Since the 1980s the number of women completing computer science degrees has plummeted, and in most large tech companies the representation of women in technical roles is below 30%. This lack of diversity prevents us from building products that work for everybody. It can foster toxic "brogrammer" cultures which harm everybody who works within them, and it deprives teams of the well-documented performance boost that women bring. Many of the early superstars in computer science were women -- from Lord Byron's polymath daughter Ada Lovelace, the first person to envisage a general purpose computer, to Rear Admiral Grace Hooper, who pioneered the use of natural language in writing computer programs. Similarly, the post-war computing scene was dominated by women.

AI saves the day again as Google gives a helping hand in the health sector. Google has developed a new artificial intelligence (AI) robot that can predict a patient's risk of heart disease and other cardiovascular events. The new robot, built in partnership with health organisation Verily, uses AI to analyse data from almost 300,000 patients and then predict cardiovascular outcomes based on the data and patient assessment. Deep learning is used to spot patterns in the information, learning the early signs of life threatening diseases from eye scans and the metrics needed to predict the risk. The tech giant's new robot reads a patient's retinas to assess the risk of heart disease.

The AI genie is out of the bottle. As we swipe on our screens and make online requests, Artificial Intelligence grants our wishes. As with child rearing, it's our responsibility to guide AI into maturity. If neglected and abused, AI can potentially become "more dangerous than nukes" as Elon Musk puts it. However, if cared for, AI can level-up society into a fourth industrial revolution.

The world is under threat from artificial intelligence and needs to do more to keep people safe, experts have urged. A new report compiled by 26 of the world's leading experts paints a terrifying picture of the world in the next 10 years. Physical attacks as well as those on our digital worlds and political system could drastically undermine the safety of humanity, it warns, and people must work together now if they want to keep the world safe. The use of artificial intelligence is likely to empower all kinds of people – including rogue states, criminals, and terrorists, the report warns. Boston Dynamics describes itself as'building dynamic robots and software for human simulation'.

Google has been tinkering in the field of medicine over the last few years, including developing a prototype electronic contact lens. The company's latest health project involves detecting cardiovascular conditions by analyzing the vasculature of the retina. The researchers built a deep learning system that processed data from two datasets containing thousands of patients, each of which included images of a patient's retina along with various risk factors and health conditions such smoking and high blood pressure. The system found correlations between various parameters measured within the retinal images and cardiovascular risk factors, as well as disease. For example, the software was able to identify smokers just by looking at the retina 71% of the time.

The brain's neural network has long inspired artificial-intelligence researchers.Credit: Alfred Pasieka/SPL/Getty Four years ago, scientists from Google showed up on neuroscientist Steve Finkbeiner's doorstep. The researchers were based at Google Accelerated Science, a research division in Mountain View, California, that aims to use Google technologies to speed scientific discovery. They were interested in applying'deep-learning' approaches to the mountains of imaging data generated by Finkbeiner's team at the Gladstone Institute of Neurological Disease in San Francisco, also in California. Deep-learning algorithms take raw features from an extremely large, annotated data set, such as a collection of images or genomes, and use them to create a predictive tool based on patterns buried inside. Once trained, the algorithms can apply that training to analyse other data, sometimes from wildly different sources.

We've co-authored a paper that forecasts how malicious actors could misuse AI technology, and potential ways we can prevent and mitigate these threats. This paper is the outcome of almost a year of sustained work with our colleagues at the Future of Humanity Institute, the Centre for the Study of Existential Risk, the Center for a New American Security, the Electronic Frontier Foundation, and others. AI challenges global security because it lowers the cost of conducting many existing attacks, creates new threats and vulnerabilities, and further complicates the attribution of specific attacks. Like our work on concrete problems in AI safety, we've grounded some of the problems motivated by the malicious use of AI in concrete scenarios, such as: persuasive ads generated by AI systems being used to target the administrator of a security systems; cybercriminals using neural networks and "fuzzing" techniques to create computer viruses with automatic exploit generation capabilities; malicious actors hacking a cleaning robot so that it delivers an explosives payload to a VIP; and rogue states using omniprescent AI-augmented surveillance systems to pre-emptively arrest people who fit a predictive risk profile. We're excited to start having this discussion with our peers, policymakers, and the general public; we've spent the last two years researching and solidifying our internal policies at OpenAI and are going to begin engaging a wider audience on these issues.

Actor-critic methods solve reinforcement learning problems by updating a parameterized policy known as an actor in a direction that increases an estimate of the expected return known as a critic. However, existing actor-critic methods only use values or gradients of the critic to update the policy parameter. In this paper, we propose a novel actor-critic method called the guide actor-critic (GAC). GAC firstly learns a guide actor that locally maximizes the critic and then it updates the policy parameter based on the guide actor by supervised learning. Our main theoretical contributions are two folds. First, we show that GAC updates the guide actor by performing second-order optimization in the action space where the curvature matrix is based on the Hessians of the critic. Second, we show that the deterministic policy gradient method is a special case of GAC when the Hessians are ignored. Through experiments, we show that our method is a promising reinforcement learning method for continuous controls.

Theory of mind (ToM; Premack & Woodruff, 1978) broadly refers to humans' ability to represent the mental states of others, including their desires, beliefs, and intentions. We propose to train a machine to build such models too. We design a Theory of Mind neural network -- a ToMnet -- which uses meta-learning to build models of the agents it encounters, from observations of their behaviour alone. Through this process, it acquires a strong prior model for agents' behaviour, as well as the ability to bootstrap to richer predictions about agents' characteristics and mental states using only a small number of behavioural observations. We apply the ToMnet to agents behaving in simple gridworld environments, showing that it learns to model random, algorithmic, and deep reinforcement learning agents from varied populations, and that it passes classic ToM tasks such as the "Sally-Anne" test (Wimmer & Perner, 1983; Baron-Cohen et al., 1985) of recognising that others can hold false beliefs about the world. We argue that this system -- which autonomously learns how to model other agents in its world -- is an important step forward for developing multi-agent AI systems, for building intermediating technology for machine-human interaction, and for advancing the progress on interpretable AI.

The roles played by learning and memorization represent an important topic in deep learning research. Recent work on this subject has shown that the optimization behavior of DNNs trained on shuffled labels is qualitatively different from DNNs trained with real labels. Here, we propose a novel permutation approach that can differentiate memorization from learning in deep neural networks (DNNs) trained as usual (i.e., using the real labels to guide the learning, rather than shuffled labels). The evaluation of weather the DNN has learned and/or memorized, happens in a separate step where we compare the predictive performance of a shallow classifier trained with the features learned by the DNN, against multiple instances of the same classifier, trained on the same input, but using shuffled labels as outputs. By evaluating these shallow classifiers in validation sets that share structure with the training set, we are able to tell apart learning from memorization. Application of our permutation approach to multi-layer perceptrons and convolutional neural networks trained on image data corroborated many findings from other groups. Most importantly, our illustrations also uncovered interesting dynamic patterns about how DNNs memorize over increasing numbers of training epochs, and support the surprising result that DNNs are still able to learn, rather than only memorize, when trained with pure Gaussian noise as input.