Each innovation challenges the norms, codes, and values of the society in which it is embedded. The industrial revolution unleashed new forces of productivity but at the cost of inhumane working conditions, leading to the creation of unions, labor laws, and the foundations of the political party structures of modern democracies. Fossil fuels powered a special century of growth before pushing governments, companies, and civil society to phase them out to protect our health, ecology, and climate. When innovations lead to disaster, it says much about the societal context. The Chernobyl nuclear disaster embodied the flaws of Soviet planning.

This display was computed in less than one hour.... view more A new machine learning algorithm is poised to help urban transportation analysts relieve bottlenecks and chokepoints that routinely snarl city traffic. The tool, called TranSEC, was developed at the U.S. Department of Energy's Pacific Northwest National Laboratory to help urban traffic engineers get access to actionable information about traffic patterns in their cities. WATCH: https://www.youtube.com/watch?v 8S4bLv9CtOo (Video by Graham Bourque Pacific Northwest National Laboratory) Currently, publicly available traffic information at the street level is sparse and incomplete. Traffic engineers generally have relied on isolated traffic counts, collision statistics and speed data to determine roadway conditions. The new tool uses traffic datasets collected from UBER drivers and other publicly available traffic sensor data to map street-level traffic flow over time.

These are just a few of the ill-defined principles commonly listed in ethical frameworks for artificial intelligence (AI), hundreds of which have now been released by organizations ranging from Google to the government of Canada to BMW. As organizations embrace AI with increasing speed, adopting these principles is widely viewed as one of the best ways to ensure AI does not cause unintended harms. Many AI ethical frameworks cannot be clearly implemented in practice, as researchers have consistently demonstrated. Without a dramatic increase in the specificity of existing AI frameworks, there's simply not much technical personnel can do to clearly uphold such high-level guidance. And this, in turn, means that while AI ethics frameworks may make for good marketing campaigns, they all too frequently fail to stop AI from causing the very harms they are meant to prevent.

The facial recognition community continues to speak in circles when ethics is the topic. Notable business-to-business tech analyst Gartner is the latest with indistinct advice that amounts to "Please use better judgment." A new post by Gartner analyst Frank Buytendijk advises, "The appropriate use of facial recognition technology depends on the prevailing culture, ethics, legislation and practices." People in public might as well be fish in a barrel analyzed for profit by machine vision. Vendors leave questions of ethics in the hands of buyers.

I have always worked with data, since high school, a long time ago in a galaxy far far away. Specifically, my background is astrophysics, with a Ph.D. in the subject. I performed astronomical data analysis, modeling, and simulation for 25 years, while also working on data repositories for space science satellite missions at NASA. I became very interested in the scientific discovery opportunities of very large datasets in the late 1990's, at which time I began my quest into machine learning, data mining, and data science. The motivation for me has always been discovery, from my early days until now.

The paper introduces the notion of an epistemic argumentation framework (EAF) as a means to integrate the beliefs of a reasoner with argumentation. Intuitively, an EAF encodes the beliefs of an agent who reasons about arguments. Formally, an EAF is a pair of an argumentation framework and an epistemic constraint. The semantics of the EAF is defined by the notion of an ω-epistemic labelling set, where ω is complete, stable, grounded, or preferred, which is a set of ω-labellings that collectively satisfies the epistemic constraint of the EAF. The paper shows how EAF can represent different views of reasoners on the same argumentation framework. It also includes representing preferences in EAF and multi-agent argumentation.

We target the problem of detecting Trojans or backdoors in DNNs. Such models behave normally with typical inputs but produce specific incorrect predictions for inputs poisoned with a Trojan trigger. Our approach is based on a novel observation that the trigger behavior depends on a few ghost neurons that activate on trigger pattern and exhibit abnormally higher relative attribution for wrong decisions when activated. Further, these trigger neurons are also active on normal inputs of the target class. Thus, we use counterfactual attributions to localize these ghost neurons from clean inputs and then incrementally excite them to observe changes in the model's accuracy. We use this information for Trojan detection by using a deep set encoder that enables invariance to the number of model classes, architecture, etc. Our approach is implemented in the TrinityAI tool that exploits the synergies between trustworthiness, resilience, and interpretability challenges in deep learning. We evaluate our approach on benchmarks with high diversity in model architectures, triggers, etc. We show consistent gains (+10%) over state-of-the-art methods that rely on the susceptibility of the DNN to specific adversarial attacks, which in turn requires strong assumptions on the nature of the Trojan attack.

We consider a wireless communication system that consists of a background emitter, a transmitter, and an adversary. The transmitter is equipped with a deep neural network (DNN) classifier for detecting the ongoing transmissions from the background emitter and transmits a signal if the spectrum is idle. Concurrently, the adversary trains its own DNN classifier as the surrogate model by observing the spectrum to detect the ongoing transmissions of the background emitter and generate adversarial attacks to fool the transmitter into misclassifying the channel as idle. This surrogate model may differ from the transmitter's classifier significantly because the adversary and the transmitter experience different channels from the background emitter and therefore their classifiers are trained with different distributions of inputs. This system model may represent a setting where the background emitter is a primary, the transmitter is a secondary, and the adversary is trying to fool the secondary to transmit even though the channel is occupied by the primary. We consider different topologies to investigate how different surrogate models that are trained by the adversary (depending on the differences in channel effects experienced by the adversary) affect the performance of the adversarial attack. The simulation results show that the surrogate models that are trained with different distributions of channel-induced inputs severely limit the attack performance and indicate that the transferability of adversarial attacks is neither readily available nor straightforward to achieve since surrogate models for wireless applications may significantly differ from the target model depending on channel effects.

We consider evaluating the causal effects of dynamic treatments, i.e. of multiple treatment sequences in various periods, based on double machine learning to control for observed, time-varying covariates in a data-driven way under a selection-on-observables assumption. To this end, we make use of so-called Neyman-orthogonal score functions, which imply the robustness of treatment effect estimation to moderate (local) misspecifications of the dynamic outcome and treatment models. This robustness property permits approximating outcome and treatment models by double machine learning even under high dimensional covariates and is combined with data splitting to prevent overfitting. In addition to effect estimation for the total population, we consider weighted estimation that permits assessing dynamic treatment effects in specific subgroups, e.g. among those treated in the first treatment period. We demonstrate that the estimators are asymptotically normal and $\sqrt{n}$-consistent under specific regularity conditions and investigate their finite sample properties in a simulation study. Finally, we apply the methods to the Job Corps study in order to assess different sequences of training programs under a large set of covariates.

Whatever the ultimate impact may be of a report by UK experts in algorithmic bias, the document already has succeeded where many analyst reports have failed. The new report, on bias in algorithmic decision making, comes from the government-funded Centre for Data Ethics and Innovation. It takes the position that AI decision making must be ethical to be successful, and AI ethics must be viewed as impacting individual people -- because it does. Thinking about AI ethics in terms of industries, regions, demographics and data sets is an easy out. It lets everyone involved spread harms across faceless multitudes.