Microsoft’s latest Internet innovation might have something to say about Google’s and Facebook’s ongoing efforts to fight fake news by using artificial intelligence.
Trouble is, Congress is unlikely to want to hear about it now as it presses Big Tech to get a handle on bogus content before the 2020 campaign begins in earnest.
That’s because Microsoft, the operating systems giant transitioning to a cloud-services business model, has created a bot that uses AI to write comments for news articles posted online. Dubbed DeepComOpens a new window , the open-source algorithm is designed to drive traffic to publishers by generating reader debates that attract the notice of search engines.
By tempting consumers to respond with comments of their own, DeepCom can help push news and feature stories higher up on lists of results. That’s because the algorithms that power Google, Microsoft’s Bing and other search services count comments as content and take them into account when compiling their responses to user queries.
Pandora’s bot
As such and paradoxically, Microsoft opens a new avenue for just the sort of fake news that Google and Facebook are taking pains to root out.
Recently, the tech giants have gone to war against deepfakes, the amalgams of images and speechOpens a new window that provocateurs can use to make contrived clips of politicians and others that look and sound authentic.
Deepfakes along with chatbots – made up of fake social media accounts and shadow domains that mirror legitimate websites – are among the digital weapons employed by Russia and other rogue nationsOpens a new window to influence the outcomes of Western political campaigns. With the 2016 election of Donald Trump among the highest profile of these elections, many members of Congress are trying to keep future votes safe.
Given the reliance on AI to produce facsimile clips of speeches and interviews that look and sound like the real things, Google is using machine learning to ferret out the falsified footage. At the end of last month, it released a massive deepfake dataset produced in conjunction with its Jigsaw incubator for IT security.
Battling benchmarks
The Alphabet subsidiaries filmed paid subjects and created deepfakes using the range of available algorithms since the technology first appeared in 2017. Researchers at universities in Germany and Italy are building the contribution into the FaceForensics benchmarkOpens a new window of 1.8 million images that they use for deepfake detection.
Citing similar datasets and in cooperation with academic partners, Facebook’s Deepfake Detection ChallengeOpens a new window launched last month aims to release its own benchmark in December. The social media conglomerate, which was roundly chastised by Congress in June and righteously spoofedOpens a new window in a deepfake of founder Mark Zuckerberg on the eve of his testimony, is funding collaborations among challenge participants.
Microsoft is participating in the Deepfake Challenge. Its Chinese arm worked with developers at Beihang University in Beijing, an aerospace institute that is home to China’s exascale computing efforts, to create DeepCom.
Going both ways
The system uses a pair of neural networks – one for reading and one for generating feedback – to digest headlines and content. It predicts the points and people of interest in the items with AI and then uses that information to produce commentary that can range from applause to provocation.
A paper co-written by members of the team that created DeepCom fails to address potential problems caused by malicious users, including the agents of foreign governments who might adapt the algorithms to drive sites spouting propaganda and disinformation to the top of results pages.
And the DeepCom team has made no effort in subsequent revisions of that document to address mitigation of the threats. As President Trump’s trade war with ChinaOpens a new window drags on, DeepCom will likely draw the attention of politicians and regulators.
When it does, it may well be too late. DeepCom debuted months agoOpens a new window on GitHub, the Microsoft-owned code repository, meaning that the software is available for good or ill to members of the global development community.