Neural networks leverage both causal and correlation-based relationships in data to learn models that optimize a given performance criterion, such as classification accuracy. This results in learned models that may not necessarily reflect the true causal relationships between input and output. When domain priors of causal relationships are available at the time of training, it is essential that a neural network model maintains these relationships as causal, even as it learns to optimize the performance criterion. We propose a causal regularization method that can incorporate such causal domain priors into the network and which supports both direct and total causal effects. We show that this approach can generalize to various kinds of specifications of causal priors, including monotonicity of causal effect of a given input feature or removing a certain influence for purposes of fairness. Our experiments on eleven benchmark datasets show the usefulness of this approach in regularizing a learned neural network model to maintain desired causal effects. On most datasets, domain-prior consistent models can be obtained without compromising on accuracy.

Artificial intelligence (AI) is often touted as the most exciting technology of our age, promising to transform our economies, lives, and capabilities. Some even see AI as making steady progress towards the development of'intelligence machines' that will soon surpass human skills in most areas. AI has indeed made rapid advances over the last decade or so, especially owing to the application of modern statistical and machine learning techniques to huge unstructured data sets. It has already influenced almost all industries: AI algorithms are now used by all online platforms and in industries that range from manufacturing to health, finance, wholesale, and retail. Government agencies have also started relying on AI, particularly in the criminal justice system and in customs and immigration control.

Who goes to a demonstration with a high-powered assault weapon? He planned on bringing his high-powered assault weapon to that demonstration with every intention of using it. He claims it was self-defense -- it was not. He was out for blood, his goal was to shoot and kill as many as possible and claim it was self-defense. Absolutely relieved that the jury in this case based their deliberations and ultimate not guilty verdict on the facts and didn't cower to the obvious intended intimidation of BLM, Antifa, the left-leaning liberal loonies, and last but not least, the lying fake media.

Automatic summarization of legal texts is an important and still a challenging task since legal documents are often long and complicated with unusual structures and styles. Recent advances of deep models trained end-to-end with differentiable losses can well-summarize natural text, yet when applied to the legal domain, they show limited results. In this paper, we propose to use reinforcement learning to train current deep summarization models to improve their performance in the legal domain. To this end, we adopt proximal policy optimization methods and introduce novel reward functions that encourage the generation of candidate summaries satisfying both lexical and semantic criteria. We apply our method to training different summarization backbones and observe a consistent and significant performance gain across three public legal datasets.

This piece was a finalist for the inaugural Gradient Prize. Machine Learning is a powerful technique to automatically learn models from data that have recently been the driving force behind several impressive technological leaps such as self-driving cars, robust speech recognition, and, arguably, better-than-human image recognition. We rely on these machine learning models daily; they influence our lives in ways we did not expect, and they are only going to become even more ubiquitous. Consider a couple of example machine learning models: 1) Detecting cats in images 2) Deciding which ads to show you online 3) Predicting which areas will suffer crime, and 4) Predicting how likely a criminal is to re-offend. The first two seem harmless enough.

In the series "Breaking down the EU AI regulations" so far we've talked about the AI regulations and specifically what the EU did on this topic. To recap, AI regulations in the EU are currently governed mostly by frameworks and rules which are still only proposals and are not yet in force. The companies that are somehow connected to the creation and use of AI will have to let their users know about their use of AI systems (or AI algorithms) and have to be able to explain why their AI model makes certain decisions. There is an AI Act about rules to put new AI-enabled systems into the market. This includes what AI systems cannot be marketed, what information should be published so that users may know about the AI systems and their operational capabilities, what technical requirements are there, requirements for datasets that are used to train AI systems, and also requirements after the AI systems are put in circulation.

Now too is the legislation proposing to regulate it. Earlier this year, the European Union outlined its proposed artificial intelligence legislation and gathered feedback from hundreds of companies and organizations. The European Commission closed the consultation period in August, and next comes further debate in the European Parliament. As well as banning some uses outright (facial recognition for identification in public spaces and social "scoring," for instance), its focus is on regulation and review, especially for AI systems deemed "high risk" -- those used in education or employment decisions, say. Any company with a software product deemed high risk will require a Conformité Européenne (CE) badge to enter the market.

More and more AI is present in our everyday lives, tech companies are using the huge amount of data available to them to make better predictions, track our behaviour and offer services they think we will use. As AI is being used in everything nowadays it begs the question of who is responsible for the decisions AI makes? What happens when AI systems fail and kill or harm someone? Can the AI system be held criminally liable for its actions? Criminal liability usually requires an action and a mental intent.

The Center for Innovation, Administration and Research in the Judiciary (CIAPJ) of the Getulio Vargas Foundation (FGV) released the report of the first phase of the research "Technology applied to conflict resolution in the Brazilian Judiciary " (click here to obtain the PDF document) carried out in December 2020. This research was coordinated by the Minister of the Superior Court of Justice (STJ) Luis Felipe Salomão. The research covered 3 of the 5 branchs of the Brazilian Judiciary: State Justice, Labor Justice, Federal Justice, Electoral Justice and Military Justice. The collection of these data was carried out with 59 (fifty-nine) courts (Federal Supreme Court -- STF, Superior Court of Justice -- STJ, Superior Labor Court -- TST, Regional Labor Courts, Federal Regional Courts and Courts of Justice) and the National Council of Justice. The report indicates that half of the courts have an artificial intelligence project under development or already implemented.

Employee rostering is a process of assigning available employees to open shifts. Automating it has ubiquitous practical benefits for nearly all industries, such as reducing manual workload and producing flexible, high-quality schedules. In this work, we develop a hybrid methodology which combines Mixed-Integer Linear Programming (MILP) with scatter search, an evolutionary algorithm, having as use case the optimization of employee rostering for Swissgrid, where it is currently a largely manual process. The hybrid methodology guarantees compliance with labor laws, maximizes employees' preference satisfaction, and distributes workload as uniformly as possible among them. Above all, it is shown to be a robust and efficient algorithm, consistently solving realistic problems of varying complexity to near-optimality an order of magnitude faster than an MILP-alone approach using a state-of-the-art commercial solver. Several practical extensions and use cases are presented, which are incorporated into a software tool currently being in pilot use at Swissgrid.