Artificial intelligence (AI) systems will increasingly be used to cause harm as they grow more capable. In fact, AI systems are already starting to be used to automate fraudulent activities, violate human rights, create harmful fake images, and identify dangerous toxins. To prevent some misuses of AI, we argue that targeted interventions on certain capabilities will be warranted. These restrictions may include controlling who can access certain types of AI models, what they can be used for, whether outputs are filtered or can be traced back to their user, and the resources needed to develop them. We also contend that some restrictions on non-AI capabilities needed to cause harm will be required. Though capability restrictions risk reducing use more than misuse (facing an unfavorable Misuse-Use Tradeoff), we argue that interventions on capabilities are warranted when other interventions are insufficient, the potential harm from misuse is high, and there are targeted ways to intervene on capabilities. We provide a taxonomy of interventions that can reduce AI misuse, focusing on the specific steps required for a misuse to cause harm (the Misuse Chain), and a framework to determine if an intervention is warranted. We apply this reasoning to three examples: predicting novel toxins, creating harmful images, and automating spear phishing campaigns.

We present a hierarchical Bayesian learning approach to infer jointly sparse parameter vectors from multiple measurement vectors. Our model uses separate conditionally Gaussian priors for each parameter vector and common gamma-distributed hyper-parameters to enforce joint sparsity. The resulting joint-sparsity-promoting priors are combined with existing Bayesian inference methods to generate a new family of algorithms. Our numerical experiments, which include a multi-coil magnetic resonance imaging application, demonstrate that our new approach consistently outperforms commonly used hierarchical Bayesian methods.

Online discourse covers a wide range of topics and many actors tailor their content to impact online discussions through carefully crafted messages and targeted campaigns. Yet the scale and diversity of online media content make it difficult to evaluate the impact of a particular message. In this paper, we present a new technique that leverages semantic similarity to quantify the change in the discussion after a particular message has been published. We use a set of press releases from environmental organisations and tweets from the climate change debate to show that our novel approach reveals a heavy-tailed distribution of response in online discourse to strategic communications.

With the Proposal for a Regulation laying down harmonised rules on Artificial Intelligence (AI Act) the European Union provides the first regulatory document that applies to the entire complex of AI systems. While some fear that the regulation leaves too much room for interpretation and thus bring little benefit to society, others expect that the regulation is too restrictive and, thus, blocks progress and innovation, as well as hinders the economic success of companies within the EU. Without a systematic approach, it is difficult to assess how it will actually impact the AI landscape. In this paper, we suggest a systematic approach that we applied on the initial draft of the AI Act that has been released in April 2021. We went through several iterations of compiling the list of AI products and projects in and from Germany, which the Lernende Systeme platform lists, and then classified them according to the AI Act together with experts from the fields of computer science and law. Our study shows a need for more concrete formulation, since for some provisions it is often unclear whether they are applicable in a specific case or not. Apart from that, it turns out that only about 30\% of the AI systems considered would be regulated by the AI Act, the rest would be classified as low-risk. However, as the database is not representative, the results only provide a first assessment. The process presented can be applied to any collections, and also repeated when regulations are about to change. This allows fears of over- or under-regulation to be investigated before the regulations comes into effect.

Many natural language processing (NLP) tasks rely on labeled data to train machine learning models to achieve high performance. However, data annotation can be a time-consuming and expensive process, especially when the task involves a large amount of data or requires specialized domains. Recently, GPT-3.5 series models have demonstrated remarkable few-shot and zero-shot ability across various NLP tasks. In this paper, we first claim that large language models (LLMs), such as GPT-3.5, can serve as an excellent crowdsourced annotator by providing them with sufficient guidance and demonstrated examples. To make LLMs to be better annotators, we propose a two-step approach, 'explain-then-annotate'. To be more precise, we begin by creating prompts for every demonstrated example, which we subsequently utilize to prompt a LLM to provide an explanation for why the specific ground truth answer/label was chosen for that particular example. Following this, we construct the few-shot chain-of-thought prompt with the self-generated explanation and employ it to annotate the unlabeled data. We conduct experiments on three tasks, including user input and keyword relevance assessment, BoolQ and WiC. The annotation results from GPT-3.5 surpasses those from crowdsourced annotation for user input and keyword relevance assessment. Additionally, for the other two tasks, GPT-3.5 achieves results that are comparable to those obtained through crowdsourced annotation.

This budget represents a maximal privacy violation that any user is willing to face by contributing their data to the training set. We argue that this approach is limited because different users may have different privacy expectations. Thus, setting a uniform privacy budget across all points may be overly conservative for some users or, conversely, not sufficiently protective for others. In this paper, we capture these preferences through individualized privacy budgets. To demonstrate their practicality, we introduce a variant of Differentially Private Stochastic Gradient Descent (DP-SGD) which supports such individualized budgets. DP-SGD is the canonical approach to training models with differential privacy. We modify its data sampling and gradient noising mechanisms to arrive at our approach, which we call Individualized DP-SGD (IDP-SGD). Because IDP-SGD provides privacy guarantees tailored to the preferences of individual users and their data points, we find it empirically improves privacy-utility trade-offs.

This image taken by NASA's Hubble Space Telescope shows a spiral galaxy (bottom left) in front of a large galaxy cluster. New research leveraged an artificial tool to estimate the masses of galaxy clusters more accurately. Astronomers at the Institute for Advanced Study and the Flatiron Institute, along with their collaborators, have utilized artificial intelligence to improve the method of calculating the mass of massive clusters of galaxies. The AI revealed that by incorporating a simple term into an existing equation, researchers can now achieve much more accurate mass estimates than before. The newly enhanced calculations will allow scientists to determine the basic characteristics of the universe with greater precision, according to a report by the astrophysicists, which was published in the Proceedings of the National Academy of Sciences.

The United States is no stranger to technological arms races, having spent much of the Cold War in a two-pronged one-upmanship effort against the Soviet Union to build bigger rockets and land those on the moon. Its rivalry now is with China, and the latest battleground is artificial intelligence. China has long sought to dominate the AI landscape, laying out a plan to become a "global leader" in the sector by 2030 and pledging billions of state dollars for research and development. U.S. breakthroughs have been more organic, illustrated most recently by the rapid global uptake of chatbots made by American companies, such as Google, Microsoft, and OpenAI, with Chinese counterparts largely playing catch-up. One fundamental distinction is the private sector's role at the forefront of developing new AI capabilities: There were no private-built rockets in the 1960s.

Just when the conversation was starting to get good and we approached borderline subjects, we repeatedly found ourselves back at square one. Even simple requests for facts about China's government or top leader Xi led it to terminate the exchange with a canned reply about being an AI that was still learning, and a link to begin a new conversation -- making conversation with Ernie Bot less smooth than with ChatGPT.

The Government Exchange Commission is cautioning customers about the most recent innovation being utilized in extortion plans. Well-informed lawbreakers are utilizing man-made brainpower (artificial intelligence) to deceive individuals by utilizing the voices of friends and family. Hoodlums record and clone voices utilizing words and expressions of an individual, then, at that point, call that individual's loved ones letting them know they're in a difficult situation and need cash immediately. They likewise use ridiculing, an innovation that permits fraudsters to copy a telephone number, email, or website page, successfully fooling casualties into accepting they're being reached by somebody they know or a genuine source. Never pay or send cash in manners that make it hard to get your cash back like wiring cash, sending digital currency, or purchasing present cards and giving them the card numbers and PINs.