Five reasons why now is the time to be thinking about artificial intelligence in your newsroom
Reason #1: Artificial intelligence (AI) is actually about data
In her presentation AI: CRAP IN CRAP OUT, Agnes Stenbom, a machine learning (ML) specialist at Swedish publisher Schibstedt, illustrates an important fact about what ‘doing ML’ means by sharing some of the findings from a creative experiment done by Google, with the goal of finding out how human perception works — across the globe.
[Spoiler: if you ask people to draw certain objects, shoes will surprise you!]
Schibstedt’s Agnes Stenbom talks about data and algorithmic bias
In her 18-minute talk Stenbom further illustrates perfectly and practically what kind of issues you encounter as a journalist when working with AI: why use it, how to think about it, with whom to debate it, what experiment to do with it and when to implement it?
Data as fuel for AI
Let’s use an arbitrary metaphor: if AI is a car’s engine — then data is the fuel. You don’t need to understand the technical workings of neural nets or natural language processing (NLP), in the same way you don’t have to understand the inner workings of your car in order to drive it.
And while there is no universal agreement on a definition of AI or ML, everyone does agree that ‘useful’ and ‘good’ applications of AI start and end with the quality of their data.
Your first action: Address these hard data-questions over at datajournalism.com 🔒🆓.
To better understand how essential quality data is for AI applications, playing with the Teachable Machine might be a good way to test-drive AI:
Reason #2: AI can support human-centred thinking
To have an informed debate about AI and make responsible decisions for your professional future, you will have to look at yourself and your context so you can both question and educate yourself.
Research shows newsrooms are under-addressing these needs, as demonstrated in Findings from a 2019 survey commissioned by the International Center For Journalists (ICFJ):
On training demand, ICFJ Report from 2019, (Survey; n=4100, 149 countries)
Of horrors and hypes…
You probably know all about the perils and promises of AI and ML. Does GTP-3 ring a bell? According to a report by The Atlantic, GTP-3 is a ‘multitool’ that “generates long-form articles as effortlessly as it composes tweets, and its output is often difficult to distinguish from the work of human beings.”
While we’re at it, let’s list some of the other common, negative assumptions:
undecipherable blackbox;
biased — check!
not a magic bullet;
power is concentrated at a handful of tech proprietors;
an unlimited disinformation future of mankind;
But everyday AI life is not that bad…
Chances are, somewhere over these last few days you’ve been using an active AI system yourself, as a regular user of apps like these…
Example taken from ‘Introduction to Machine Learning Problem Framing’
While sorting vegetables might not be your day-to-day business, our own industry anticipated and already applies AI in very practical ways, on some very mundane tasks…
Exhibit A: Last year the Reuters Institute at Oxford asked:
Exhibit B: Actually, according to the director of that same institute, Rasmus Kleis Nielsen, this is how and where we’re seeing it today:
Exhibit C: Where we, as an industry our putting our most recent efforts, was shown by John Keefe (New York Times) on October 14, 2020, at a ONA20-conference session titled ‘AI in Journalism’(🔒). Keefe showed these numbers on AI Trends and Patterns in Journalism:
Projects by purpose:
Augmenting reporting capacity (45%)
Reducing variable costs (27%)
Optimising revenue (12%)
Engagement (5%)
News stories (5%)
Self-critique (5%)
Keefe told the online audience that around 130 projects from newsrooms all across the globe (hat tipping the folks at JournalismAI for this list) have applied some form of AI or machine learning already.
Examples of Applied AI in Journalism range from automating what content to push to augmented writing of headlines, from robo-generated stock market reports, housing market updates and election results to handling millions of data points to spot emerging patterns regarding the spread of mis- and disinformation.
At Quartz’s AI Studio they’ve boiled down the assessment of potential use cases to “situations and feelings you might have where machine learning could help:
We’ll never be able to read all of these documents.
What’s unique about this text compared to all the rest?
My eyes sting from searching these images for the same thing.
We need to find more records like these in a huge pile of data.
I could really use a heads-up before this happens again.”
The end-user’s needs
During that previously mentioned ONA20-conference session another interesting example of AI in Journalism was presented by Ashley Alvarado (KPCC) on how they managed to increase audience engagement by gathering, analysing (and answering) over 4,000 reader’s questions on Covid-19, using tools from Hearken.
Tools
There are many organisations- from all across the board- offering helpful, machine learning powered tools for journalists.
Built by the International Consortium of Investigative Journalists, Datashare is an application that allows you to efficiently search and organise your documents
When it comes to data and analytics, there’s Power BI from Microsoft, DataShare developed by the ICIJ, Pinpoint from Google and Tabula by the Knight Foundation. For comment moderation you can use Perspective API from JigSaw. If you need to automate content, go check Sophi.io and United Robots.
Basic concepts and experiments
If you do want to get a grasp of the core concepts, go take this Introduction to Machine Learning for journalists, developed by the VRT, YLE and the LSE.
First-hand, guided experiments. If you feel more like experimenting, Quartz AI Studio has done us all a favour by guiding you on every step along the way.
Reason #3: AI can make you a better journalist
Agnes Stenbom is great at explaining the concept of algorithmic bias to journalists. But she also explains how she is currently using AI to understand, identify and mitigate bias:
To know what type of org you need to be in, what kind of people you want to work with, and what role and responsibilities you should take up, consider your professional context for a moment.
Emotional intelligence
During another ONA20 session, Journalism’s Technology Problem, panel member Mark Little mentioned thinking about emotional intelligence when it comes to working with new technology.
Question: How can you take away (unfounded but very real) misperceptions and fears about any new technology from people in your newsroom?
These fears can come in many shapes and forms, ranging from job stability, changing responsibilities, taking orders from machines or non-journalists and/or having data — the dashboards — take-over the creative work — the white- and storyboards.
Cross-functional teams
Now that data scientists, engineers, marketeers, interaction designers and publishers are all joining core editorial teams, we are going to have to accommodate for all these different types of specialists on the same team. Professionals who come from different backgrounds, have different expertise and bring diverse modes of work and cultural practices.
As journalists, we need to take all this into account and explain to everyone involved why and how this is a boon to journalism.
Finding common ground and overcoming differences is hard and complex and takes a lot of patience. Once you are able to freely, openly discuss and collaboratively articulate potentially valuable uses and fears about AI, there needs to be buy-in from leadership too.
Reason #4: AI informs your overall tech strategy
How will this new technology affect current structures in your org, the processes and the workflows? What will it do for your products; when do you need to change the responsibilities and make-up of your team(s)?
What kind of executive do you need (or: need to become)?
In other words: what’s your strategy here?
The implementation of AI can be your ‘vehicle’ for a greater cause: to create and/or foster a culture with a shared vocabulary, shared goals and unambiguous principles.
It might be a bigger challenge than ‘doing AI right’. And this time it could just be the right moment for you…
Reason #5: You’re late… but not too late
Somewhere during that seminal ONA20 AI in Journalism session, a panel member mentioned that most efforts are done by global and national newsrooms. You seem to need to have a certain scale (and resources, obviously) in order to pull it off.
Question: Will machine learning lead to a new chapter in the digital divide?
Au contraire! I sense there’s a somewhat unfamiliar but fruitful open culture emerging in newsrooms (and online spaces) these days, when it comes to eliciting feedback from co-workers and sharing best practices with others.
Vital donors
My feeling is, this trend is caused by a growing respect for, need of and familiarity with engineers, designers and marketeers in the newsroom, and that those ‘non-journalists’ brought their work ethic, cultures and iterative ways of creating things to the newsroom, planted it there and allowed journalists to take a few pages from those playbooks.
Most of the machine learning pioneers in these newsrooms are more than willing to educate others. They love to talk about their mistakes and are intrinsically motivated to collaborate and share their code, their data, their workflow and their set-up of infrastructure for ‘doing AI’.
This attitude lifts all boats and helps us all to have an informed debate, learn faster, and get everyone to navigate yet another paradigm shift for journalism.
There still is time to come aboard; the barriers to entry are relatively low.
#Bonus: a nifty little action list for takeaway
If you want to get into the action straight away (in no particular order), start with these ten things first:
Play with it:
Teachable Machine
Karen Palmer’s RIOTUse the tools that are mentioned;
Take online courses:
Helsinki University’s Elements of AI
GNI’s Introduction to machine learning, for journalistsWatch videos:
AI: Crap in, Crap out
ONA20's Emerging Technology TrackStay updated:
Collab diaries from Mattia and Charlie over at JournalismAI
Read reports from Reuters, Polis/LSE and ICFJFollow people on Twitter.
Discuss; assess potential uses; team up with people in and outside of your newsroom.
Start your experiments at Quartz AI Studio;
Do some further book reading:
‘Automating the News’, by N. Diakopoulos
‘Newsmakers — AI and the Future of Journalism’ by F. Marconi;Inquire with me if you need help or feel lost:
info[at] laurensvreekamp[dot]nl
Laurens Vreekamp is an Associate Consultant at Fathm. If you are interested in how Fathm can help your organisation, please email us at hello@fathm.co