The proliferation of applications using artificial intelligence (AI) systems has led to a growing number of users interacting with these systems through sophisticated interfaces. Human-computer interaction research has long shown that interfaces shape both user behavior and user perception of technical capabilities and risks. Yet, practitioners and researchers evaluating the social and ethical risks of AI systems tend to overlook the impact of anthropomorphic, deceptive, and immersive interfaces on human-AI interactions. Here, we argue that design features of interfaces with adaptive AI systems can have cascading impacts, driven by feedback loops, which extend beyond those previously considered. We first conduct a scoping review of AI interface designs and their negative impact to extract salient themes of potentially harmful design patterns in AI interfaces. Then, we propose Design-Enhanced Control of AI systems (DECAI), a conceptual model to structure and facilitate impact assessments of AI interface designs. DECAI draws on principles from control systems theory -- a theory for the analysis and design of dynamic physical systems -- to dissect the role of the interface in human-AI systems. Through two case studies on recommendation systems and conversational language model systems, we show how DECAI can be used to evaluate AI interface designs.

The widespread use of machine learning and data-driven algorithms for decision making has been steadily increasing over many years. The areas in which this is happening are diverse: healthcare, employment, finance, education, the legal system to name a few; and the associated negative side effects are being increasingly harmful for society. Negative data \emph{bias} is one of those, which tends to result in harmful consequences for specific groups of people. Any mitigation strategy or effective policy that addresses the negative consequences of bias must start with awareness that bias exists, together with a way to understand and quantify it. However, there is a lack of consensus on how to measure data bias and oftentimes the intended meaning is context dependent and not uniform within the research community. The main contributions of our work are: (1) a general algorithmic framework for defining and efficiently quantifying the bias level of a dataset with respect to a protected group; and (2) the definition of a new bias measure. Our results are experimentally validated using nine publicly available datasets and theoretically analyzed, which provide novel insights about the problem. Based on our approach, we also derive a bias mitigation algorithm that might be useful to policymakers.

People use language for various purposes. Apart from sharing information, individuals may use it to express emotions or to show respect for another person. In this paper, we focus on the formality level of machine-generated translations and present FAME-MT -- a dataset consisting of 11.2 million translations between 15 European source languages and 8 European target languages classified to formal and informal classes according to target sentence formality. This dataset can be used to fine-tune machine translation models to ensure a given formality level for each European target language considered. We describe the dataset creation procedure, the analysis of the dataset's quality showing that FAME-MT is a reliable source of language register information, and we present a publicly available proof-of-concept machine translation model that uses the dataset to steer the formality level of the translation. Currently, it is the largest dataset of formality annotations, with examples expressed in 112 European language pairs. The dataset is published online: https://github.com/laniqo-public/fame-mt/ .

Machine Learning algorithms (ML) impact virtually every aspect of human lives and have found use across diverse sectors, including healthcare, finance, and education. Often, ML algorithms have been found to exacerbate societal biases presented in datasets, leading to adversarial impacts on subsets/groups of individuals, in many cases minority groups. To effectively mitigate these untoward effects, it is crucial that disparities/biases are identified and assessed early in a ML pipeline. This proactive approach facilitates timely interventions to prevent bias amplification and reduce complexity at later stages of model development. In this paper, we introduce DispaRisk, a novel framework designed to proactively assess the potential risks of disparities in datasets during the initial stages of the ML pipeline. We evaluate DispaRisk's effectiveness by benchmarking it with commonly used datasets in fairness research. Our findings demonstrate the capabilities of DispaRisk to identify datasets with a high-risk of discrimination, model families prone to biases, and characteristics that heighten discrimination susceptibility in a ML pipeline. The code for our experiments is available in the following repository: https://github.com/jovasque156/disparisk

The rapid advancement of foundation models (FMs) across language, image, audio, and video domains has shown remarkable capabilities in diverse tasks. However, the proliferation of FMs brings forth a critical challenge: the potential to generate hallucinated outputs, particularly in high-stakes applications. The tendency of foundation models to produce hallucinated content arguably represents the biggest hindrance to their widespread adoption in real-world scenarios, especially in domains where reliability and accuracy are paramount. This survey paper presents a comprehensive overview of recent developments that aim to identify and mitigate the problem of hallucination in FMs, spanning text, image, video, and audio modalities. By synthesizing recent advancements in detecting and mitigating hallucination across various modalities, the paper aims to provide valuable insights for researchers, developers, and practitioners. Essentially, it establishes a clear framework encompassing definition, taxonomy, and detection strategies for addressing hallucination in multimodal foundation models, laying the foundation for future research in this pivotal area.

Social media abounds with multimodal sarcasm, and identifying sarcasm targets is particularly challenging due to the implicit incongruity not directly evident in the text and image modalities. Current methods for Multimodal Sarcasm Target Identification (MSTI) predominantly focus on superficial indicators in an end-to-end manner, overlooking the nuanced understanding of multimodal sarcasm conveyed through both the text and image. This paper proposes a versatile MSTI framework with a coarse-to-fine paradigm, by augmenting sarcasm explainability with reasoning and pre-training knowledge. Inspired by the powerful capacity of Large Multimodal Models (LMMs) on multimodal reasoning, we first engage LMMs to generate competing rationales for coarser-grained pre-training of a small language model on multimodal sarcasm detection. We then propose fine-tuning the model for finer-grained sarcasm target identification. Our framework is thus empowered to adeptly unveil the intricate targets within multimodal sarcasm and mitigate the negative impact posed by potential noise inherently in LMMs. Experimental results demonstrate that our model far outperforms state-of-the-art MSTI methods, and markedly exhibits explainability in deciphering sarcasm as well.

Google announced this week that it would begin the international rollout of its new artificial intelligence-powered search feature, called AI Overviews. When billions of people search a range of topics from news to recipes to general knowledge questions, what they see first will now be an AI-generated summary. Google touted AI Overviews at its annual I/O developer conference as a way of delivering customers quick answers and simplifying the online search experience, but it also has another effect on the way that people engage with the internet: keeping users, and advertisers, on Google.com. "Google will do the googling for you," said Liz Reid, head of Google Search. While Google was once mostly a portal to reach other parts of the internet, it has spent years consolidating content and services to make itself into the web's primary destination.

Many recent studies have shown that the perception of speech can be decoded from brain signals and subsequently reconstructed as continuous language. However, there is a lack of neurological basis for how the semantic information embedded within brain signals can be used more effectively to guide language reconstruction. The theory of predictive coding suggests that human brain naturally engages in continuously predicting future word representations that span multiple timescales. This implies that the decoding of brain signals could potentially be associated with a predictable future. To explore the predictive coding theory within the context of language reconstruction, this paper proposes a novel model \textsc{PredFT} for jointly modeling neural decoding and brain prediction. It consists of a main decoding network for language reconstruction and a side network for predictive coding. The side network obtains brain predictive coding representation from related brain regions of interest with a multi-head self-attention module. This representation is fused into the main decoding network with cross-attention to facilitate the language models' generation process. Experiments are conducted on the largest naturalistic language comprehension fMRI dataset Narratives. \textsc{PredFT} achieves current state-of-the-art decoding performance with a maximum BLEU-1 score of $27.8\%$.

Laser warfare, among all the long-unfulfilled imaginings of science fiction writers, is right up there with flying cars. After decades of research, the US military is actively deploying laser defense systems in the Middle East to shoot down drones launched by adversaries like Yemen's Houthi rebels, one of several recent deployments of laser tech in actual combat situations. In less pew-pew-oriented security news, the debate continues over the extension of Section 702 of the Foreign Intelligence Surveillance Act, signed by President Biden last month, as 20 civil liberties organizations sent a letter to the Justice Department demanding more clarity on when the NSA can demand US tech companies cooperate in its wiretaps. Elsewhere, WIRED obtained emails showing how New York City decided to deploy a gun-detection system called Evolv in subways despite false-positive rates as high as 85 percent. At the Google I/O developer conference, meanwhile, the search giant debuted a new AI-based feature in Android that's designed to detect if a phone has been stolen and automatically lock it down.

In the context of enterprises accumulating proprietary unstructured data, AI-driven information retrieval solutions have emerged as vital tools for extracting relevant answers to employee queries. Traditional methods for developing such solutions often involve choosing between Retrieval Augmented Generation (RAG) or fine-tuned Large Language Models (LLMs). However, fine-tuned LLMs, comprising only generative models, lack a guarantee of factual accuracy, while RAG, comprising an embedding model and a generative model, assures factual precision (Lewis at al., 2020 [1]). Despite their superior performance in general, RAG based solutions often rely on pre-trained models, potentially leading to suboptimal alignment with enterprise-specific data. Addressing this challenge entails exploring two potential avenues: Firstly, recent studies such as RAFT (Zhang et al., 2024 [2]) explore the integration of fine-tuned generative models within a RAG pipeline to enhance accuracy, albeit requiring substantial domain-specific data to fine-tune the generative models. Alternatively, leveraging domain-specific embedding models within a RAG pipeline to enhance accuracy remains an underexplored area. Earlier efforts, such as BioBERT (Lee et al., 2019 [3]), SciBERT (Beltagy et al., 2019 [4]), and LEGAL-BERT (Chalkidis et al., 2020 [5]) have effectively demonstrated the efficacy of domain-specific embeddings in information retrieval tasks. These endeavors primarily investigated two methodologies: (a) extending the pre-training of BERT and (b) pre-training BERT from scratch, both employing domain-specific corpora. Despite yielding commendable results, these methodologies necessitated substantial domainspecific corpora, with figures as staggering as 21.3B words for BioBERT, 3.17B tokens for SciBERT, and 11.5GB of text data for LEGAL-BERT, thereby posing significant challenges, particularly in low-resource domains like enterprises.