At the same time, AI is being twisted by authoritarian regimes to violate rights. The Chinese Communist Party is reportedly using AI to uncover and punish those who criticize the regime's pandemic response and to institute a type of coronavirus social-credit score--assigning people color codes to determine who is free to go out and who will be forced into quarantine. As the world begins to recover from the pandemic, nations face a stark choice about what vision of artificial intelligence will prevail. As Group of Seven nations meet this year under the organization's U.S. presidency, there is a critical opportunity to shape the evolution of AI in a way that respects fundamental rights and upholds our shared values. That is why G-7 technology ministers will agree Thursday to launch the Global Partnership on Artificial Intelligence, or GPAI, together with other democratic countries.

In February, a lobbyist friend urged Erik Sartorius, the executive director of the Kansas League of Municipalities, to look at a newly introduced bill that would affect cities. The legislation involved "personal delivery devices"--robots that, as if in a sci-fi movie, might deliver a bag of groceries, a toolbox, or a prescription to your doorstep. It would have limited their weight to 150 pounds, not including the cargo inside. And it would have allowed them to operate on any sidewalk or crosswalk in Kansas at speeds up to 6 miles per hour, the pace of a quick human jog. Lawmakers and lobbyists say the bill was drafted with help from Amazon.

MIT SMR Connections is the custom content creation unit within MIT Sloan Management Review. In this Q&A, Michelle K. Lee, vice president of the Amazon Web Services (AWS) Machine Learning Solutions Lab, shares real-world examples of machine learning in action, describes four key implementation challenges, and offers other advice. This conversation has been condensed and edited for clarity, length, and editorial style. Q: Can you provide an overview of how artificial intelligence (AI) and machine learning (ML) are driving digital transformation? Lee: AI and machine learning went from being aspirational technology to mainstream extremely fast.

Enterprises around the world are rapidly incorporating artificial intelligence (AI) into existing and new products and processes. This effort is not just to improve such offerings and services, but to achieve a qualitatively higher level of capability not possible before. It is clear that AI carries the potential for many new opportunities, across all industries, but it is also already recognized that it brings numerous risks as well. As with any technology, senior management and board directors need to be aware of both the opportunity and the risk in order to successfully and responsibly manage the enterprise. The opportunities are great--AI can assist in robotic process automation (RPA), machine learning, natural language processing, finding new drugs and therapies, and will be essential for driverless transportation--but if the risks are downplayed or overlooked, there can be serious reputational and/or legal consequences.

Jaynarayan H. Lala (jay.lala@rtx.com) is a Senior Principal Engineering Fellow at Raytheon Technologies, San Diego, CA, USA. Carl E. Landwehr (carl.landwehr@gmail.com) is a Research Scientist at George Washington University and a Visiting Professor at University of Michigan, Ann Arbor, MI, USA. John F. Meyer (jfm@umich.edu) is a Professor Emeritus of Computer Science and Engineering at University of Michigan, Ann Arbor, MI, USA. This Viewpoint is derived from material produced as part of the Intelligent Vehicle Dependability and Security (IVDS) project of IFIP Working Group 10.4.

Back in 2014, the Los Angeles Times published a report about an earthquake three minutes after it happened. This feat was possible because a staffer had developed a bot (a software robot) called Quakebot to write automated articles based on data generated by the US Geological Survey. Today, AIs write hundreds of thousands of the articles that are published by mainstream media outlets every week. At first, most of the Natural Language Generation (NLG) tools producing these articles were provided by software companies like Narrative Science. Today, many media organisations have developed in-house versions.

The amount of data for processing and categorization grows at an ever increasing rate. At the same time the demand for collaboration and transparency in organizations, government and businesses, drives the release of data from internal repositories to the public or 3rd party domain. This in turn increase the potential of sharing sensitive information. The leak of sensitive information can potentially be very costly, both financially for organizations, but also for individuals. In this work we address the important problem of sensitive information detection. Specially we focus on detection in unstructured text documents. We show that simplistic, brittle rule sets for detecting sensitive information only find a small fraction of the actual sensitive information. Furthermore we show that previous state-of-the-art approaches have been implicitly tailored to such simplistic scenarios and thus fail to detect actual sensitive content. We develop a novel family of sensitive information detection approaches which only assumes access to labeled examples, rather than unrealistic assumptions such as access to a set of generating rules or descriptive topical seed words. Our approaches are inspired by the current state-of-the-art for paraphrase detection and we adapt deep learning approaches over recursive neural networks to the problem of sensitive information detection. We show that our context-based approaches significantly outperforms the family of previous state-of-the-art approaches for sensitive information detection, so-called keyword-based approaches, on real-world data and with human labeled examples of sensitive and non-sensitive documents.

We study the multiple manifold problem, a binary classification task modeled on applications in machine vision, in which a deep fully-connected neural network is trained to separate two low-dimensional submanifolds of the unit sphere. We provide an analysis of the one-dimensional case, proving for a simple manifold configuration that when the network depth $L$ is large relative to certain geometric and statistical properties of the data, the network width $n$ grows as a sufficiently large polynomial in $L$, and the number of i.i.d. samples from the manifolds is polynomial in $L$, randomly-initialized gradient descent rapidly learns to classify the two manifolds perfectly with high probability. Our analysis demonstrates concrete benefits of depth and width in the context of a practically-motivated model problem: the depth acts as a fitting resource, with larger depths corresponding to smoother networks that can more readily separate the class manifolds, and the width acts as a statistical resource, enabling concentration of the randomly-initialized network and its gradients. The argument centers around the neural tangent kernel and its role in the nonasymptotic analysis of training overparameterized neural networks; to this literature, we contribute essentially optimal rates of concentration for the neural tangent kernel of deep fully-connected networks, requiring width $n \gtrsim L\,\mathrm{poly}(d_0)$ to achieve uniform concentration of the initial kernel over a $d_0$-dimensional submanifold of the unit sphere $\mathbb{S}^{n_0-1}$, and a nonasymptotic framework for establishing generalization of networks trained in the NTK regime with structured data. The proof makes heavy use of martingale concentration to optimally treat statistical dependencies across layers of the initial random network. This approach should be of use in establishing similar results for other network architectures.

The importance of algorithmic fairness grows with the increasing impact machine learning has on people's lives. Recent work on fairness metrics shows the need for causal reasoning in fairness constraints. In this work, a practical method named FairTrade is proposed for creating flexible prediction models which integrate fairness constraints on sensitive causal paths. The method uses recent advances in variational inference in order to account for unobserved confounders. Further, a method outline is proposed which uses the causal mechanism estimates to audit black box models. Experiments are conducted on simulated data and on a real dataset in the context of detecting unlawful social welfare. This research aims to contribute to machine learning techniques which honour our ethical and legal boundaries.

Philadelphia-based software and analytics firm Azavea received a full $150,000 award for its project, "Advancing Flood Extent Delineation Modeling Using Synthetic Aperture Radar (SAR) Data." Using the grant, Azavea will work to resolve a key problem with timely responses to flood events: seeing through the heavy cloud cover that often accompanies flooding. To combat the cloud cover, Azavea will apply synthetic-aperture radar (or SAR), which uses radar to reconstruct images and landscapes, in combination with deep learning techniques to interpret SAR imagery in real-time. "By combining these two technologies," Azavea writes, "this project will support the rapid delivery of accurate flood inundation maps that will enable first responders, humanitarian relief organizations, and other decision-makers on the ground to effectively route resources and identify highly impacted areas, both during and following extreme weather events."