); and it valued an owned item more than a newly found item even though the two items were identical (Endowment Effect, Study 4). In each study, human participants showed similar effects. Heuristics and related decision effects in humans are thought to be driven by cognitive and affective processes such as loss aversion and effort reduction. The fact that an LLM - which lacks these processes - also shows such effects invites consideration of the possibility that language may play a role in generating these effects in humans.
相關內容
In this work, we explore a framework for contextual decision-making to study how the relevance and quantity of past data affects the performance of a data-driven policy. We analyze a contextual Newsvendor problem in which a decision-maker needs to trade-off between an underage and an overage cost in the face of uncertain demand. We consider a setting in which past demands observed under ``close by'' contexts come from close by distributions and analyze the performance of data-driven algorithms through a notion of context-dependent worst-case expected regret. We analyze the broad class of Weighted Empirical Risk Minimization (WERM) policies which weigh past data according to their similarity in the contextual space. This class includes classical policies such as ERM, k-Nearest Neighbors and kernel-based policies. Our main methodological contribution is to characterize exactly the worst-case regret of any WERM policy on any given configuration of contexts. To the best of our knowledge, this provides the first understanding of tight performance guarantees in any contextual decision-making problem, with past literature focusing on upper bounds via concentration inequalities. We instead take an optimization approach, and isolate a structure in the Newsvendor loss function that allows to reduce the infinite-dimensional optimization problem over worst-case distributions to a simple line search. This in turn allows us to unveil fundamental insights that were obfuscated by previous general-purpose bounds. We characterize actual guaranteed performance as a function of the contexts, as well as granular insights on the learning curve of algorithms.
Large language models (LLMs) have demonstrated impressive capabilities in general scenarios, exhibiting a level of aptitude that approaches, in some aspects even surpasses, human-level intelligence. Among their numerous skills, the translation abilities of LLMs have received considerable attention. In contrast to traditional machine translation that focuses solely on source-target mapping, LLM-based translation can potentially mimic the human translation process that takes many preparatory steps to ensure high-quality translation. This work aims to explore this possibility by proposing the MAPS framework, which stands for Multi-Aspect Prompting and Selection. Specifically, we enable LLMs to first analyze the given source text and extract three aspects of translation-related knowledge: keywords, topics and relevant demonstrations to guide the translation process. To filter out the noisy and unhelpful knowledge, we employ a selection mechanism based on quality estimation. Experiments suggest that MAPS brings significant and consistent improvements over text-davinci-003 and Alpaca on eight translation directions from the latest WMT22 test sets. Our further analysis shows that the extracted knowledge is critical in resolving up to 59% of hallucination mistakes in translation. Code is available at //github.com/zwhe99/MAPS-mt.
In the modern world, we are permanently using, leveraging, interacting with, and relying upon systems of ever higher sophistication, ranging from our cars, recommender systems in e-commerce, and networks when we go online, to integrated circuits when using our PCs and smartphones, the power grid to ensure our energy supply, security-critical software when accessing our bank accounts, and spreadsheets for financial planning and decision making. The complexity of these systems coupled with our high dependency on them implies both a non-negligible likelihood of system failures, and a high potential that such failures have significant negative effects on our everyday life. For that reason, it is a vital requirement to keep the harm of emerging failures to a minimum, which means minimizing the system downtime as well as the cost of system repair. This is where model-based diagnosis comes into play. Model-based diagnosis is a principled, domain-independent approach that can be generally applied to troubleshoot systems of a wide variety of types, including all the ones mentioned above, and many more. It exploits and orchestrates i.a. techniques for knowledge representation, automated reasoning, heuristic problem solving, intelligent search, optimization, stochastics, statistics, decision making under uncertainty, machine learning, as well as calculus, combinatorics and set theory to detect, localize, and fix faults in abnormally behaving systems. In this thesis, we will give an introduction to the topic of model-based diagnosis, point out the major challenges in the field, and discuss a selection of approaches from our research addressing these issues.
Just as other disciplines, the humanities explore how computational research approaches and tools can meaningfully contribute to scholarly knowledge production. We approach the design of computational tools through the analytical lens of 'human-AI collaboration.' However, there is no generalizable concept of what constitutes 'meaningful' human-AI collaboration. In terms of genuinely human competencies, we consider criticality and reflection as guiding principles of scholarly knowledge production. Although (designing for) reflection is a recurring topic in CSCW and HCI discourses, it has not been centered in work on human-AI collaboration. We posit that integrating both concepts is a viable approach to supporting 'meaningful' human-AI collaboration in the humanities. Our research, thus, is guided by the question of how critical reflection can be enabled in human-AI collaboration. We address this question with a use case that centers on computer vision (CV) tools for art historical image retrieval. Specifically, we conducted a qualitative interview study with art historians and extended the interviews with a think-aloud software exploration. We observed and recorded our participants' interaction with a ready-to-use CV tool in a possible research scenario. We found that critical reflection, indeed, constitutes a core prerequisite for 'meaningful' human-AI collaboration in humanities research contexts. However, we observed that critical reflection was not fully realized during interaction with the CV tool. We interpret this divergence as supporting our hypothesis that computational tools need to be intentionally designed in such a way that they actively scaffold and support critical reflection during interaction. Based on our findings, we suggest four empirically grounded design implications for 'critical-reflective human-AI collaboration'.
Large foundation models have demonstrated a great ability to achieve general human-level intelligence far beyond traditional approaches. As the technique keeps attracting attention from the AI community, more and more large foundation models have become publically available. However, most of those models exhibit a major deficiency in specialized-task applications, where the step of finetuning is still required for obtaining satisfactory performance. As the number of available models and specialized tasks keeps growing, the job of general finetuning becomes highly nontrivial. In this paper, we take the first step to address this issue. We introduce an extensible and lightweight toolkit, LMFlow, which aims to simplify the finetuning and inference of general large foundation models. LMFlow offers a complete finetuning workflow for a large foundation model to support personalized training with limited computing resources. Furthermore, it supports continuous pretraining, instruction tuning, parameter-efficient finetuning, alignment tuning, and large model inference, along with carefully designed and extensible APIs. This toolkit has been thoroughly tested and is available at //github.com/OptimalScale/LMFlow.
Cyber incident response is critical to business continuity -- we describe a new exercise that challenges professionals to play the role of Chief Information Security Officer (CISO) for a major financial organisation. Teams must decide how organisational team and budget resources should be deployed across Enterprise Architecture (EA) upgrades and cyber incidents. Every choice made has an impact -- some prevent whilst others may trigger new or continue current attacks. We explain how the underlying platform supports these interactions through a reactionary event mechanism that introduces events based on the current attack surface of the organisation. We explore how our platform manages to introduce randomness on top of triggered events to ensure that the exercise is not deterministic and better matches incidents in the real world. We conclude by describing next steps for the exercise and how we plan to use it in the future to better understand risk decision making.
Convolutional Neural Networks (CNNs) are frequently and successfully used in medical prediction tasks. They are often used in combination with transfer learning, leading to improved performance when training data for the task are scarce. The resulting models are highly complex and typically do not provide any insight into their predictive mechanisms, motivating the field of 'explainable' artificial intelligence (XAI). However, previous studies have rarely quantitatively evaluated the 'explanation performance' of XAI methods against ground-truth data, and transfer learning and its influence on objective measures of explanation performance has not been investigated. Here, we propose a benchmark dataset that allows for quantifying explanation performance in a realistic magnetic resonance imaging (MRI) classification task. We employ this benchmark to understand the influence of transfer learning on the quality of explanations. Experimental results show that popular XAI methods applied to the same underlying model differ vastly in performance, even when considering only correctly classified examples. We further observe that explanation performance strongly depends on the task used for pre-training and the number of CNN layers pre-trained. These results hold after correcting for a substantial correlation between explanation and classification performance.
Foundation models pretrained on diverse data at scale have demonstrated extraordinary capabilities in a wide range of vision and language tasks. When such models are deployed in real world environments, they inevitably interface with other entities and agents. For example, language models are often used to interact with human beings through dialogue, and visual perception models are used to autonomously navigate neighborhood streets. In response to these developments, new paradigms are emerging for training foundation models to interact with other agents and perform long-term reasoning. These paradigms leverage the existence of ever-larger datasets curated for multimodal, multitask, and generalist interaction. Research at the intersection of foundation models and decision making holds tremendous promise for creating powerful new systems that can interact effectively across a diverse range of applications such as dialogue, autonomous driving, healthcare, education, and robotics. In this manuscript, we examine the scope of foundation models for decision making, and provide conceptual tools and technical background for understanding the problem space and exploring new research directions. We review recent approaches that ground foundation models in practical decision making applications through a variety of methods such as prompting, conditional generative modeling, planning, optimal control, and reinforcement learning, and discuss common challenges and open problems in the field.
Games and simulators can be a valuable platform to execute complex multi-agent, multiplayer, imperfect information scenarios with significant parallels to military applications: multiple participants manage resources and make decisions that command assets to secure specific areas of a map or neutralize opposing forces. These characteristics have attracted the artificial intelligence (AI) community by supporting development of algorithms with complex benchmarks and the capability to rapidly iterate over new ideas. The success of artificial intelligence algorithms in real-time strategy games such as StarCraft II have also attracted the attention of the military research community aiming to explore similar techniques in military counterpart scenarios. Aiming to bridge the connection between games and military applications, this work discusses past and current efforts on how games and simulators, together with the artificial intelligence algorithms, have been adapted to simulate certain aspects of military missions and how they might impact the future battlefield. This paper also investigates how advances in virtual reality and visual augmentation systems open new possibilities in human interfaces with gaming platforms and their military parallels.
Causality can be described in terms of a structural causal model (SCM) that carries information on the variables of interest and their mechanistic relations. For most processes of interest the underlying SCM will only be partially observable, thus causal inference tries to leverage any exposed information. Graph neural networks (GNN) as universal approximators on structured input pose a viable candidate for causal learning, suggesting a tighter integration with SCM. To this effect we present a theoretical analysis from first principles that establishes a novel connection between GNN and SCM while providing an extended view on general neural-causal models. We then establish a new model class for GNN-based causal inference that is necessary and sufficient for causal effect identification. Our empirical illustration on simulations and standard benchmarks validate our theoretical proofs.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191