A bill that aims to essentially ban law enforcement and intelligence agencies from buying data from Clearview AI has drawn bipartisan support from 20 senators. If the Fourth Amendment is Not For Sale Act were to pass as is, agencies wouldn't be able to buy location data from third-party brokers without a warrant. The bill would prevent agencies from purchasing data on people in the US and Americans outside of the country if the information was procured from "a user's account or device, or via deception, hacking, violations of a contract, privacy policy, or terms of service," according to Senator Ron Wyden's office. If the bill becomes law, Clearview AI would no longer be able to sell much of the data it has obtained to US government agencies. To power its facial recognition technology, Clearview AI reportedly scraped billions of images from social media platforms without consent.

As governments around the world consider how to regulate AI, the European Union is planning first-of-its-kind legislation that would put strict limits on the technology. On Wednesday, the European Commission, the body's executive branch, detailed a regulatory approach that calls for a four-tier system that groups AI software into separate risk categories and applies an appropriate level of regulation to each. At the top would be systems that pose an "unacceptable" risk to people's rights and safety. The EU would outright ban these types of algorithms under the Commission's proposed legislation. An example of software that would fall under this category is any AI that would allow governments and companies to implement social scoring systems.

This opinion piece is forthcoming in Il Sole 24 Ore. The Muller-Lyer optical illusion consists of two lines of equal length that differ only in the direction of arrowheads at either end. Yet, to most observers, the line with arrowheads pointing outwards looks longer than the other. If you grew up in and among buildings with straight walls and 90 degree angles, you have learned to perceive lines according to geometric patterns. Your view of the Muller-Lyer lines is biased.

The European Union's executive arm proposed a bill that would limit police use of facial-recognition software in public and ban the marketing or use of certain kinds of AI systems, in one of the broadest efforts yet to regulate high-stakes applications of artificial intelligence. The bill proposed on Wednesday would also create a list of so-called high-risk uses of AI that would be subject to new supervision and standards for their development and use, such as critical infrastructure, college admissions and loan applications. Regulators could fine a company up to 6% of its annual world-wide revenue for the most severe violations, though in practice EU officials rarely if ever mete out their maximum fines. The bill is one of the broadest of its kind to be proposed by a Western government, and part of the EU's expansion of its role as a global tech enforcer. In recent years, the EU has sought to take a global lead in drafting and enforcing new regulations aimed at taming the alleged excesses of big tech companies and curbing potential dangers of new technologies, in areas ranging from digital competition to online-content moderation.

BRUSSELS (AFP) - The European Union unveils a plan on Wednesday (April 21) to regulate the sprawling field of artificial intelligence, aimed at making Europe a leader in the new tech revolution while reassuring the public against Big Brother-like abuses. "Whether it's precision farming in agriculture, more accurate medical diagnosis or safe autonomous driving, artificial intelligence will open up new worlds for us. But this world also needs rules," European Commission President Ursula von der Leyen said in her state-of-the-union speech in September last year. "We want a set of rules that puts people at the centre." The Commission, the EU's executive arm, has been preparing the proposal for over a year and a debate involving the European Parliament and 27 member states is to go on for months more before a definitive text is in force.

The European Union's executive arm proposed a bill that would limit police use of facial-recognition software in public and ban the marketing or use of certain kinds of AI systems, in one of the broadest efforts yet to regulate high-stakes applications of artificial intelligence. The bill proposed on Wednesday would also create a list of so-called high-risk uses of AI that would be subject to new supervision and standards for their development and use, such as critical infrastructure, college admissions and loan applications. Regulators could fine a company up to 6% of its annual world-wide revenue for the most severe violations, though in practice EU officials rarely if ever mete out their maximum fines. The bill is one of the broadest of its kind to be proposed by a Western government, and part of the EU's expansion of its role as a global tech enforcer. In recent years, the EU has sought to take a global lead in drafting and enforcing new regulations aimed at taming the alleged excesses of big tech companies and curbing potential dangers of new technologies, in areas ranging from digital competition to online-content moderation.

It's "misleading and counterproductive" to block the use of machine-learning algorithms in the justice system on the grounds that some of them may be subject to racial bias, according to a forthcoming study in the American Criminal Law Review. The use of artificial intelligence by judges, prosecutors, police and other justice authorities remains "the best means to overcome the pervasive bias and discrimination that exists in all parts of the deeply flawed criminal justice system," said the study. Algorithmic systems are used in a variety of ways in the U.S. justice system in practices ranging from identifying and predicting crime "hot spots" to real-time surveillance. More than 60 kinds of risk assessment tools are currently in use by court systems around the country, usually to weigh whether individuals should be held in detention before trial or can be released on their own recognizance. The risk assessment tools, which assign weights to data points such as previous arrests and the age of the offender, have come under fire from activists, judges, prosecutors, and some criminologists who say they are susceptible to bias themselves.

Artificial intelligence has the power to change the world -- and it, as it advances from useful to essential, then from essential to mandatory and finally from mandatory to directive. "Machines will tell us what to do," warns entrepreneur and author Bill Bishop. Gillian K. Hadfield, the director of the Schwartz Reisman Institute for Technology and Society, agrees AI will present new problems but is confident society will be bold and come up with new ideas for regulating AI. To paraphrase Tolstoy, all nice, helpful robots are alike, but all dangerous robots are dangerous in their own unique way. In this debate, I will show that we need to fear the dangerous robots (AIs), not by acting like paranoid Luddites, but to ensure we fear AIs enough to pay attention to potential threats, and to take proactive steps to mitigate them.

Why can the Expectation-Maximization (EM) algorithm for mixture models converge? Why can different initial parameters cause various convergence difficulties? The Q-L synchronization theory explains that the observed data log-likelihood L and the complete data log-likelihood Q are positively correlated; we can achieve maximum L by maximizing Q. According to this theory, the Deterministic Annealing EM (DAEM) algorithm's authors make great efforts to eliminate locally maximal Q for avoiding L's local convergence. However, this paper proves that in some cases, Q may and should decrease for L to increase; slow or local convergence exists only because of small samples and unfair competition. This paper uses marriage competition to explain different convergence difficulties and proposes the Fair Competition Principle (FCP) with an initialization map for improving initializations. It uses the rate-verisimilitude function, extended from the rate-distortion function, to explain the convergence of the EM and improved EM algorithms. This convergence proof adopts variational and iterative methods that Shannon et al. used for analyzing rate-distortion functions. The initialization map can vastly save both algorithms' running times for binary Gaussian mixtures. The FCP and the initialization map are useful for complicated mixtures but not sufficient; we need further studies for specific methods.

With increasingly more data and computation involved in their training, machine learning models constitute valuable intellectual property. This has spurred interest in model stealing, which is made more practical by advances in learning with partial, little, or no supervision. Existing defenses focus on inserting unique watermarks in a model's decision surface, but this is insufficient: the watermarks are not sampled from the training distribution and thus are not always preserved during model stealing. In this paper, we make the key observation that knowledge contained in the stolen model's training set is what is common to all stolen copies. The adversary's goal, irrespective of the attack employed, is always to extract this knowledge or its by-products. This gives the original model's owner a strong advantage over the adversary: model owners have access to the original training data. We thus introduce $dataset$ $inference$, the process of identifying whether a suspected model copy has private knowledge from the original model's dataset, as a defense against model stealing. We develop an approach for dataset inference that combines statistical testing with the ability to estimate the distance of multiple data points to the decision boundary. Our experiments on CIFAR10, SVHN, CIFAR100 and ImageNet show that model owners can claim with confidence greater than 99% that their model (or dataset as a matter of fact) was stolen, despite only exposing 50 of the stolen model's training points. Dataset inference defends against state-of-the-art attacks even when the adversary is adaptive. Unlike prior work, it does not require retraining or overfitting the defended model.