In the United Kingdom there's been some landmark AI news recently involving government use of the technology. First, use of facial recognition by South Wales Police was ruled unlawful by a Court of Appeal judge in part for violating privacy, human rights, and failure by police to verify the tech did not exhibit race or gender bias. How the U.K. treats facial recognition is important since London has more CCTV cameras than any major city outside of China. Then, U.K. government officials used an algorithm that ended up benefiting kids who go to private schools and downgrading students from disadvantaged backgrounds. Prime Minister Boris Johnson defended the algorithm grading results as "robust" and "dependable for employers."

Artificial intelligence is becoming a major feature of educational practice and policymaking, but researchers are beginning to raise critical questions about its ethics and effects. Artificial Intelligence (AI) has become the subject of both hype and horror in education. During the 2020 Covid-19 pandemic, AI in education (AIed) attracted serious investor interest, market speculation, and enthusiastic technofuturist predictions. At the same time, algorithms and statistical models were implicated in several major controversies over predictive grading based on historical performance data, raising serious questions about privileging data-driven assessment over teacher judgment. In the new special issue AI in education: Critical perspectives and alternative futures published in Learning, Media and Technology, Rebecca Eynon and I pulled together a collection of cutting edge social scientific analyses of AIed.

"With great power, comes great responsibility" I remember hearing this dialogue in the Spider-Man movie where Uncle Ben schools a young Peter Parker on the travails of wielding power in life. In the context of Artificial Intelligence, this phrase could not be more relevant. Artificial Intelligence is widely being heralded as General Purpose Technology by many economists – one with a range of characteristics that make it poised to generate long-term productivity and economic growth. The onus of bailing out economies out of turmoil and provide a foundation for a new world order falls on AI, in these COVID times especially. This makes the debate around ethical AI is extremely pertinent today. How can AI be ethical?

In 2017, U.S. state trial courts received a gastronomical 83 million court cases. The Chinese Civil Law system sees over 19 million cases per year, with only 120,000 judges to rule over them. In the OECD area (consisting of most high-income economies), the average length for civil proceedings is 240 days in the first instance; the final disposition of cases often involves a long process of appeals, which in some countries can go up to 7 years. It's no secret that the judiciary system in many countries is long, tedious, slow, and can cause months of misery, pain, and anxiety to individuals, families, corporations, and litigators. Moreover, when cases do see the light of day in court, the outcome is not always satisfactory, with high-profile cases especially receiving criticism for being plagued by judge biases' and personal preferences. Scholarly research suggests that in the United States, judges' personal backgrounds, professional experiences, life experiences, and partisan ideologies might impact their decision-making.

Israel will allow autonomous vehicle driving without the presence of a human driver as required nowadays, said a statement issued by Israeli Ministry of Transport on Thursday. The ministry said that a new planned legislation will allow advanced tests as well as free or paid passenger rides in autonomous vehicles via a dedicated application. "This is a huge step forward in promoting autonomous vehicles in Israel," the ministry said in its statement. So far, all testing of autonomous vehicles in Israel are carried out with a safety driver to take over the vehicle in case of emergency. "The new legislation will advance activities of leading technology companies in Israel, as they will be able to conduct advanced testing and operate autonomous vehicles," according to the ministry.

Classifiers deployed in high-stakes real-world applications must output calibrated confidence scores, i.e. their predicted probabilities should reflect empirical frequencies. Recalibration algorithms can greatly improve a model's probability estimates; however, existing algorithms are not applicable in real-world situations where the test data follows a different distribution from the training data, and privacy preservation is paramount (e.g. protecting patient records). We introduce a framework that abstracts out the properties of recalibration problems under differential privacy constraints. This framework allows us to adapt existing recalibration algorithms to satisfy differential privacy while remaining effective for domain-shift situations. Guided by our framework, we also design a novel recalibration algorithm, accuracy temperature scaling, that outperforms prior work on private datasets. In an extensive empirical study, we find that our algorithm improves calibration on domain-shift benchmarks under the constraints of differential privacy. On the 15 highest severity perturbations of the ImageNet-C dataset, our method achieves a median ECE of 0.029, over 2x better than the next best recalibration method and almost 5x better than without recalibration.

Learning algorithms trained on large amounts of data are increasingly adopted in applications with significant individual and social consequences such as selecting loan applicants, filtering resumes of job applicants, estimating the likelihood for a defendant to commit future crimes, or deciding where to deploy police officers. Analyzing the risk of bias in these systems is therefore crucial. In fact, that is also critical for seemingly less socially consequential applications such as ads placement, recommendation systems, speech recognition, and many other common applications of machine learning. Such biases can appear due to the way the training data has been collected, due to an improper choice of the loss function optimized, or as a result of some other algorithmic choices.

The July/August issue of Costs Lawyer is now available here. Topics covered include whether case and costs management hearings should be separated; the reported and unreported rulings of the Senior Costs Judge in Marbrow; termination clauses in damages-based agreements; and costs orders in the employment tribunal. Costs Lawyer Carl Bird has launched a new firm, Agile Costs, which aims to "bring legal costs into the 21st century" by automating the drafting of Precedent H and Precedent S. Mr Bird has worked in legal costs since 2010 and qualified as a Costs Lawyer in 2017, working in-house at Pinsent Masons before leaving to set up Agile. He explained how he has been "exploring technological solutions in legal costs for the past four years" and learnt to code so he could build the system himself. He said: "Two things have struck me as significant in the work I have undertaken: the sheer amount of time costs draftsmen and Costs Lawyers spend in restructuring data when preparing costs pleadings, and how incredibly useful historic costs data could be for the prediction of future costs."

The social media company had in July raised its settlement offer by $100 million to $650 million in relation to the lawsuit, in which Illinois users accused it of violating the U.S. state's Biometric Information Privacy Act. The revised settlement agreement resolved the court's concerns, leading to the preliminary approval of the class action settlement, Judge James Donato wrote in an order filed in the U.S. District Court for the Northern District of California. "Preliminary approval of the amended stipulation of class action settlement, Dkt. No. 468, is granted, and a final approval hearing is set for January 7, 2021," the judge said in the eight-page order. Facebook allegedly violated the state's law through its "Tag Suggestions" feature, which allowed users to recognize their Facebook friends from previously uploaded photos, according to the lawsuit, which began in 2015.

Lawyers were trained to think that practicing law is a zero-sum game, with no space for collaboration. Some recent events gave me hope that this is no longer the case. I know I am biased, because in the past I've been an open innovation manager in a global corporation, but even now that I'm working on legal technology adoption for a big law firm I think we might have a good chance. I recently attended a working breakfast in Milan (Italy), organized by Luminance, a leading contract review company, which was the first in a sequence of similar events. On that occasion, several partners and innovation heads from big Italian law firms openly discussed ideas and best practices on the matter.