The Brazilian Supreme Court receives tens of thousands of cases each semester. Court employees spend thousands of hours to execute the initial analysis and classification of those cases -- which takes effort away from posterior, more complex stages of the case management workflow. In this paper, we explore multimodal classification of documents from Brazil's Supreme Court. We train and evaluate our methods on a novel multimodal dataset of 6,510 lawsuits (339,478 pages) with manual annotation assigning each page to one of six classes. Each lawsuit is an ordered sequence of pages, which are stored both as an image and as a corresponding text extracted through optical character recognition. We first train two unimodal classifiers: a ResNet pre-trained on ImageNet is fine-tuned on the images, and a convolutional network with filters of multiple kernel sizes is trained from scratch on document texts. We use them as extractors of visual and textual features, which are then combined through our proposed Fusion Module. Our Fusion Module can handle missing textual or visual input by using learned embeddings for missing data. Moreover, we experiment with bi-directional Long Short-Term Memory (biLSTM) networks and linear-chain conditional random fields to model the sequential nature of the pages. The multimodal approaches outperform both textual and visual classifiers, especially when leveraging the sequential nature of the pages.

Soon after Roe v. Wade was overturned, a neonatal nurse took to a local Ohio newspaper to share how strongly she agreed with the Supreme Court's opinion. Instead of explicitly expressing religious views or personal beliefs, she shared that in her "professional experience" the 1973 cementing of national abortion rights "led to the utter demise of respect for humanity at any lifestage and has, singlehandedly, led to a demise in our societal culture and ethical values." She noted that 99 percent of people seeking abortions are doing so "as a birth control method." The newspaper piece is a startling artifact of the anti-choice movement. The American College of Obstetricians and Gynecologists is firm in its own stance: "Abortion is an essential component of comprehensive, evidence-based health care."

The UK data watchdog is to investigate whether artificial intelligence systems are showing racial bias when dealing with job applications. The Information Commissioner's Office said AI-driven discrimination could have "damaging consequences for people's lives" and lead to someone being rejected for a job or being wrongfully denied a bank loan or a welfare benefit. It will investigate the use of algorithms to sift through job applications, amid concerns that they are affecting employment opportunities for people from ethnic minorities. "We will be investigating concerns over the use of algorithms to sift recruitment applications, which could be negatively impacting employment opportunities of those from diverse backgrounds," said the ICO. The investigation is being announced as part of a three-year plan for the ICO under the UK's new information commissioner, John Edwards, who joined the ICO in January after running its New Zealand counterpart.

Toxic language detection systems often falsely flag text that contains minority group mentions as toxic, as those groups are often the targets of online hate. Such over-reliance on spurious correlations also causes systems to struggle with detecting implicitly toxic language. To help mitigate these issues, we create ToxiGen, a new large-scale and machine-generated dataset of 274k toxic and benign statements about 13 minority groups. We develop a demonstration-based prompting framework and an adversarial classifier-in-the-loop decoding method to generate subtly toxic and benign text with a massive pretrained language model. Controlling machine generation in this way allows ToxiGen to cover implicitly toxic text at a larger scale, and about more demographic groups, than previous resources of human-written text. We conduct a human evaluation on a challenging subset of ToxiGen and find that annotators struggle to distinguish machine-generated text from human-written language. We also find that 94.5% of toxic examples are labeled as hate speech by human annotators. Using three publicly-available datasets, we show that finetuning a toxicity classifier on our data improves its performance on human-written data substantially. We also demonstrate that ToxiGen can be used to fight machine-generated toxicity as finetuning improves the classifier significantly on our evaluation subset. Our code and data can be found at https://github.com/microsoft/ToxiGen.

Yet many of these systems are not accessible to people who experience different kinds of disability exclusion. Different personal characteristics may impact any or all of the physical (DNA, fingerprints, face or retina) and behavioral (gesture, gait, voice) characteristics listed in the RFI as examples of biometric signals. We define disability here in terms of the discriminatory and often systemic problems with available infrastructure's ability to meet the needs of all people [UN 2017, Oliver, 2013). Using this definition, "[biometrics] could either mitigate or amplify disability depending on how they are designed." (Guo, 2019). As Whittaker and colleauges (2019) state, this is not simply a matter of algorithmic accuracy: "...discrimination against people of color, women, and other historically marginalized groups has often been justified by representing these groups as disabled . Thus disability is entwined with, and serves to justify, practices of marginalization." It is critical that we look beyond inclusion to full and fully accommodated participation.

Investigating cooperativity of interlocutors is central in studying pragmatics of dialogue. Models of conversation that only assume cooperative agents fail to explain the dynamics of strategic conversations. Thus, we investigate the ability of agents to identify non-cooperative interlocutors while completing a concurrent visual-dialogue task. Within this novel setting, we study the optimality of communication strategies for achieving this multi-task objective. We use the tools of learning theory to develop a theoretical model for identifying non-cooperative interlocutors and apply this theory to analyze different communication strategies. We also introduce a corpus of non-cooperative conversations about images in the GuessWhat?! dataset proposed by De Vries et al. (2017). We use reinforcement learning to implement multiple communication strategies in this context and find empirical results validate our theory.

Meta has announced that it has built and open-sourced'No Language Left Behind' NLLB-200, a single Artificial Intelligence (AI) model that is the first to translate across 200 different languages, including 55 African languages with state-of-the-art results. Meta is using the modelling techniques and learnings from the project to improve and extend translations on Facebook, Instagram, and Wikipedia. In an effort to develop high-quality machine translation capabilities for most of the world's low-resource languages, this single AI model was designed with a focus on African languages. They are challenging from a machine translation perspective. AI models require lots and lots of data to help them learn, and there's not a lot of human-translated training data for these languages.

We are excited to bring Transform 2022 back in-person July 19 and virtually July 20 - 28. Join AI and data leaders for insightful talks and exciting networking opportunities. Could artificial intelligence (AI) help companies meet growing expectations for environmental, social and governance (ESG) reporting? Certainly, over the past couple of years, ESG issues have soared in importance for corporate stakeholders, with increasing demands from investors, employees and customers. According to S&P Global, in 2022 corporate boards and government leaders "will face rising pressure to demonstrate that they are adequately equipped to understand and oversee ESG issues -- from climate change to human rights to social unrest." ESG investing, in particular, has been a big part of this boom: Bloomberg Intelligence found that ESG assets are on track to exceed $50 trillion by 2025, representing more than a third of the projected $140.5 trillion in total global assets under management.

NASA has released the second set of images taken by the James Webb Space Telescope, revealing galaxies, planets, and stars in unprecedented detail. The new pictures include an incredible study of the atmosphere of a gas giant planet 1,000 light-years from Earth, called WASP-96 b, a dying star throwing off its outer layers, and an exquisite view of Stephan's Quintet, a group of five galaxies about 300 million light-years from Earth. Take a look at them here. The term "artificial intelligence" really has two meanings. AI refers both to the fundamental scientific quest to build human intelligence into computers and to the work of modeling massive quantities of data.

Regulation of data privacy may be changing in Canada, with the proposed creation of a new enforcement body with the power to impose very significant fines. The proposed reform also includes new regulation of the use of artificial intelligence. On 16 June 2022, the Canadian government introduced a proposal for three-pronged legislation to strengthen Canada's data privacy framework and create new regulations for the responsible development of artificial intelligence (AI), while continuing to implement Canada's Digital Charter. The Digital Charter establishes ten key principles for the digital landscape, such as universal access, safety, and control and consent in relation to personal data. The proposal (Bill C-27, the Digital Charter Implementation Act, 2022) would also introduce significant changes to enforcement.