CSI
Machine learning and AI systems are set to become a dominant feature across the media ecosystem. How is AI being applied to media in practice today and in planning for tomorrow?
p30 http://www.csimagazine.com/eblast/Digital_Editions/Feb-2017/Digital_EditionFeb2017.pdf
Artificial Intelligence was arguably the hottest topic at the CES with products from Amazon Alexa voice controlled speakers to toothbrushes boasting the new capability. In media, such systems are already being used to develop new compression schemes and better analyse big data for services like content recommendation.
Although alive in concept since the 1940s, AI began its renaissance around 2010 by being applied to a number of problems that had been extremely hard to solve: object recognition, voice recognition and text translation to name but a few. To illustrate the scale of the progress, the performance of object recognition algorithms on the benchmark ImageNet database went from an error rate of 28% in 2010 to less than 3% in 2016, lower than the human error rate on the same data (see www.image-net. org/challenges/LSVRC/). Such extraordinary advances have been possible mostly thanks to technical and intellectual advances that have allowed the development of very large Artificial Neural Networks (ANN) coupled with the availability of a huge quantity of data to train the networks.
Following the money An unprecedented amount of money subsequently poured into machine learning research, led by the largest internet companies - Google, Facebook, Apple, IBM, Twitter - to develop the software and hardware tools to support the design and deployment of ANNs. Equity funding of AI-focused start-ups reached an all-time high in the second quarter of 2016 of more than $1 billion, according to researcher CB Insights. There were 121 funding rounds for such start-ups, representing more than $7.5bn of investment.
Among the investments, Facebook acquired FacioMetrics in November. This spin-off from Carnegie Mellon University developed IntraFace, a technology that can detect seven different emotions on peoples’ faces. Speculation is that the acquisition will feed into Facebook’s work on life-like avatars that convey emotions via ‘VR emoji’. In August, Intel bought machine learning start-up Nervana Systems and in October Samsung acquired AI software developer Viv Labs [http://viv.ai], with plans to equip its next Galaxy S smartphones with a Siri-like digital assistant. In September, Amazon, Facebook, Google, IBM, and Microsoft formed the non-profit Partnership on AI to advance public understanding of the subject and conduct research on ethics and best practices.
“The reality [is] that every person, business and nation must deal with the emergence of AI as a competitive advantage,” says James Canton, CEO at the Institute for Global Futures. “Those that have it will gain a vital global strategic advantage over others.” If that sounds like overkill, here’s Google chief exec Sundar Pichai; “We are at a seminal moment in computing. We are evolving from a mobile-first to an AI-first world.” Speaking at the launch of Google’s Pixel phone in October, Pichai said that the company’s shift to AI is as fundamental “as the invention of the Web or the smartphone.”
“When I look at where computing is heading, I see how machine learning and artificial intelligence are unlocking capabilities that were unthinkable only a few years ago. This means that the power of the software — the ‘smarts’ — really matter for hardware more than ever before.”
Augmented Intelligence Machine learning and AI are used interchangeably, but in reality most so-called AI apps are productised machine learning (ML) applications for which AI is a catch-all but misleading phrase. ML is best understood as learning machines and should be distinguished from AI, or machines that think. AI is a branch of computer science attempting to build machines capable of intelligent behaviour, while Stanford University defines machine learning as “the science of getting computers to act without being
explicitly programmed”.
You need robust ML built on ANNs before you get to AI but currently there are few true mainstream AI applications outside of autonomous cars and, to a lesser extent, virtual assistants like Alexa, Siri and the Google Assistant. “More complex types of cognitive technology — neural or deep learning networks, natural language processing, and algorithms — can seem like black boxes even to the data scientists who create them,” says Thomas Davenport, a professor in IT at Massachusetts’ Babson College, co-founder of the International Institute for Analytics and a senior advisor to Deloitte Analytics.
“AI used to be relegated to only the fastest supercomputers, but recent advances in software and the use of GPUs to process the algorithms mean that the cost of AI assistance is no longer a barrier to entry,” says Paul Turner VP, enterprise product management, Telestream. “You can certainly imagine that AI systems will be able to analyse the actual content for metadata gathering. Given that metadata is key to automated workflows, this could vastly expand our capability to ‘mine’ content for other purposes.” IBM prefers to talk about augmented intelligence.
“It’s an approach which asks how AI supports decision making and demands a societal change in how we look at technology,” says Carrie Lomas, IBM’s cognitive solutions and IoT executive. “Through personal devices like tablets to all manner of items with sensors, the (industry as a whole) is taking in lots of data and combining it with different types of information to enable a genuinely new understanding of the world.”
IBM’s cognitive computer system Watson, for example, has combined its Alchemy Language APIs with a speech to text platform, to create a tool for video owners to analyse video – forming IBM Cloud Video. It is able to scan social media in real time to monitor reactions to live streaming events. Much of the ML and AI applications a CES are built on Nvidia chipsets. Nvidia describes its TensorRT product as a “high performance neural network inference engine for production deployment of deep learning applications”.
It is targeting delivery of super fast “inferences and significantly reduced latency”, as demanded by real-time services such as streaming video categorisation in the cloud. AI technologies will undoubtedly broadly impact the media sector at all stages, from media production to delivery. The biggest corporations on the planet are embedding predictive intelligence into everyday apps to make all our lives easier. The most obvious examples are digital agents or chatbots like Amazon Alexa, Apple Siri, Google Assistant, Facebook M, and Microsoft Cortana. Siri, for example, acts as a personal assistant, using voice processing.
The bulk of AI’s potential, however, has still to reach the mainstream. Here are a few examples of the most promising applications that are being developed:
Accessibility. Automatic description of photos and movie scenes for the blind. Facebook delivered a first cut of this technology for static photos. Microsoft and Google have similar functionality in the works.
Subtitling. Automatic subtitles for hearingimpaired people, both from speech recognition and lip reading.
Content production. AI should not be confused with intelligent creation, yet even here the edges are being blurred. Editing software such as Magisto, with 80 million users, takes in raw GoPro or smartphone shot video and automates the process of editing and packaging it with a narrative timeline, tonal grade and background music for consumers and even marketeers facing huge demands on their time and too much video to process and publish online. A number of recent developments in ML research will allow picture and movie content editing with Photoshop-like tools that edit conceptual elements of an image instead of individual pixels. It will soon be possible to directly edit facial expressions.
Ad insertion in movies. Mirriad, a Londonbased startup, has developed an application for product placement (replacement or insertion) in movie scenes depending on target audience, location, context, etc.
Content creation. In the near future, AI methods will be used to create new content. The first pioneering applications are already available. A documentary assembled by the Lumberjack AI system is hoped to be presented before the SMPTE-backed Hollywood Professional Association (HPA) by 2018 and has already helped create Danish channel STV ‘s 69 x10’ episodes of semi-scripted kids series Klassen. In the more distant future, AI will be able to generate new written and video content on the fly, based on the interaction with the user.
Content distribution. According to Berkes, new ANN techniques are being developed to beat traditional encoding and decoding algorithms for pictures and movies. “They will allow the transmission of high quality media content even in regions with low internet and mobile bandwidth,” he says. “ANNs are being used not only to build better compression methods but also to artificially clean up and increase the resolution of transmitted images (known as ‘super-resolution’; one example is from London start-up Magic Pony Technologies, acquired by Twitter last June for $150m to reconstruct a HD video from a low-definition, compressed stream).
While some AI critics imagine catastrophic scenarios, others like to imagine a future in which AI helps us all to be more productive and constructive. “Overall, end-users will benefit from the increasing role of AI, in particular in interacting with the media,” says Berkes. “Virtual assistants will understand their preferences and respond to vocal command, facilitating content discovery from multiple sources. As traditional media becomes increasingly connected, AI will enable content providers to interact with end-users. AI assistants will help consumers select personalised camera angles for sport events and they will deliver automatic summaries of the latest news and missed TV shows. He also predicts that AI will make creative tools accessible to everyone, assisting users in editing pictures and movies and creating musical compositions.
One fear is that AI will inevitably force humans out of work. This was realised in Japan last month when Fukoku Mutual Life Insurance replaced 34 employees with IBM’s Watson. “You still need people to train Watson,” emphasises Lomas. “That’s where the magic and specialism of creatives is of huge value. AI has a huge opportunity in the creative space. It doesn’t mean we don’t need creatives, it means they can move to higher level jobs and focus on being even more creative.”
No comments:
Post a Comment