The traditional definition of artificial intelligence is the ability of machines to execute tasks and solve problems in ways normally attributed to humans. Some tasks that we consider simple--recognizing an object in a photo, driving a car--are incredibly complex for AI. Machines can surpass us when it comes to things like playing chess, but those machines are limited by the manual nature of their programming; a 30 gadget can beat us at a board game, but it can't do--or learn to do--anything else. This is where machine learning comes in. Show millions of cat photos to a machine, and it will hone its algorithms to improve at recognizing pictures of cats.

AlphaGo, the board-game-playing AI from Google's DeepMind subsidiary, is one of the most famous examples of deep learning – machine learning using neural networks – to date. So it may be surprising to learn that some of the code that led to the machine's victory was created by good old-fashioned humans. The software, which beat Korean Go Champion Lee Sedol 4–1 in March, taught itself to play the ancient Asian game by running millions of simulations against itself. AlphaGo is one of two neural networks, taught by a mixture of supervised learning (studying previous games played by humans) and reinforcement learning (playing against itself and learning from its mistakes). But some things, it turns out, just can't be taught.

Last week MPs warned that the UK is facing a digital skills crisis. Its report explained that 90% of jobs require digital skills to some degree, but flagged that the UK needs another 745,000 workers with digital skills by 2017 in order to remain competitive against other countries. At the moment the skills gap is reportedly costing the UK around 63bn a year. In tandem, we are seeing more companies invest in automation and artificial intelligence (AI), which reduces the need for humans in the decision making process. For example, Twitter recently announced the acquisition of machine learning startup, Magic Pony Technology.

Experts at the Annual Meeting of the New Champions tackled the issue of artificial intelligence and what it means for humans, concluding that they can and should coexist. The pertinent issue is how humans can leverage artificial intelligence to enhance the outcome of new technologies and improve quality of life, and not focus on the narrative of human vs machine. However, rapid technological advances underline the urgency for policy-makers to redesign educational systems so that younger generations are adequately prepared for a workplace that will see more automated processes. "By some estimates, 47% of existing jobs in the US could be replaced by automation," said Wendell Wallach, Scholar, Interdisciplinary Center for Bioethics, Yale University, USA. "When the World Bank used similar methodology, it came up with 69% in India, and 77% in China. If that's truly the case, we are talking about tremendous jobs being lost," he added.

The little striker wearing a crescent moon and star jersey lines up his penalty and kicks right, netting his goal as the keeper dives the wrong way and hits the ground yelping in pain. Both players are teammates practising to represent Pakistan in a major world football tournament. Unlike their low-ranked flesh-and-blood counterparts, however, these are advanced robots whose programmers are set to compete against students from the world's top universities as they look to showcase what their country can do in the world of Artificial Intelligence. Students at Pakistan's National University of Science and Technology (NUST) will this year for the first time send a team to the annual RoboCup, an event featuring 32 universities that will be held in Leipzig, Germany from June 27 to July 4. The six machines are NAO humanoid robots purchased from France's Aldebaran Robotics at a cost of roughly 17,000. It is in fact the third year that NUST, Pakistan's premier engineering institute, has qualified for the prestigious cup.

But this hand-wringing is a distraction from the very real problems with artificial intelligence today, which may already be exacerbating inequality in the workplace, at home and in our legal and judicial systems. Sexism, racism and other forms of discrimination are being built into the machine-learning algorithms that underlie the technology behind many "intelligent" systems that shape how we are categorized and advertised to. Take a small example from last year: Users discovered that Google's photo app, which applies automatic labels to pictures in digital photo albums, was classifying images of black people as gorillas. Google apologized; it was unintentional. This is fundamentally a data problem.

At the end of April, we published an article on getting up to speed on deep learning, which included 20 resources to catch up on rapid advancements in the field. Much has happened since then, so we thought we'd pull together a few of the excellent resources that have emerged this month in June. As always, this list is not comprehensive, so let us know if there's something we should add, or if you're interested in discussing this area further. Facebook introduces DeepText, its deep learning engine that understands textual content on Facebook with near-human accuracy and at a speed of several thousand posts per second, in more than 20 languages. Google DeepMind learns to play Montezuma's Revenge via intrinsic motivation techniques (video).

The word Artificial Intelligence is increasing. The latest example is Picturesqe, a tool for photographers that uses AI-powered automation to help pick out the best snaps and filter out the dross. Founders of Picturesqe, a machine-learning powered piece of software, are confident that it can select the good photos from your large stack. But unlike similar mobile Apps, Picturesqe is targeted specifically at professional photographers and semi-pros. Features of the application include smart grouping, which automatically groups similar photos based on visual content, intelligent zoom so that you can quickly compare the same spot on multiple shots, and aesthetic ranking.

The human auditory system gives us the extraordinary ability to converse above the chatter of a lively cocktail party. Selective listening in such conditions is an extremely challenging task for computers, and has been the holy grail of speech processing for more than 50 years. Previously, no practical method existed in the case of single channel mixtures of speech, especially when the speakers are unknown, but now Mitsubishi Electric Research Labs (MERL) are addressing the problem of acoustic source separation with a deep learning framework they call "deep clustering". At the Deep Learning Summit in Boston last month John Hershey, Senior Principal Research Scientist at MERL, presented'Cracking the Cocktail Party Problem: Deep Clustering for Speech Separation' and shared their breakthrough, using their deep clustering network to assign embedding vectors to different sonic elements of the noisy signal. With this technology, MERL are on the verge of solving the general audio separation problem, opening up a new era in spontaneous human-machine communication.

Infinite–Layer Networks (ILN) have recently been proposed as an architecture that mimics neural networks while enjoying some of the advantages of kernel methods. ILN are networks that integrate over infinitely many nodes within a single hidden layer. It has been demonstrated by several authors that the problem of learning ILN can be reduced to the kernel trick, implying that whenever a certain integral can be computed analytically they are efficiently learnable. In this work we give an online algorithm for ILN, which avoids the kernel trick assumption. More generally and of independent interest, we show that kernel methods in general can be exploited even when the kernel cannot be efficiently computed but can only be estimated via sampling. We provide a regret analysis for our algorithm, showing that it matches the sample complexity of methods which have access to kernel values.