An internal report on Artificial Intelligence recently approved by a special committee of the European Parliament embodies a push from EU lawmakers and member states to make regulation on artificial intelligence less burdensome and more innovation-friendly. Christian Democrat MEP Axel Voss has been leading the charge against "overburdening" companies with excessive regulation, arguing that the EU regulatory environment should leave more room for innovation. That was the underlying motive of an own-initiative report on Artificial Intelligence in a Digital Age, recently approved in the AIDA committee, a parliamentary body set up in 2020, under Voss' leadership. "We need a better regulatory framework that learns also from the mistakes of the GDPR," Voss said while presenting the report. Instead of overburdening companies, the AI Act should give clear guidance and should leave space for innovation, he added.

An AI tool can quickly suggest possible candidates for the chemical structures of psychoactive "designer drugs" from a simple analysis. "Our method could cut down the time required to identify a new designer drug from weeks or months to just hours," says Michael Skinnider at the University of British Columbia in Canada. Skinnider and his colleagues created a machine learning tool called DarkNPS by training it with chemical structures of around 1700 known designer drugs, collected from forensic labs around the world. The training set included tandem mass spectrometry results for each drug, which is a common technique that provides information on the mass of a molecule and the elements it contains. This allowed the AI to identify patterns between tandem mass spectrometry data and chemical structures.

"A job advertised online today attracts dozens to hundreds to thousands of applications, mainly due to the amplifying effect of the internet," said Laurie Cumbo, Democratic majority leader of the council and sponsor of the bill. In response, she said, employers have adopted software that sifts through applications rapidly and makes recommendations through an "opaque" process. "My bill will require audits that will clear the fog around these processes and help limit implicit bias and discrimination in the hiring process," Cumbo said. Harvard Business School surveyed 2,250 executives earlier this year and found that 90% of companies were using software to screen applicants, even as a majority of those firms admitted that the screening process eliminated qualified candidates. That happens for reasons beyond gender or racial discrimination, the report noted.

We present PredProp, a method for bidirectional, parallel and local optimisation of weights, activities and precision in neural networks. PredProp jointly addresses inference and learning, scales learning rates dynamically and weights gradients by the curvature of the loss function by optimizing prediction error precision. PredProp optimizes network parameters with Stochastic Gradient Descent and error forward propagation based strictly on prediction errors and variables locally available to each layer. Neighboring layers optimise shared activity variables so that prediction errors can propagate forward in the network, while predictions propagate backwards. This process minimises the negative Free Energy, or evidence lower bound of the entire network. We show that networks trained with PredProp resemble gradient based predictive coding when the number of weights between neighboring activity variables is one. In contrast to related work, PredProp generalizes towards backward connections of arbitrary depth and optimizes precision for any deep network architecture. Due to the analogy between prediction error precision and the Fisher information for each layer, PredProp implements a form of Natural Gradient Descent. When optimizing DNN models, layer-wise PredProp renders the model a bidirectional predictive coding network. Alternatively DNNs can parameterize the weights between two activity variables. We evaluate PredProp for dense DNNs on simple inference, learning and combined tasks. We show that, without an explicit sampling step in the network, PredProp implements a form of variational inference that allows to learn disentangled embeddings from low amounts of data and leave evaluation on more complex tasks and datasets to future work.

VP Data & AI at ECS, roles have included co-founder at a data analytics startup, VP AI at Booz Allen, and Global Analytics Lead at Accenture. As the influence of artificial intelligence grows, it is increasingly vital to design processes and systems to harness AI while counterbalancing risk. Our charge is to eliminate bias, codify objectives and represent values. Responsible AI ensures alignment to our standards spanning data, algorithms, operations, technology and Human Computer Interaction. I am examining the importance of each of these elements in a series of articles.

An AI tool can quickly suggest possible candidates for the chemical structures of psychoactive "designer drugs" from a simple analysis. The tool could fast-track the development of lab tests which screen the use of drugs that have similar effects to substances such as cocaine and heroin, but have been designed to evade detection. "Our method could cut down the time required to identify a new designer drug from weeks or months to just hours," says Michael Skinnider at the University of British Columbia in Vancouver. Skinnider and his colleagues created a machine learning tool called DarkNPS by training it with chemical structures of around 1700 known designer drugs, collected from forensic labs around the world. The training set included tandem mass spectrometry results for each drug, which is a common technique that provides information on the mass of a molecule and the elements it contains.

Omnia's Trustworthy AI Module, Deloitte's unique artificial intelligence evaluation technology, has been recognized as'Digital Innovation of the Year' at the Digital Accountancy Forum and Awards 2021 in London earlier this week. This marks the second consecutive year Deloitte has garnered top honors for delivering innovative and disruptive technologies by The Accountant and International Accounting Bulletin. It also marks the fourth time Deloitte has won the award overall. Omnia DNAV, a digital cloud-based solution that revolutionizes the audit of securities and investments, was honored with the award in 2020. Deloitte won the 2018 'Audit Innovation of the Year' for its audit-transforming Cortex data platform and in 2015 for functionality using artificial intelligence that quickly identifies, extracts, and analyzes information across an entire population of documents.

In a quiet year of technology design, development, and deployment, we would expect to see the main drivers of enterprise investments (in talent, tools, and technologies) to be internal. That is, companies would quietly stoke the fires of innovation, generate new products and services, and grow their workforce capabilities, without a lot of external forces, other than the perennial aim to achieve competitive advantage in their industry. In the Pandemic era, those internal forces have taken a back seat to other external factors, or at least the passenger seat. Various issues have been loud and clear in the popular media and among the general populace that have influenced the AI adoption equation in 2021. These "hot" issues include AI ethics and bias, the "Great Resignation", global competition, climate impacts, and the role of government in regulating AI technologies.

Deep models trained through maximum likelihood have achieved state-of-the-art results for survival analysis. Despite this training scheme, practitioners evaluate models under other criteria, such as binary classification losses at a chosen set of time horizons, e.g. Brier score (BS) and Bernoulli log likelihood (BLL). Models trained with maximum likelihood may have poor BS or BLL since maximum likelihood does not directly optimize these criteria. Directly optimizing criteria like BS requires inverse-weighting by the censoring distribution, estimation of which itself also requires inverse-weighted by the failure distribution. But neither are known. To resolve this dilemma, we introduce Inverse-Weighted Survival Games to train both failure and censoring models with respect to criteria such as BS or BLL. In these games, objectives for each model are built from re-weighted estimates featuring the other model, where the re-weighting model is held fixed during training. When the loss is proper, we show that the games always have the true failure and censoring distributions as a stationary point. This means models in the game do not leave the correct distributions once reached. We construct one case where this stationary point is unique. We show that these games optimize BS on simulations and then apply these principles on real world cancer and critically-ill patient data.

As \emph{artificial intelligence} (AI) systems are increasingly involved in decisions affecting our lives, ensuring that automated decision-making is fair and ethical has become a top priority. Intuitively, we feel that akin to human decisions, judgments of artificial agents should necessarily be grounded in some moral principles. Yet a decision-maker (whether human or artificial) can only make truly ethical (based on any ethical theory) and fair (according to any notion of fairness) decisions if full information on all the relevant factors on which the decision is based are available at the time of decision-making. This raises two problems: (1) In settings, where we rely on AI systems that are using classifiers obtained with supervised learning, some induction/generalization is present and some relevant attributes may not be present even during learning. (2) Modeling such decisions as games reveals that any -- however ethical -- pure strategy is inevitably susceptible to exploitation. Moreover, in many games, a Nash Equilibrium can only be obtained by using mixed strategies, i.e., to achieve mathematically optimal outcomes, decisions must be randomized. In this paper, we argue that in supervised learning settings, there exist random classifiers that perform at least as well as deterministic classifiers, and may hence be the optimal choice in many circumstances. We support our theoretical results with an empirical study indicating a positive societal attitude towards randomized artificial decision-makers, and discuss some policy and implementation issues related to the use of random classifiers that relate to and are relevant for current AI policy and standardization initiatives.