Causal explanations of the predictions of NLP systems are essential to ensure safety and establish trust. Yet, existing methods often fall short of explaining model predictions effectively or efficiently and are often model-specific. In this paper, we address model-agnostic explanations, proposing two approaches for counterfactual (CF) approximation. The first approach is CF generation, where a large language model (LLM) is prompted to change a specific text concept while keeping confounding concepts unchanged. While this approach is demonstrated to be very effective, applying LLM at inference-time is costly. We hence present a second approach based on matching, and propose a method that is guided by an LLM at training-time and learns a dedicated embedding space. This space is faithful to a given causal graph and effectively serves to identify matches that approximate CFs. After showing theoretically that approximating CFs is required in order to construct faithful explanations, we benchmark our approaches and explain several models, including LLMs with billions of parameters. Our empirical results demonstrate the excellent performance of CF generation models as model-agnostic explainers. Moreover, our matching approach, which requires far less test-time resources, also provides effective explanations, surpassing many baselines. We also find that Top-K techniques universally improve every tested method. Finally, we showcase the potential of LLMs in constructing new benchmarks for model explanation and subsequently validate our conclusions. Our work illuminates new pathways for efficient and accurate approaches to interpreting NLP systems.

The increased use of AI systems is associated with multi-faceted societal, environmental, and economic consequences. These include non-transparent decision-making processes, discrimination, increasing inequalities, rising energy consumption and greenhouse gas emissions in AI model development and application, and an increasing concentration of economic power. By considering the multi-dimensionality of sustainability, this paper takes steps towards substantiating the call for an overarching perspective on "sustainable AI". It presents the SCAIS Framework (Sustainability Criteria and Indicators for Artificial Intelligence Systems) which contains a set 19 sustainability criteria for sustainable AI and 67 indicators that is based on the results of a critical review and expert workshops. This interdisciplinary approach contributes a unique holistic perspective to facilitate and structure the discourse on sustainable AI. Further, it provides a concrete framework that lays the foundation for developing standards and tools to support the conscious development and application of AI systems.

While Convolutional Neural Networks (CNNs) have been widely successful in 2D human pose estimation, Vision Transformers (ViTs) have emerged as a promising alternative to CNNs, boosting state-of-the-art performance. However, the quadratic computational complexity of ViTs has limited their applicability for processing high-resolution images. In this paper, we propose three methods for reducing ViT's computational complexity, which are based on selecting and processing a small number of most informative patches while disregarding others. The first two methods leverage a lightweight pose estimation network to guide the patch selection process, while the third method utilizes a set of learnable joint tokens to ensure that the selected patches contain the most important information about body joints. Experiments across six benchmarks show that our proposed methods achieve a significant reduction in computational complexity, ranging from 30% to 44%, with only a minimal drop in accuracy between 0% and 3.5%.

Effective and rapid decision-making from randomized controlled trials (RCTs) requires unbiased and precise treatment effect inferences. Two strategies to address this requirement are to adjust for covariates that are highly correlated with the outcome, and to leverage historical control information via Bayes' theorem. We propose a new Bayesian prognostic covariate adjustment methodology, referred to as Bayesian PROCOVA, that combines these two strategies. Covariate adjustment in Bayesian PROCOVA is based on generative artificial intelligence (AI) algorithms that construct a digital twin generator (DTG) for RCT participants. The DTG is trained on historical control data and yields a digital twin (DT) probability distribution for each RCT participant's outcome under the control treatment. The expectation of the DT distribution, referred to as the prognostic score, defines the covariate for adjustment. Historical control information is leveraged via an additive mixture prior with two components: an informative prior probability distribution specified based on historical control data, and a weakly informative prior distribution. The mixture weight determines the extent to which posterior inferences are drawn from the informative component, versus the weakly informative component. This weight has a prior distribution as well, and so the entire additive mixture prior is completely pre-specifiable without involving any RCT information. We establish an efficient Gibbs algorithm for sampling from the posterior distribution, and derive closed-form expressions for the posterior mean and variance of the treatment effect parameter conditional on the weight, in Bayesian PROCOVA. We evaluate efficiency gains of Bayesian PROCOVA via its bias control and variance reduction compared to frequentist PROCOVA in simulation studies that encompass different discrepancies. These gains translate to smaller RCTs.

Sequential Monte Carlo methods can characterize arbitrary distributions using sequential importance sampling and resampling, but typically require very large sample sizes to mitigate weight collapse [Snyder et al., 2008, 2015]. By contrast, ensemble Kalman-type methods avoid the use of weights, but are based on affine prior-to-posterior updates that are consistent only if all distributions involved are Gaussian. In the context of smoothing, such methods include the ensemble Kalman smoother (EnKS) [Evensen and Van Leeuwen, 2000], which has inspired numerous algorithmic variations such as the ensemble smoother with multiple data assimilation [Emerick and Reynolds, 2013] and the iterative ensemble Kalman smoother (iEnKS) [Bocquet and Sakov, 2014, Evensen et al., 2019], as well as backwards smoothers such as the ensemble Rauch-Tung-Striebel smoother (EnRTSS) [Raanes, 2016]. These two classes of methods occupy opposite ends of a spectrum that ranges from an emphasis on statistical generality at one end to an emphasis on computational efficiency at the other. This trade-off complicates design decisions for smoothing problems that are at once non-Gaussian and computationally expensive.

Smoothers are algorithms for Bayesian time series re-analysis. Most operational smoothers rely either on affine Kalman-type transformations or on sequential importance sampling. These strategies occupy opposite ends of a spectrum that trades computational efficiency and scalability for statistical generality and consistency: non-Gaussianity renders affine Kalman updates inconsistent with the true Bayesian solution, while the ensemble size required for successful importance sampling can be prohibitive. This paper revisits the smoothing problem from the perspective of measure transport, which offers the prospect of consistent prior-to-posterior transformations for Bayesian inference. We leverage this capacity by proposing a general ensemble framework for transport-based smoothing. Within this framework, we derive a comprehensive set of smoothing recursions based on nonlinear transport maps and detail how they exploit the structure of state-space models in fully non-Gaussian settings. We also describe how many standard Kalman-type smoothing algorithms emerge as special cases of our framework. A companion paper (Ramgraber et al., 2023) explores the implementation of nonlinear ensemble transport smoothers in greater depth.

Nearly two decades ago, a think-tank in Washington D.C. invited past and present government officials from the CIA, the Defense Intelligence Agency (DIA), the Pentagon and elsewhere to debate the risks of revealing the truth about UFOs. The 2004 event -- according to a former CIA scientist who went public with the shocking story Friday -- broke into working groups to weigh the positive and negative ramifications of declassifying America's top secret UFO programs. Every working group according to that scientist, Dr. Hal Puthoff, came back with the same conclusion: the societal risks of UFO'disclosure' were just too great. But now, a host of Washington insiders are calling for a strategic'campaign' to drag these alleged UFO reverse-engineering programs out into public view. The Sol Foundation, a new nonprofit dedicated to exploring the broad implications of what are now called'Unidentified Anomalous Phenomena' or UAP, convened its first ever symposium Friday, sponsored by Stanford University's School of Medicine Sol's lofty goal, as described by its chief operating officer - the now famous UFO whistleblower and US Air Force and intel veteran David Grusch - is to'open ourselves to a future where truth, unity, technological advancements and a deeper understanding of our existence converge' The pivot emerged this weekend at an invite-only conference of former government officials, tenured physicists and other academic researchers, activists and reporters, held at Stanford University and attended by DailyMail.com. The most explosive moments from the UFO event -- the first ever symposium of the new nonprofit Sol Foundation, which is dedicated to exploring the broad implications of what are now called'Unidentified Anomalous Phenomena' or UAP -- came from recently retired US Army Colonel Karl E. Nell.

'The Big Weekend' panelists discuss rumors around Vice President Kamala Harris's future as predictions mount that former First Lady Michelle Obama could replace President Biden. Former Democratic California Gov. Jerry Brown refused to comment on Vice President Kamala Harris after expressing confidence in President Biden as the "man of the hour," during a recent media interview. While talking about President Biden and overall feelings of the electorate, Brown told NBC News in a recent interview that the president was the "man of the hour." "I would say he's the man of the hour. He's there," Brown said, adding that he didn't have a political strategy for the Democrats.

In an interview, Meta's president of global affairs, Nick Clegg, said the tools "are really quite important" in that they provide, in a lot of ways, "the most comprehensive access to publicly available content across Facebook and Instagram of anything that we've built to date." The Content Library will also help the company meet new regulatory requirements and obligations on data sharing and transparency, as the company notes in a blog post Tuesday. The library and associated API were first released as a beta version several months ago and allow researchers to access near-real-time data about pages, posts, groups, and events on Facebook and creator and business accounts on Instagram, as well as the associated numbers of reactions, shares, comments, and post view counts. While all this data is publicly available--as in, anyone can see public posts, reactions, and comments on Facebook--the new library makes it easier for researchers to search and analyze this content at scale. Meta says that to protect user privacy, this data will be accessible only through a virtual "clean room" and not downloadable.

Canopy CMO Yaron Litwin discusses how criminals are using deepfake technology to blackmail teens and generate child pornography. Today's hot trend for policymakers is talking about artificial intelligence. This incredibly powerful technology is here to stay, and new research shows that most of us are optimistic about how generative AI will be able to improve our lives. But there are some new and concerning threats to which policymakers must pay attention. This includes a horrific misuse of this positive tech: bad actors abusing AI to put real people in sexually explicit situations, including minors.