We let AI organise the newsfeeds that got Trump elected. Now we’re letting it write the news.

Last year I made a note to myself to bring more balance into my writing about Artificial Intelligence (AI) which was a little negative and dark.

AI promises us great benefits in medicine, health and other fields. But all I was writing about was my worries about bias and automating unfairness. So it’s not ideal that my first AI article of 2019 is a worried one yet again, but here it is.

AI already controls your news feed.

Artificial Intelligence already organises your Facebook news feed deciding which stories should get shown to you and in which order. This can be exploited by fake news bots and enemy actors has been widely blamed for getting Trump elected and also influencing the UK’s Brexit vote.

Now, hard pressed media outlets – under severe budget pressure and pressure to be first with the news are developing AI to write, summarise and edit whole articles. These may be published without ever being seen by humans.

AI journalists are already mainstream with Bloomberg, The Washington Post and other major outlets using them.

In September last year, Bloomberg announced that their mobile app would showing summaries of longer news stories for busy readers. These single line summaries are created by an AI which the company claims comprehends the meaning of the story in order to generate a summary.

Last month, A Japanese newspaper announced it was using AI to produce summaries of articles without human oversight. The system saves 5 minutes of editor time per story.

Heliograf, the Washington Post’s news bot, produced about 300 briefs on the Rio Olympics of 2016. It has since covered U.S. elections and high school football game and produced over 800 articles in its first year. Rival, The Associated Press uses AI to write reports covering earnings reports from companies and publishing them within seconds of the report being released.

Until this month these AI bots have been charged primarily with summarising or drafting articles. But last week, OpenAI, showcased new research called “GPT-2”,  an AI system that can write whole stories and articles from scratch.

The system is “unsupervised AI” which means it learned from reading 8 million internet articles rather than being explicitly trained for the task. The team has put the system through a standard test, the Winograd schemas, a tough reading comprehension test where it set a record for performance. It also achieves near-human performance on the Children’s Book test, another test of reading comprehension.

Many of the journalists attending the demo marvelled at the highly convincing news articles it was able to write, mostly whilst loudly worrying about the fate of their own jobs and profession.

The system is by no means perfect. Prose can be rough and the articles get less coherent the longer they get. And while the AI can write news articles that are sometimes convincing, it can’t yet write true news articles; the quotes and statistics it usesare all made up.

So we’re on the cusp of an AI powered revolution where not only does AI control the order of and appearance of stories in our daily news feeds but it even writes the articles itself based on whatever it finds, or whatever data it is led to.

The scope for fake news production and promotion is vast.

This is a point not missed on the team an OpenAI themselves who took the unusual decision to showcase the work but not to release it publicly. It would be normal at this stage to release a public model that other researchers could interact with. They stated “needing more time for ethical considerations” as the reason.

“Due to our concerns about malicious applications of the technology, we are not releasing the trained model,”

OpenAI spokesperson

Such concerns go beyond just generating misleading news articles. OpenAI also worries about deception, bias and large-scale abuse. Malicious people might exploit the technology to impersonate others online, automate the production of faked social media content, and automate the production of spam or phishing attacks.

We’re on the cusp of an AI revolution and the vast majority of us have no understanding of even what it is, how it works or how we can make sure we get the best of it and not the worst. We have precious little time to catch up with what is going on before the technology overtakes us and we literally become just pawns in the matrix.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.