We need to talk! About artificial intelligence and ethics in journalism

2018 — Year of Artificial Intelligence: Elon Musk warns in Austin of the dangers of autonomous weapon systems — a few months later the Cambridge Analytica scandal shakes the world. There are people out there who say that the unethical use of AI is the real threat to humanity.

Artificial intelligence is all about digitizing human knowledge and human abilities. What many people are not aware of: media companies are already using artificial intelligence in many different ways.

The Washington Post has been trying for years to find out to what extent automation can be useful in journalism. Their Bot Heliograf, for example, analyzes data, records trends and then enriches templates to finally create short stories. This works quite well for sports, finance and election reporting — in other words, it is already used in domains where there is enough structured data available. Quakebot, a bot of the Los Angeles Times, works in a similar way, generating reports on regional earthquakes based on data and text templates practically in real time.

Other media outlets are also increasingly using AI: AP, for example, has automated the production of earnings stories, CNBC and many other newsrooms are using semi-automatic video creation platforms such as Wochit to to produce engaging social videos, and Al Jazeera is considering having reporter robots on the ground to report from war zones.

Further applications of artificial intelligence in journalism:

• Recommendation widgets based on user movement data for videos and other content (e.g. BBC)
• Smart text-to-speech features for voice assistants (e.g. DW)
• Active and passive personalization of news content (Washington Post, FAZ), paid content purchase offers based on probability models (SZ) and advertising based on usage patterns/profiling (Outbrain and other advertising marketers)
• Social listening: Early detection of trends and news on the net and on social platforms with the help of listening tools (e.g. Listening Center RP online, Crowdtangle)
• Information extraction, clustering and visualization of complex, unstructured data masses, e.g. by machine learning (e.g. SZ/Paradise Papers)

Learning to talk to each other

What really fascinates me is Natural Language Processing, NLP for short. Some of my colleagues at Deutsche Welle have been working on this technology for quite some time. NLP’s goal is to enable people and machines to communicate with each other at eye level. The way chatbots or digital language assistants work is based on this principle. This year NLP has really made a lot of progress: The prototype of the tool news.bridge permits an editor to transcribe a video using speech-to-text technology in a matter of seconds. In the next step, the tool translates the script (in still very different quality) into all possible target languages (e.g. via an API to Google Translate) and finally does a new over-voice using synthetic speech generation.

A demonstration I witnessed at Salesforce in San Francisco this year was similarly magical: a machine that can summarize a complex text almost in real time — Chief Scientist Richard Socher himself says that he hadn’t thought this possible until recently.

Machine generated texts are not (yet) a Pulitzer Prize suspect, even though the boundaries between human and automatic authors are blurring faster and faster.

But still, many advantages for the media outlets are pretty obvious: automation saves time and enables the “repackaging” of journalistic products for different platforms — and this is urgently needed, considering how rapidly the number of channels and platforms that have to be served simultaneously and specifically has risen in recent years.

Automation brings page views

To the extent that automation creates new opportunities to publish similar content quickly and cost-effectively on different platforms and in different languages, the possibilities of scalability in journalism improve. Or, to put it less complicated: automation brings — if things go well — higher engagement rates, longer retention times and more page views. And this is vital for survival in times when the business model of many media houses is still based on reach. “Right now, automated journalism is about producing volume. Ultimately, media companies will have to figure out how to go beyond the pageview,” says Seth Lewis, journalism professor at the University of Oregon specializing in AI.

But even those who focus on impact instead of reach do see competitive advantages by using AI: There is already evidence of a significant increase in audience engagement as soon as content is personalized. Users appreciate it when media content is tailored to their interests. According to an Accenture study from 2017, around 77% of users of digital news expect some form of personalization. Perhaps because it makes it easier for them to cope with the overabundance of information out there. Or perhaps because they are used to comparable forms of personalization from social networks such as Facebook.

No personalization without data

However, no personalization without data. Facebook has shown the way. The company collects so much data — even far beyond its own platforms — that it knows pretty much everything about its users. Media organizations are still lagging behind, but are also tracking the behavior of their users, as this short excerpt from the Washington Post’s Privacy Policy (dated 24 May 2018) makes clear:

“We may collect personal information about our users in various ways. For example, we may collect information that you provide to us, information that we collect through your use of the Services, and information that we collect from publicly available sources or third parties.”

Not many users are aware of the fact that established media companies are now trying to collect data just as intensively as technology companies, which for this reason are repeatedly criticized by journalists — and rightly so. It’s all about anonymous data and — even more valuable — personal data.

News organizations however are struggling to get their user’s consent to track their data — especially here in Europe, where the GDPR is aiming to give individuals full control over their personal data. While the large platforms require registration from the outset and tie the extensive transfer of personal data to this, media users hardly reveal anything about themselves when reading articles or watching videos. This only changes once users get persuaded to subscribe to a newsletter, join a reader club or even use paid content. In one of the few articles on this topic, Natasha Singer from The New York Times writes: “Many other companies, including news organizations like The New York Times, mine information about users for marketing purposes. If Facebook is being singled out for such practices, it is because it is a market leader and its stockpiling of personal data is at the core of its $40.6 billion annual business.”

We need to talk about trust!

Is it the same for you? At this point I have a hunch that we need to talk: About the way we — the media — collect user data, and above all about how we deal with this data. In the end, nothing less depends on it than the most important asset of media companies: The trust of our users.

I did a completely non-representative survey in my personal environment: Not a single person agreed to receive personalized content without their explicit consent. And not a single person had previously considered that the information provider they trusted could theoretically pass on personal data to marketing partners — especially if that person had previously paid for a digital subscription. In short, while many of my German friends are deeply suspicious of social networks, they all trust that established media providers will be much more responsible with their private data. We media thrive on this trust. People spend money and time on our products because they trust us. Because they assume that journalism serves society — regardless of whether it’s privately or publicly funded.

Still 70% of US citizens have “a lot of trust” or “some trust” in News Organizations. In Germany, 75% of respondents continued to trust the media this year (2017: 75%). Public media are by far the most trusted, followed by the national daily newspapers.

After all, 48% of respondents in Germany explicitly do not want the media to pass on their data, while around 40% see this as less problematic if, for example, they can continue to receive free services. I personally assume that people’s sensitivity to the value of their data will grow in the future — also fuelled by a growing distrust of social networks, whose reputation has suffered greatly in recent years.

So what needs to be done?

1. We need to have a debate about the ethical dimension of artificial intelligence in journalism. Up to now, I have not been able to find a single code of conduct that would contain principles on the ethical use of AI in journalism. Furthermore, there is no open debate about which ethical principles could form the basis for such self-imposed rules. And one more thing: In contrast to traditional code of conducts, agreements upon the use of artificial intelligence in the media are not something that journalists can debate amongst themselves. There needs to be a comprehensive debate that includes the technology and business side as well.
2. Ethical aspects should be included in product development. More and more media companies are deploying agile product development to get to market faster with innovative products. Artificial intelligence is an increasingly common component of these products. This means that there must also be room for an ethical debate in product development: What consequences does it have, for example, if we personalize news content (keyword: filter bubbles)? What would products look like that offer users broad access to information while being attractive enough to prolong their stay and allow the media to monetize high-quality content?
3. We journalists need to become more competent and must not leave AI to data scientists and software engineers alone. Code is never free of values, because it is ultimately written (or commissioned) by people who pursue a particular goal. Machine learning, too, is done with the help of data that people have chosen. So, how can we prevent the bias and prejudice it may contain from entering our reporting and thus becoming entrenched?
4. We need to approach this debate from the perspective of our users.What do they need to continue to have trust in the media? What do they need from us to remain convinced that journalism serves the society?

Since the takeover of Jeff Bezos the Washington Post proudly calls itself a “media and tech company”. Only this strategy has brought the newspaper into the black — and has made it possible until today that socially relevant journalism can emerge, which is committed to the common good. In September 2018, Marc Benioff — founder and CEO of Salesforce — and his wife Lynne bought Time Magazine. Benioff doesn’t just have gigantic AI expertise (which is probably good news for the paper, which is struggling with circulation and advertising shrinkage). He is also a CEO who is committed to the common good and who says of his new journalistic acquisition: “I feel our values are aligned. Trust is my highest value and it is Time’s as well.”

He is thus setting the tone for a debate that we must now move forward with. Artificial intelligence in itself is neither good nor bad. But if journalism is to remain committed to society, we need to talk about why and where we use AI— and how we can maintain the trust of our users at the same time.

Or, as Benioff puts it:

“Are you committed to the state of the world? And committed to improving the state of the world? Or not? You have to choose. You have to choose.”

NOTE: I have also published this article in German.