This is the web version of Eye on A.I., Fortune's weekly newsletter covering artificial intelligence and business. To get it delivered weekly to your in-box, sign up here. In January 2020, in a Fortune magazine cover story, I chronicled the corporate race for artificial general intelligence, a kind of human-like or even superhuman A.I. that is the staple of science fiction. The pursuit of AGI, as it's more commonly called, has led to many of the machine learning innovations that underpin the current A.I. boom. But that boom is centered around narrow A.I--software that can perform one, specific task well.

'Outnumbered' panel discusses the multiple crises including the Afghanistan drone strike and border surge that could define Biden's presidency After yet another weekend out of sight, Americans should be wondering if our president even wants the job. Joe Biden's campaign was famous for calling early "lids," that is sending his traveling press home for the day before the day was even half over. Campaigns are normally furiously busy events known for crushing schedules and exhausting programming. His was a mellow affair, mostly done on a video feed from his home. The official excuse was COVID-19 and while it certainly made sense that a then-77 year old man might be concerned about catching the virus and would therefore limit his in-person exposure, it didn't make sense that even his remote events were extremely limited and rare.

Artificial intelligence (AI) has long been a feature of modern technology and is becoming increasingly common in workplace technologies. According to ManageEngine's recent 2021 Digital Readiness Survey, more than 86% of organisations in Australia and New Zealand reported increasing their use of AI even as recently as two years ago. But despite an increased uptake across organisations in the A/NZ region, only 25% said their confidence in the technology had significantly increased. One possible reason for the lack of overall confidence in AI is the potential for unethical biases to work their way into developing AI technologies. While it may be true that nobody sets out to build an unethical AI model, it may only take a few cases for disproportionate or accidental weighting to be applied to certain data types over others, creating unintentional biases.

Causal effect estimation for dynamic treatment regimes (DTRs) contributes to sequential decision making. However, censoring and time-dependent confounding under DTRs are challenging as the amount of observational data declines over time due to a reducing sample size but the feature dimension increases over time. Long-term follow-up compounds these challenges. Another challenge is the highly complex relationships between confounders, treatments, and outcomes, which causes the traditional and commonly used linear methods to fail. We combine outcome regression models with treatment models for high dimensional features using uncensored subjects that are small in sample size and we fit deep Bayesian models for outcome regression models to reveal the complex relationships between confounders, treatments, and outcomes. Also, the developed deep Bayesian models can model uncertainty and output the prediction variance which is essential for the safety-aware applications, such as self-driving cars and medical treatment design. The experimental results on medical simulations of HIV treatment show the ability of the proposed method to obtain stable and accurate dynamic causal effect estimation from observational data, especially with long-term follow-up. Our technique provides practical guidance for sequential decision making, and policy-making.

Modern streaming services are increasingly labeling videos based on their visual or audio content. This typically augments the use of technologies such as AI and ML by allowing to use natural speech for searching by keywords and video descriptions. Prior research has successfully provided a number of solutions for speech to text, in the case of a human speech, but this article aims to investigate possible solutions to retrieve sound events based on a natural language query, and estimate how effective and accurate they are. In this study, we specifically focus on the YamNet, AlexNet, and ResNet-50 pre-trained models to automatically classify audio samples using their respective melspectrograms into a number of predefined classes. The predefined classes can represent sounds associated with actions within a video fragment. Two tests are conducted to evaluate the performance of the models on two separate problems: audio classification and intervals retrieval based on a natural language query. Results show that the benchmarked models are comparable in terms of performance, with YamNet slightly outperforming the other two models. YamNet was able to classify single fixed-size audio samples with 92.7% accuracy and 68.75% precision while its average accuracy on intervals retrieval was 71.62% and precision was 41.95%. The investigated method may be embedded into an automated event marking architecture for streaming services.

We investigate causal inference in the asymptotic regime as the number of variables approaches infinity using an information-theoretic framework. We define structural entropy of a causal model in terms of its description complexity measured by the logarithmic growth rate, measured in bits, of all directed acyclic graphs (DAGs), parameterized by the edge density d. Structural entropy yields non-intuitive predictions. If we randomly sample a DAG from the space of all models, in the range d = (0, 1/8), almost surely the model is a two-layer DAG! Semantic entropy quantifies the reduction in entropy where edges are removed by causal intervention. Semantic causal entropy is defined as the f-divergence between the observational distribution and the interventional distribution P', where a subset S of edges are intervened on to determine their causal influence. We compare the decomposability properties of semantic entropy for different choices of f-divergences, including KL-divergence, squared Hellinger distance, and total variation distance. We apply our framework to generalize a recently popular bipartite experimental design for studying causal inference on large datasets, where interventions are carried out on one set of variables (e.g., power plants, items in an online store), but outcomes are measured on a disjoint set of variables (residents near power plants, or shoppers). We generalize bipartite designs to k-partite designs, and describe an optimization framework for finding the optimal k-level DAG architecture for any value of d \in (0, 1/2). As edge density increases, a sequence of phase transitions occur over disjoint intervals of d, with deeper DAG architectures emerging for larger values of d. We also give a quantitative bound on the number of samples needed to reliably test for average causal influence for a k-partite design.

Customer segmentation has long been a productive field in banking. However, with new approaches to traditional problems come new opportunities. Fine-grained customer segments are notoriously elusive and one method of obtaining them is through feature extraction. It is possible to assign coefficients of standard personality traits to financial transaction classes aggregated over time. However, we have found that the clusters formed are not sufficiently discriminatory for micro-segmentation. In this study, we extract temporal features with continuous values from the hidden states of neural networks predicting customers' spending personality from their financial transactions. We consider both temporal and non-sequential models, using long short-term memory (LSTM) and feed-forward neural networks, respectively. We found that recurrent neural networks produce micro-segments where feed-forward networks produce only course segments. Finally, we show that classification using these extracted features performs at least as well as bespoke models on two common metrics, namely loan default rate and customer liquidity index.

The deep learning approach to detecting malicious software (malware) is promising but has yet to tackle the problem of dataset shift, namely that the joint distribution of examples and their labels associated with the test set is different from that of the training set. This problem causes the degradation of deep learning models without users' notice. In order to alleviate the problem, one approach is to let a classifier not only predict the label on a given example but also present its uncertainty (or confidence) on the predicted label, whereby a defender can decide whether to use the predicted label or not. While intuitive and clearly important, the capabilities and limitations of this approach have not been well understood. In this paper, we conduct an empirical study to evaluate the quality of predictive uncertainties of malware detectors. Specifically, we re-design and build 24 Android malware detectors (by transforming four off-the-shelf detectors with six calibration methods) and quantify their uncertainties with nine metrics, including three metrics dealing with data imbalance. Our main findings are: (i) predictive uncertainty indeed helps achieve reliable malware detection in the presence of dataset shift, but cannot cope with adversarial evasion attacks; (ii) approximate Bayesian methods are promising to calibrate and generalize malware detectors to deal with dataset shift, but cannot cope with adversarial evasion attacks; (iii) adversarial evasion attacks can render calibration methods useless, and it is an open problem to quantify the uncertainty associated with the predicted labels of adversarial examples (i.e., it is not effective to use predictive uncertainty to detect adversarial examples).

Sum-of-norms clustering is a convex optimization problem whose solution can be used for the clustering of multivariate data. We propose and study a localized version of this method, and show in particular that it can separate arbitrarily close balls in the stochastic ball model. More precisely, we prove a quantitative bound on the error incurred in the clustering of disjoint connected sets. Our bound is expressed in terms of the number of datapoints and the localization length of the functional.

A new report about artificial intelligence and its effects warns AI has reached a turning point and its negative effects can no longer be ignored. The big picture: For all the sci-fi worries about ultra-intelligent machines or wide-scale job loss from automation -- both of which would require artificial intelligence that is far more capable than what has been developed so far -- the larger concern may be about what happens if AI doesn't work as intended. Background: The AI100 project -- which was launched by Eric Horvitz, who served as Microsoft's first chief scientific officer, and is hosted by the Stanford Institute on Human-Centered AI (HAI) -- is meant to provide a longitudinal study of a technology that seems to be advancing by the day. What's happening: The panel found AI has exhibited remarkable progress over the past five years, especially in the area of natural language processing (NLP) -- the ability of AI to analyze and generate human language. The catch: That means AI has reached a point where its downsides in the real world are becoming increasingly difficult to miss -- and increasingly difficult to stop.