Whether and how data scientists, statisticians and modellers should be accountable for the AI systems they develop remains a controversial and highly debated topic, especially given the complexity of AI systems and the difficulties in comparing and synthesising competing claims arising from their deployment for data analysis. This paper proposes to address this issue by decreasing the opacity and heightening the accountability of decision making using AI systems, through the explicit acknowledgement of the statistical foundations that underpin their development and the ways in which these dictate how their results should be interpreted and acted upon by users. In turn, this enhances (1) the responsiveness of the models to feedback, (2) the quality and meaning of uncertainty on their outputs and (3) their transparency to evaluation. To exemplify this approach, we extend Posterior Belief Assessment to offer a route to belief ownership from complex and competing AI structures. We argue that this is a significant way to bring ethical considerations into mathematical reasoning, and to implement ethical AI in statistical practice. We demonstrate these ideas within the context of competing models used to advise the UK government on the spread of the Omicron variant of COVID-19 during December 2021.

Like other extreme quantities (for example, a distance of 34 parsecs), unfamiliar dollar amounts can be hard to fathom without comparison to something else [7, 18]. To address this issue, it can be useful to employ perspectives: re-phrasings of measurements that make them easier to understand, via a change of units to express the focal number on a different scale, or a comparison to a reference object. For instance, $330 billion can be re-expressed using perspectives of "about $1,000 per person in the United States" or "about 5% of the United States Federal Budget". In addition to being intuitively appealing and perceived as helpful [6, 10, 13], perspectives have been shown to aid numerical comprehension by boosting recall, estimation, error detection, and prediction [3, 12, 27], which could find relevance in a wide variety of downstream applications. These demonstrations of the benefits of perspectives have led to questions around what makes some analogies better than others, and if and how one can generate high-quality perspectives at scale for naturally occurring mentions of measurements. Approaches to automated perspective generation have varied, but they generally rely on first constructing a database of reference objects to compare measurements to and then prioritizing analogies to these reference objects that are both familiar and helpful to the reader [10, 27]. Prioritizing reference objects is complicated by the fact that what is most helpful for understanding a measurement can be difficult to quantify and can depend on the context in which the measurement occurs.

During the 2016 U.S. presidential race, a Russian "troll-farm" calling itself the Internet Research Agency sought to harm Hillary Clinton's election chances and help Donald Trump reach the White House by using Twitter to spread false news stories and other disinformation, according to a 2020 report from the Senate Intelligence Committee. Most of that content apparently was produced by human beings, a supposition supported by the fact that activity dropped off on Russian holidays. Soon, though, if not already, such propaganda will be produced automatically by artificial intelligence (AI) systems such as ChatGPT, a chatbot capable of creating human-sounding text. "Imagine a scenario where you have ChatGPT generating these tweets. The number of fake accounts you could manage for the same price would be much larger," says V.S. Subrahmanian, a professor of computer science at Northwestern University, whose research focuses on the intersection of AI and security problems.

When artificial intelligence software like ChatGPT writes, it considers many options for each word, taking into account the response it has written so far and the question being asked. It assigns a score to each option on the list, which quantifies how likely the word is to come next, based on the vast amount of human-written text it has analyzed. ChatGPT, which is built on what is known as a large language model, then chooses a word with a high score, and moves on to the next one. The model's output is often so sophisticated that it can seem like the chatbot understands what it is saying -- but it does not. Every choice it makes is determined by complex math and huge amounts of data.

AI-generated text, from tools like ChatGPT, is starting to impact daily life. Teachers are testing it out as part of classroom lessons. Marketers are champing at the bit to replace their interns. Memers are going buck wild. It would be a lie to say I'm not a little anxious about the robots coming for my writing gig.

The hottest startup in Silicon Valley right now is OpenAI, the Microsoft-backed developer of ChatGPT, a much-hyped chatbot that can write a poem, college essay or even a line of software code. Tesla tycoon Elon Musk was an early investor in OpenAI and Microsoft is reported to be in talks to up an initial investment of $1 billion to $10 billion in a goal to challenge Google's world-dominating search engine. If agreed, the cash injection by the Windows-maker would value OpenAI at a whopping $29 billion, making it a rare tech-world success when major players such as Amazon, Meta and Twitter are cutting costs and laying off staff. "Microsoft is clearly being aggressive on this front and not going to be left behind on what could be a potential game-changing AI investment," said analyst Dan Ives of Wedbush Securities. Before the release of ChatGPT, OpenAI had wowed tech geeks with Dall-E 2, a software that creates digital images with a simple instruction.

From medtech to fintech, Artificial Intelligence (AI) is rapidly transforming our world, with the potential to revolutionise industries and fundamentally alter how we live our lives. Whether it be self-driving cars or smart homes, AI has infiltrated nearly every aspect of modern life – and its influence is growing every day. In recent years, AI tools have become more sophisticated and are now being used to help individuals and organisations in various ways. One such tool that has developed a buzz in recent times on the Internet is none other than ChatGPT, developed by OpenAI. The tool has quickly grabbed attention for its detailed responses and articulate answers across many knowledge domains.

We introduce ShortcutGen, a new data poisoning attack that generates sample-dependent, error-minimizing perturbations by learning a generator. The key novelty of ShortcutGen is the use of a randomly-initialized discriminator, which provides spurious shortcuts needed for generating poisons. Different from recent, iterative methods, our ShortcutGen can generate perturbations with only one forward pass in a label-free manner, and compared to the only existing generative method, DeepConfuse, our ShortcutGen is faster and simpler to train while remaining competitive. We also demonstrate that integrating a simple augmentation strategy can further boost the robustness of ShortcutGen against early stopping, and combining augmentation and non-augmentation leads to new state-of-the-art results in terms of final validation accuracy, especially in the challenging, transfer scenario. Lastly, we speculate, through uncovering its working mechanism, that learning a more general representation space could allow ShortcutGen to work for unseen data.

Computer models are widely used across a range of scientific disciplines to describe various complex physical systems, however to perform full uncertainty quantification we often need to employ emulators. An emulator is a fast statistical construct that mimics the slow to evaluate computer model, and greatly aids the vastly more computationally intensive uncertainty quantification calculations that an important scientific analysis often requires. We examine the problem of emulating computer models that possess multiple, partial discontinuities occurring at known non-linear location. We introduce the TENSE framework, based on carefully designed correlation structures that respect the discontinuities while enabling full exploitation of any smoothness/continuity elsewhere. This leads to a single emulator object that can be updated by all runs simultaneously, and also used for efficient design. This approach avoids having to split the input space into multiple subregions. We apply the TENSE framework to the TNO Challenge II, emulating the OLYMPUS reservoir model, which possess multiple such discontinuities.

Counting the number of shortest paths on a grid is a simple procedure with close ties to Pascal's triangle. We show how path counting can be used to select relatively direct grid paths for AI-related applications involving navigation through spatial environments. Typical implementations of Dijkstra's algorithm and A* prioritize grid moves in an arbitrary manner, producing paths which stray conspicuously far from line-of-sight trajectories. We find that by counting the number of paths which traverse each vertex, then selecting the vertices with the highest counts, one obtains a path that is reasonably direct in practice and can be improved by refining the grid resolution. Central Dijkstra and Central A* are introduced as the basic methods for computing these central grid paths. Theoretical analysis reveals that the proposed grid-based navigation approach is related to an existing grid-based visibility approach, and establishes that central grid paths converge on clear sightlines as the grid spacing approaches zero. A more general property, that central paths converge on direct paths, is formulated as a conjecture.