NLP benchmarks have largely focused on short texts, such as sentences and paragraphs, even though long texts comprise a considerable amount of natural language in the wild. We introduce SCROLLS, a suite of tasks that require reasoning over long texts. We examine existing long-text datasets, and handpick ones where the text is naturally long, while prioritizing tasks that involve synthesizing information across the input. SCROLLS contains summarization, question answering, and natural language inference tasks, covering multiple domains, including literature, science, business, and entertainment. Initial baselines, including Longformer Encoder-Decoder, indicate that there is ample room for improvement on SCROLLS. We make all datasets available in a unified text-to-text format and host a live leaderboard to facilitate research on model architecture and pretraining methods.

A new generation of increasingly autonomous and self-learning systems, which we call embodied systems, is about to be developed. When deploying these systems into a real-life context we face various engineering challenges, as it is crucial to coordinate the behavior of embodied systems in a beneficial manner, ensure their compatibility with our human-centered social values, and design verifiably safe and reliable human-machine interaction. We are arguing that raditional systems engineering is coming to a climacteric from embedded to embodied systems, and with assuring the trustworthiness of dynamic federations of situationally aware, intent-driven, explorative, ever-evolving, largely non-predictable, and increasingly autonomous embodied systems in uncertain, complex, and unpredictable real-world contexts. We are also identifying a number of urgent systems challenges for trustworthy embodied systems, including robust and human-centric AI, cognitive architectures, uncertainty quantification, trustworthy self-integration, and continual analysis and assurance.

While Automatic Speech Recognition has been shown to be vulnerable to adversarial attacks, defenses against these attacks are still lagging. Existing, naive defenses can be partially broken with an adaptive attack. In classification tasks, the Randomized Smoothing paradigm has been shown to be effective at defending models. However, it is difficult to apply this paradigm to ASR tasks, due to their complexity and the sequential nature of their outputs. Our paper overcomes some of these challenges by leveraging speech-specific tools like enhancement and ROVER voting to design an ASR model that is robust to perturbations. We apply adaptive versions of state-of-the-art attacks, such as the Imperceptible ASR attack, to our model, and show that our strongest defense is robust to all attacks that use inaudible noise, and can only be broken with very high distortion.

Artificial intelligence or AI technology has taken over almost every organization, available on this planet. Be it a big company or a start-up, business leaders choose AI technology over any other traditional IT practices, as they believe artificial intelligence or AI will be the key element behind their business' success. As artificial intelligence is gradually gaining popularity in the Indian domestic market, several AI startups in India, have started emerging. According to the AIM Research, AI Start-ups in India successfully raised US$836.3 million in 2020, and in the same year, the Government of India had increased the expenditure for Digital India to US$477 million, to develop artificial intelligence or AI models for the newly emerging artificial intelligence start-ups in India. Several AI start-ups have sprung up in India, but only a few were able to create a mark in the Indian domestic market.

This article is part of a series on AI for Boards of Directors. As AI grows in impact in the business world, the US Government is finally wading in to influence the future of regulation. Businesses have significant challenges effectively operating in a regulatory free-for-all world. Companies actually want some regulation. As mentioned in "Why Are Technology Companies Quitting Facial Recognition?", the providers of AI solutions want federal regulations because, in the absence of leadership at the federal level, states and municipalities will create those regulations.

From January 3rd, 2022, Amazon will be solving most of its packaging issues with the help of AI based machine learning tools. Meaning, the Jeff Bezos led company will be amalgamating computer vision and natural language processing to'guestimate' the right amount of packaging required to pack millions of products it ships to its customers. According to an update released to the media, Amazon expressed that the use of AI tech has reduced the packaging consumption per shipment by over 33% that accounts for 3 million tons of packaging required to prepare over 2 billion different sized boxes. From the year 2019, Amazon tested the Machine Learning model of packaging in its facilities located across the United States and was happy to announce that it was 100% satisfied with the results. To achieve its vision, the American retail giant had to upgrade its packaging and distribution tunnels with some software driven cameras and some sensors.

This article was contributed by Valerias Bangert, strategy and innovation consultant, founder of three media outlets, and published author. The debate around whether AI will automate jobs away is heating up. AI critics claim that these statistical models lack the creativity and intuition of human workers and that they are thus doomed to specific, repetitive tasks. While AI job automation has already replaced around 400,000 factory jobs in the U.S. from 1990 to 2007, with another 2 million on the way, AI today is automating the economy in a much more subtle way. Take the example of writing jobs.

However, it's only really in the last year that we have really seen digital ethics hit the mainstream, with organisations in the private and public sector focusing their attention and, increasingly, resources, on these matters. So, what's caused this shift? Many organisations underwent an overnight transformation during the pandemic to survive and at the heart of this was the accelerated adoption of more advanced technologies such as automation and artificial intelligence (AI). Organisations are now taking a more serious look at what being data-driven means for them, developing data strategies that could shift entire business models. Without integrating digital ethics into this acceleration, the ethical risks proliferate.

WASHINGTON – The Biden administration and its allies are assembling a punishing set of financial, technology and military sanctions against Russia that they say would go into effect within hours of an invasion of Ukraine, hoping to make clear to President Vladimir Putin the high cost he would pay if he sends troops across the border. In interviews, officials described details of those plans for the first time, just before a series of diplomatic negotiations to defuse the crisis with Moscow, one of the most perilous moments in Europe since the end of the Cold War. The talks begin Monday in Geneva and then move across Europe. The plans the United States has discussed with allies in recent days include cutting off Russia's largest financial institutions from global transactions, imposing an embargo on American-made or American-designed technology needed for defense-related and consumer industries, and arming insurgents in Ukraine who would conduct what would amount to a guerrilla war against a Russian military occupation, if it comes to that. Such moves are rarely telegraphed in advance.

It is well known that translations of songs and poems not only breaks rhythm and rhyming patterns, but also results in loss of semantic information. The Bhagavad Gita is an ancient Hindu philosophical text originally written in Sanskrit that features a conversation between Lord Krishna and Arjuna prior to the Mahabharata war. The Bhagavad Gita is also one of the key sacred texts in Hinduism and known as the forefront of the Vedic corpus of Hinduism. In the last two centuries, there has been a lot of interest in Hindu philosophy by western scholars and hence the Bhagavad Gita has been translated in a number of languages. However, there is not much work that validates the quality of the English translations. Recent progress of language models powered by deep learning has enabled not only translations but better understanding of language and texts with semantic and sentiment analysis. Our work is motivated by the recent progress of language models powered by deep learning methods. In this paper, we compare selected translations (mostly from Sanskrit to English) of the Bhagavad Gita using semantic and sentiment analyses. We use hand-labelled sentiment dataset for tuning state-of-art deep learning-based language model known as \textit{bidirectional encoder representations from transformers} (BERT). We use novel sentence embedding models to provide semantic analysis for selected chapters and verses across translations. Finally, we use the aforementioned models for sentiment and semantic analyses and provide visualisation of results. Our results show that although the style and vocabulary in the respective Bhagavad Gita translations vary widely, the sentiment analysis and semantic similarity shows that the message conveyed are mostly similar across the translations.