A growing ecosystem of large, open-source foundation models has reduced the labeled data and technical expertise necessary to apply machine learning to many new problems. Yet foundation models pose a clear dual-use risk, indiscriminately reducing the costs of building both harmful and beneficial machine learning systems. Policy tools such as restricted model access and export controls are the primary methods currently used to mitigate such dual-use risks. In this work, we review potential safe-release strategies and argue that both policymakers and AI researchers would benefit from fundamentally new technologies enabling more precise control over the downstream usage of open-source foundation models. We propose one such approach: the task blocking paradigm, in which foundation models are trained with an additional mechanism to impede adaptation to harmful tasks without sacrificing performance on desirable tasks. We call the resulting models self-destructing models, inspired by mechanisms that prevent adversaries from using tools for harmful purposes. We present an algorithm for training self-destructing models leveraging techniques from meta-learning and adversarial learning, which we call meta-learned adversarial censoring (MLAC). In a small-scale experiment, we show MLAC can largely prevent a BERT-style model from being re-purposed to perform gender identification without harming the model's ability to perform profession classification.

The legality of training language models (LMs) on copyrighted or otherwise restricted data is under intense debate. However, as we show, model performance significantly degrades if trained only on low-risk text (e.g., out-of-copyright books or government documents), due to its limited size and domain coverage. We present SILO, a new language model that manages this risk-performance tradeoff during inference. SILO is built by (1) training a parametric LM on Open License Corpus (OLC), a new corpus we curate with 228B tokens of public domain and permissively licensed text and (2) augmenting it with a more general and easily modifiable nonparametric datastore (e.g., containing copyrighted books or news) that is only queried during inference. The datastore allows use of high-risk data without training on it, supports sentence-level data attribution, and enables data producers to opt out from the model by removing content from the store. These capabilities can foster compliance with data-use regulations such as the fair use doctrine in the United States and the GDPR in the European Union. Our experiments show that the parametric LM struggles on domains not covered by OLC. However, access to the datastore greatly improves out of domain performance, closing 90% of the performance gap with an LM trained on the Pile, a more diverse corpus with mostly high-risk text. We also analyze which nonparametric approach works best, where the remaining errors lie, and how performance scales with datastore size. Our results suggest that it is possible to build high quality language models while mitigating their legal risk.

Knowledge Representation and Reasoning is a central, longstanding, and active area of Artificial Intelligence. Over the years it has evolved significantly; more recently it has been challenged and complemented by research in areas such as machine learning and reasoning under uncertainty. In July 2022 a Dagstuhl Perspectives workshop was held on Knowledge Representation and Reasoning. The goal of the workshop was to describe the state of the art in the field, including its relation with other areas, its shortcomings and strengths, together with recommendations for future progress. We developed this manifesto based on the presentations, panels, working groups, and discussions that took place at the Dagstuhl Workshop. It is a declaration of our views on Knowledge Representation: its origins, goals, milestones, and current foci; its relation to other disciplines, especially to Artificial Intelligence; and on its challenges, along with key priorities for the next decade.

Prompting is used to guide or steer a language model in generating an appropriate response that is consistent with the desired outcome. Chaining is a strategy used to decompose complex tasks into smaller, manageable components. In this study, we utilize prompt chaining for extensive legal document classification tasks, which present difficulties due to their intricate domain-specific language and considerable length. Our approach begins with the creation of a concise summary of the original document, followed by a semantic search for related exemplar texts and their corresponding annotations from a training corpus. Finally, we prompt for a label - based on the task - to assign, by leveraging the in-context learning from the few-shot prompt. We demonstrate that through prompt chaining, we can not only enhance the performance over zero-shot, but also surpass the micro-F1 score achieved by larger models, such as ChatGPT zero-shot, using smaller models.

The combination of clinical judgement and predictive risk models crucially assist social workers to segregate children at risk of maltreatment and decide when authorities should intervene. Predictive risk modelling to address this matter has been initiated by several governmental welfare authorities worldwide involving administrative data and machine learning algorithms. While previous studies have investigated risk factors relating to child maltreatment, several gaps remain as to understanding how such risk factors interact and whether predictive risk models perform differently for children with different features. By integrating Principal Component Analysis and K-Means clustering, this paper presents initial findings of our work on the identification of such features as well as their potential effect on current risk modelling frameworks. This approach allows examining existent, unidentified yet, clusters of New Zealand (NZ) children reported with care and protection concerns, as well as to analyse their inner structure, and evaluate the performance of prediction models trained cluster wise. We aim to discover the extent of clustering degree required as an early step in the development of predictive risk models for child maltreatment and so enhance the accuracy of such models intended for use by child protection authorities. The results from testing LASSO logistic regression models trained on identified clusters revealed no significant difference in their performance. The models, however, performed slightly better for two clusters including younger children. our results suggest that separate models might need to be developed for children of certain age to gain additional control over the error rates and to improve model accuracy. While results are promising, more evidence is needed to draw definitive conclusions, and further investigation is necessary.

This is an experiential study of investigating a consistent method for deriving the correlation between sentence vector and semantic meaning of a sentence. We first used three state-of-the-art word/sentence embedding methods including GPT-3, Word2Vec, and Sentence-BERT, to embed plain text sentence strings into high dimensional spaces. Then we compute the pairwise distance between any possible combination of two sentence vectors in an embedding space and map them into a matrix. Based on each distance matrix, we compute the correlation of distances of a sentence vector with respect to the other sentence vectors in an embedding space. Then we compute the correlation of each pair of the distance matrices. We observed correlations of the same sentence in different embedding spaces and correlations of different sentences in the same embedding space. These observations are consistent with our hypothesis and take us to the next stage.

The advancement of biomedical research heavily relies on access to large amounts of medical data. In the case of histopathology, Whole Slide Images (WSI) and clinicopathological information are valuable for developing Artificial Intelligence (AI) algorithms for Digital Pathology (DP). Transferring medical data "as open as possible" enhances the usability of the data for secondary purposes but poses a risk to patient privacy. At the same time, existing regulations push towards keeping medical data "as closed as necessary" to avoid re-identification risks. Generally, these legal regulations require the removal of sensitive data but do not consider the possibility of data linkage attacks due to modern image-matching algorithms. In addition, the lack of standardization in DP makes it harder to establish a single solution for all formats of WSIs. These challenges raise problems for bio-informatics researchers in balancing privacy and progress while developing AI algorithms. This paper explores the legal regulations and terminologies for medical data-sharing. We review existing approaches and highlight challenges from the histopathological perspective. We also present a data-sharing guideline for histological data to foster multidisciplinary research and education.

Many sets of ethics principles for responsible AI have been proposed to allay concerns about misuse and abuse of AI/ML systems. The underlying aspects of such sets of principles include privacy, accuracy, fairness, robustness, explainability, and transparency. However, there are potential tensions between these aspects that pose difficulties for AI/ML developers seeking to follow these principles. For example, increasing the accuracy of an AI/ML system may reduce its explainability. As part of the ongoing effort to operationalise the principles into practice, in this work we compile and discuss a catalogue of 10 notable tensions, trade-offs and other interactions between the underlying aspects. We primarily focus on two-sided interactions, drawing on support spread across a diverse literature. This catalogue can be helpful in raising awareness of the possible interactions between aspects of ethics principles, as well as facilitating well-supported judgements by the designers and developers of AI/ML systems.

In April, lawyers for the airline Avianca noticed something strange. A passenger, Robert Mata, had sued the airline, alleging that a serving cart on a flight had struck and severely injured his left knee, but several cases cited in Mata's lawsuit didn't appear to exist. The judge couldn't verify them, either. It turned out that ChatGPT had made them all up, fabricating names and decisions. One of Mata's lawyers, Steven A. Schwartz, had used the chatbot as an assistant--his first time using the program for legal research--and, as Schwartz wrote in an affidavit, "was unaware of the possibility that its content could be false."

After being charged in court with robbery and carjacking, Ms. Woodruff was released that evening on a $100,000 personal bond. In an interview, she said she went straight to the hospital where she was diagnosed with dehydration and given two bags of intravenous fluids. A month later, the Wayne County prosecutor dismissed the case against her. The ordeal started with an automated facial recognition search, according to an investigator's report from the Detroit Police Department. Ms. Woodruff is the sixth person to report being falsely accused of a crime as a result of facial recognition technology used by police to match an unknown offender's face to a photo in a database.