May 31, 2019 / Méta-Media

How Artificial Intelligence is Transforming the Media Industry? – Part 1

By Kati Bremme

Algorithms are designed to solve problems. Artificial intelligence (AI) is a source of defiance for some and a miracle solution for others. It is omnipresent and impacts all industries. However, certain industries have more difficulty seizing the opportunities the technology offers. It’s the case of the media, which are less solvent and dynamic than the fields of finance and health to equip themselves with the tools they need to integrate the technology. PwC’s most recent AI Predictions report demonstrates these differences eloquently: 20% of all executives interviewed plan to deploy AI in their companies, compared to only 7% of all media outlets.

Nevertheless, there are vast applications of AI in the written press, cinema, radio, television and advertising fields: automation of trade processes and customer relations, social network monitoring and listening, information verification, predictive success analytics, video creation and postproduction, voice assistants and conversations, automated drafting, personalization, recommendation, optimization of content broadcasting, emotional tracking and accessibility.

By no way exhaustive, this two-part overview presents uses of AI throughout the information and entertainment media value chain. These are applications that could very well rebuild momentum within an industry that is reinventing itself.

1. Artificial intelligence as a tool for augmented information

Journalists are not the only ones who have feared being replaced by robots for a long time. Indeed, AI will replace certain tasks and lead to the obsolescence of certain trades. In that regard, the year 2020 will be a pivot year. According to Gartner, AI will eliminate 1.8 million existing jobs while creating 2.3 million new jobs. However, the future of journalists is in no way jeopardized. Although ‘robot reporters’ are already a reality and are used in many newsrooms to speed up the production process, they remain confined to very specific content typologies.

Since 2015, the Associated Press agency publishes dispatches that are produced by robot reporters for standardized financial news announcements. That same year, Le Monde called upon a Syllabs robo-journalist to cover the departmental and regional elections. With Heliograf, developed in 2016 for the Olympic Games, The Washington Post uses AI to ensure coverage of small-scale events such as local student sporting events that attract audiences that are too small to justify sending a human reporter to cover them. Finland’s YLE television outlet uses it Voitto bot to generate 100 articles and 250 images per week. However, cultural differences are observable in how newsrooms adopt new technology. It varies between Northern and Southern countries as well as between public and private media services, private services being more focused on a performance-based logic.

Robot editors? In reality, they don’t actually create, but rather assemble existing content that is then fit into predetermined templates. However, the technology is making progress and language generators are increasingly able to take the content into consideration to select the adapted format.

AI can also help journalists to analyze data and detect trends based on multiple sources of information varying from conventional open sources to new sources such as the data published by Wikileaks. With its capacity to scan and analyze massive quantities of data, AI enables the constant monitoring of trends on social networks and detect weak signals. In that, it contributes to the accomplishment of one of the public service’s mission: enabling the public to easily find the information it is searching for to be better informed. Associated Press uses NewsWhip to detect trends on Twitter, Facebook, Pinterest and LinkedIn. The Reuters agency uses News Tracer to detect trends and breaking news on Twitter as well as to facilitate the production of content. The system designed with Alibaba collects, categorizes, annotates and sorts news items.

Seeing as part of the journalist’s work is automated, AI forces us to rethink and reaffirm journalistic values to return to an ‘authentic’ form of journalism by taking individual users into consideration. However, beware of adding uselessly to an already enormous mass of information: the content generated by AI must remain relevant and that is only possible through smart collaboration between man and machine. The right balance needs to be struck between human judgment and automation, intuition, experience and creativity to increase efficiency levels when it comes to the collection, processing and validation of information.

2. Artificial intelligence as a tool to counter fake news

Although AI is capable of generating fake news, it can also contribute to detecting fake news. From false information disseminated by bots with a Slavic accent to Deep Fakes imitating Barack Obama delivering a speech, the progress made by the negative aspects of AI is impressive. So much so that Open AI recently put an end to its GPT-2 project because the AI had become so sophisticated that it ended up instilling fear in its creators. AI has sometimes been announced as a miracle remedy, including by Mark Zuckerberg at his first appearance before the US Congress in the wake of the Cambridge Analytica scandal during which he answered all of the embarrassing questions: “I don’t know, our AI team will fix it.” Of course, the truth does not magically result from Big Data. However, the technology used to fabricate a fake being the same technology used to detect it, AI is an important asset to counter misinformation.

As we all know, the problem with fake news is not so much that people no longer trust the media, but rather that they place their trust in any form of fake news. With its extensive analytical capabilities, AI can automate—at least partially—the verification of the information and the validation of the authenticity of photos/videos thanks to image recognition, metadata analysis, and real-time comparisons of information with databanks.

Combined to the blockchain, AI can also be used to authenticate information. Facebook more or less successfully uses AI to detect ‘semantic patterns’ that are reported to be characteristic of fake news. The blockchain-based Truepic and Serelay image authentication tools are both used by the Wall Street Journal’s team in charge of verifying information. Adobe uses an algorithm to detect altered images. DeepNews.ai is a tool that is largely aimed at aggregation platforms. It selects the more relevant news items on the Internet, and the algorithm then takes into consideration the breadth of treatment of the topic, the expertise, the analytical qualities as well as the means implemented—the whole based on a neural translation network.

Again, the algorithm is not a miracle solution. Most initiatives and tools function in combination with human beings whose abilities to analyze and verify sources—be it through a simple phone call—still exceed robots’ capacities. To optimize searches, algorithms are simple to train using a given content’s click rate data. However, that is of no use when it comes to detecting fake news. In this regard, the datasets used to train the algorithm to detect fake news must be coded by human fact checkers.

3. Artificial intelligence as a tool to improve conversations on the Internet

Hate speech, discrimination, violence, trolls plague the Internet. AI can use natural language processing (NLP) to automatically analyze contents, sort them and implement automatic moderation on a 24/7 basis. However, the automatic analysis of content has its limits. Even the most sophisticated forms of AI used by the platforms cannot prevent the dissemination of violent images in real time, as was recently witnessed once again during the Christchurch shooting. Platforms do not fully depend on AI moderation and instead moderate content through a combination of AI and human moderators. AI is not about to, and may never, replace humans in this regard because the technology is not capable of interpreting certain nuances such as humour, to the great dismay of Facebook’s human moderators for that matter.

Automatic systems are nevertheless a must to analyze the massive amounts of content available on social networks, detect nuisances, determine which contents are to be deleted (while referring to humans in cases of doubt) and even prevent the uploading of dubious content by preventing hated-filled images from being uploaded. The algorithms also enable the return of comments on sites whose editors had often shut down because of a lack of moderation tools. The New York Times uses the Perspective tool to assess the level of comment toxicity through keyword recognition. It thereby hopes to pass from 10% of all articles open to comments to 80%. The Guardian and The Economist have also adopted this tool.

AI can therefore be used to give audiences greater opportunities to express themselves by automating a certain number of tasks, without however replacing humans when it comes to dealing with nuances that robots are not capable of interpreting.

4. Artificial intelligence at the service of voice

Natural language processing and voice recognition have led to the development of conversational agents (chatbots, smart speakers) that are capable of dialoguing with humans. Already 20% of all searches are voice-activated (Meeker) and that percentage is expected to reach 50% by 2020 (ThinkWithGoogle). Voice assistants are a new media portal.

When we give voice commands to Google Home, Amazon’s Alexa or Apple’s Siri, AI is used to process our voice. This same neural network technology and Natural Language Processing can be used to design specific concepts and determine keywords that will trigger actions. Conversely, through Natural Language Generation, AI can transform text into voice. Hundreds of thousands of pieces of data are needed to train algorithms to translate our accents, dialects, outlandish formulations and other originalities of the language into mathematical formulas that can be understood by a robot. That is the reason why Alexa needs to listen to all of our conversations, according to Jeff Bezos.

The arrival of BERT (Bidirectional Encoder Representations from Transformers) developed by Google marks a significant evolution in the development of AI for voice: with an accuracy rate of 93.2%, computers are today capable of understanding language variables and can apply what they learn to a multitude of tasks.

Many tools are being developed to more effectively exploit voice, the most natural of all means of communication: Lyrebird is a Canadian start-up that creates ultra-realistic artificial voices as well as vocal avatars. Alexa now has the voice of a professional news anchor to read the news. Google’s AI is capable of recognizing a voice even if it never heard it before. The AI-boosted voice takes on the intonations and formulas of a human news anchor after only a few hours of text-to-speech training. Snips.ai proposes a vocal assistant service for professional builders that is fully embedded—regardless of the support—and respectful of users’ privacy. VSO is becoming the new SEO, a major issue for the media, and Google now proposes podcasts in its search results.

However, even behind Duplex, Google’s virtual assistant capable of imitating your voice and your faults to make appointments, there are 25% of human beings working in a call centre.

5. Artificial intelligence as a tool to create interactivity and engagement

In 1960, MIT’s Artificial Intelligence Laboratory created the ELIZA machine which simulated a Rogerian psychotherapist by rephrasing most of the claims made by the ‘patient’ into questions and then asking these questions. Thanks to AI, today’s possible interactions are much more developed. The chatbot originally used banks of questions and answers, but progresses made in artificial intelligence enable it to increasingly ‘analyze’ and ‘understand’ messages through natural language processing technology and to have machine learning capabilities. Whether it be to consume information or to interact with clients (Gartner Marketing predicts 85% of interactions will not involve humans by 2020), the automation of dialogue is increasingly sophisticated and personalized.

Also, the production of basic bots is more easily accessible: Facebook proposes a turnkey solution in Messenger, whereas platforms such as Omnibot, Politibot and Sently distribute plug-and-play solutions with specially dedicated media formats.

Whether bots are incorporated into messaging systems to directly reach users (1.6 billion WhatsApp users and 1.3 billion Facebook Messenger users) or developed directly in sites and apps, the conversational interaction represents for the media a means of proposing a closer user experience.

The chatbots automate the relationship, foster engagement and are immediately personalized. Quartz developed its Bot Studio to propose personalized conversational narratives. The Guardian has its chatbot since 2016, CNN and The Wall Street Journal use Facebook Messenger to disseminate information, and NBC shares breaking news through the Slack app. The BBC incorporated a bot in its articles for the purpose of interacting with the audience.

Interactive fiction content is also developed: The Inspection Chamber is a format created by the BBC to interact through conversation, StoryFlow proposes interactive audio stories targeting children, The Wayne Investigation is an interactive audio fiction available through connected speakers using Amazon’s Alexa. Alexa also adapts the stories of which you’re the hero into an audio version. With OLI, Radio France proposes night stories for the connected speaker in the child’s bedroom.

Beyond these examples, as a simple assistant or content creator, AI can lead to innovation with respect to storytelling in the advertising, marketing, film and audio sectors.

6. Artificial intelligence in extended reality (XR)

Thanks to advances in technology, chatbots are transforming themselves into virtual companions that are actually capable of discussing and debating. Artificial intelligence and virtual reality appear to be two different fields of research, but technological developments show that both fields are increasingly interconnected. Initially reserved for the gaming world, these new technologies are slowly entering the realm of audiovisual creation. AI will transform storytelling with virtual characters that are capable of advanced interactions with human beings.

With its Whispers in the Night project, Fable has begun to create virtual characters that are animated by artificial intelligence. Computer-generated animated drawings are augmented by AI and based on the same technology as the one used by Epic Games or Magic Leap for immersive storytelling purposes. Emoshape uses the “Emotion Processing Unit” (EPU) component to determine the user’s emotions in real time and enable the robots to respond with an emotional state that is in tune with the user’s. The technology even uses science to optimize the interactions and make them as realistic as possible. The MIT Media Lab customized a VR headset that incorporates a device capable of detecting the user’s emotions. The physiological capture module is comprised of electrodes that collect ‘galvanic skin response’ (GSR) data as well as of photoplethysmography (PPG)-type sensors that collect heart rate data.

Being as it is less apprehensive of humanoid robots than European countries, China has launched AI-boosted news presenters through its Xinhua news agency: a male version named Qiu Hao (that speaks Chinese and English) was first introduced on November 9, 2018 and was followed by a female version (Xin Xiamomeng) on February 19, 2019. Boosted by artificial intelligence and machine learning, they are able to autonomously comment live videos and read texts displayed on a teleprompter.

Next week, be there for the second part of the uses of artificial intelligence in the media industry!