The quest for this Holy Grail of a universal Code of ethics in AI has left in its wake a remarkable, if not worrying, quantity of projects aiming to establish a corpus of ethical standards to frame its development. But it is vital that we question the basis on which this corpus is established. And the fast-increasing number of initiatives requiring this tool makes the necessity of ensuring the basis all the more urgent. We must ask two fundamental questions. Is it possible to create one single tool for everything and is there a real widespread desire to create such a tool?

A military drone may have autonomously attacked humans for the first time without being instructed to do so, according to a recent report by the UN Security Council. The report, published in March, claimed that the AI drone – Kargu-2 quadcopter – produced by Turkish military tech company STM, attacked retreating soldiers loyal to Libyan General Khalifa Haftar. The 548-page report by the UN Security Council's Panel of Experts on Libya has not delved into details on if there were any deaths due to the incident, but it raises questions on whether global efforts to ban killer autonomous robots before they are built may be futile. Over the course of the year, the UN-recognized Government of National Accord pushed the Haftar Affiliated Forces (HAF) back from the Libyan capital Tripoli, and the drone may have been operational since January 2020, the experts noted. "Logistics convoys and retreating HAF were subsequently hunted down and remotely engaged by the unmanned combat aerial vehicles or the lethal autonomous weapons systems such as the STM Kargu-2," the UN report noted.

For some, artificial intelligence and data science are fantastic technologies that will benefit people and society. The point is, these latest innovations are evolving fast in the hotbeds of business, science and government, and it can be difficult for regular citizens and civil society to keep up, particularly nonprofits and others involved in addressing the spectrum of society's needs. What is philanthropy's role in the growth, use and regulation of these powerful and protean technologies? These are some of the questions Vilas Dhar considers in his role as president of the Patrick J. McGovern Foundation, a relatively young grantmaker whose late founder built a fortune in publishing and industry research, tracking the expanding computer industry in the late 20th and early 21st centuries. McGovern's International Data Group published a number of popular computer industry magazines such as Computerworld, PC World and InfoWorld.

A "lethal" weaponized drone "hunted down" and "remotely engaged" human targets without its handlers' say-so during a conflict in Libya last year, according to a United Nations report first covered by New Scientist this week. Whether there were any casualties remains unclear, but if confirmed, it would likely be the first recorded death carried out by an autonomous killer robot. In March 2020, a Kargu-2 attack quadcopter, which the agency called a "lethal autonomous weapon system," targeted retreating soldiers and convoys led by Libyan National Army's Khalifa Haftar during a civil conflict with Libyan government forces. "The lethal autonomous weapons systems were programmed to attack targets without requiring data connectivity between the operator and the munition: in effect, a true'fire, forget and find' capability," the UN Security Council's Panel of Experts on Libya wrote in the report. It remains unconfirmed whether any soldiers were killed in the attack, although the UN experts imply as much.

Google has developed a new tool that uses AI to help people identify common skin conditions. Every year, people reach out to Google Search almost 10 billion times to ask questions about skin, nails, and hair. While the information is there, it remains difficult for many to precisely describe the visible symptoms with words alone. Statistics show that over two billion people across the globe are affected by dermatological issues, and there is a global shortage of specialists. That's why Google developed the AI-powered dermatology assist tool – a web-based application that works with the camera on your phone.

Explaining the decisions of an Artificial Intelligence (AI) model is increasingly critical in many real-world, high-stake applications. Hundreds of papers have either proposed new feature attribution methods, discussed or harnessed these tools in their work. However, despite humans being the target end-users, most attribution methods were only evaluated on proxy automatic-evaluation metrics [52, 66, 68]. In this paper, we conduct the first, large-scale user study on 320 lay and 11 expert users to shed light on the effectiveness of state-of-the-art attribution methods in assisting humans in ImageNet classification, Stanford Dogs fine-grained classification, and these two tasks but when the input image contains adversarial perturbations. We found that, in overall, feature attribution is surprisingly not more effective than showing humans nearest training-set examples. On a hard task of fine-grained dog categorization, presenting attribution maps to humans does not help, but instead hurts the performance of human-AI teams compared to AI alone. Importantly, we found automatic attribution-map evaluation measures to correlate poorly with the actual human-AI team performance. Our findings encourage the community to rigorously test their methods on the downstream human-in-the-loop applications and to rethink the existing evaluation metrics.

In this work we study how to learn good algorithms for selecting reasoning steps in theorem proving. We explore this in the connection tableau calculus implemented by leanCoP where the partial tableau provides a clean and compact notion of a state to which a limited number of inferences can be applied. We start by incorporating a state-of-the-art learning algorithm -- a graph neural network (GNN) -- into the plCoP theorem prover. Then we use it to observe the system's behaviour in a reinforcement learning setting, i.e., when learning inference guidance from successful Monte-Carlo tree searches on many problems. Despite its better pattern matching capability, the GNN initially performs worse than a simpler previously used learning algorithm. We observe that the simpler algorithm is less confident, i.e., its recommendations have higher entropy. This leads us to explore how the entropy of the inference selection implemented via the neural network influences the proof search. This is related to research in human decision-making under uncertainty, and in particular the probability matching theory. Our main result shows that a proper entropy regularisation, i.e., training the GNN not to be overconfident, greatly improves plCoP's performance on a large mathematical corpus.

The Neural Machine Translation (NMT) model is essentially a joint language model conditioned on both the source sentence and partial translation. Therefore, the NMT model naturally involves the mechanism of the Language Model (LM) that predicts the next token only based on partial translation. Despite its success, NMT still suffers from the hallucination problem, generating fluent but inadequate translations. The main reason is that NMT pays excessive attention to the partial translation while neglecting the source sentence to some extent, namely overconfidence of the LM. Accordingly, we define the Margin between the NMT and the LM, calculated by subtracting the predicted probability of the LM from that of the NMT model for each token. The Margin is negatively correlated to the overconfidence degree of the LM. Based on the property, we propose a Margin-based Token-level Objective (MTO) and a Margin-based Sentencelevel Objective (MSO) to maximize the Margin for preventing the LM from being overconfident. Experiments on WMT14 English-to-German, WMT19 Chinese-to-English, and WMT14 English-to-French translation tasks demonstrate the effectiveness of our approach, with 1.36, 1.50, and 0.63 BLEU improvements, respectively, compared to the Transformer baseline. The human evaluation further verifies that our approaches improve translation adequacy as well as fluency.

Answering complex questions from long documents requires aggregating multiple pieces of evidence and then predicting the answers. In this paper, we propose a multi-hop retrieval method, DocHopper, to answer compositional questions over long documents. At each step, DocHopper retrieves a paragraph or sentence embedding from the document, mixes the retrieved result with the query, and updates the query for the next step. In contrast to many other retrieval-based methods (e.g., RAG or REALM) the query is not augmented with a token sequence: instead, it is augmented by "numerically" combining it with another neural representation. This means that model is end-to-end differentiable. We demonstrate that utilizing document structure in this was can largely improve question-answering and retrieval performance on long documents. We experimented with DocHopper on three different QA tasks that require reading long documents to answer compositional questions: discourse entailment reasoning, factual QA with table and text, and information seeking QA from academic papers. DocHopper outperforms all baseline models and achieves state-of-the-art results on all datasets. Additionally, DocHopper is efficient at inference time, being 3~10 times faster than the baselines.

Modern neural machine translation (NMT) models have achieved competitive performance in standard benchmarks such as WMT. However, there still exist significant issues such as robustness, domain generalization, etc. In this paper, we study NMT models from the perspective of compositional generalization by building a benchmark dataset, CoGnition, consisting of 216k clean and consistent sentence pairs. We quantitatively analyze effects of various factors using compound translation error rate, then demonstrate that the NMT model fails badly on compositional generalization, although it performs remarkably well under traditional metrics.