![Black and white image of a bearded man in a dark shirt and blazer, looking directly at the camera.](https://www.siliconrepublic.com/wp-content/uploads/2020/11/Neil-Brady-credit-Alan-Rowlette-Photography-718x523.png)
Neil Brady, co-founder of CaliberAI. Image: Alan Rowlette Photography
Digital publishing is likely to see regulators clamp down on defamation and hate speech, so CaliberAI is setting out to stop costly content before it’s published.
Neil Brady describes himself as “a mixture of technology and media specialist and academic researcher”. In all, he has almost two decades’ experience in news media, spanning journalism at The Guardian and curation at social news monitoring platform Storyful.
This was also once the domain of Paul Watson, who served as chief technology officer both at Storyful and at Kinzen, an app offering users a personalised news experience.
Brady’s father also happens to be Conor Brady, the man who held the editorship of The Irish Times for 15 years. The father-son team have now joined forces with Watson and others to form something of a news start-up supergroup in the form of CaliberAI.
‘There’s a delicate balance to be struck here between freedom of expression and censorship, but we think CaliberAI meets this perfectly’
– NEIL BRADY
CaliberAI was founded by the Bradys with Watson coming in as CTO. “Our computational linguist, Alan Reid, also has a decade of experience working in translation and language technology solutions, while our machine learning engineer, Gül Kamisli, came to us from the University of Oxford where she researched the use of neural networks to study gene interactions. Lastly we have a stellar advisory panel, composed of some of the world’s thought leaders in the fields of philosophy and ethics, editing, computer science and law.”
What this dream team has come up with is technology to flag defamation and hate speech before it’s published.
“Our main product right now is a browser extension,” Brady explained. “It’s a bit like a cross between the spellcheck function of a word processor and the notification functions of Grammarly or Textio. It warns a user when something with a high risk of being defamatory or hateful has been typed, prompting she or he to think before publishing.”
The defamation dataset
CaliberAI’s immediate target market is traditional news publishers, including newspapers, television and radio stations, as well as the internet platforms used to propagate their content. “They’re most at risk for defamation actions and suffering the direct costs as well as the ancillary ones associated with publication of defamatory and toxic content,” said Brady.
As with all tools based on machine learning, data is key to how this technology works. “We’ve taken best practice in the field of computational linguistics and used some novel techniques. The advantages of smaller, expertly crafted datasets as a compliment to big data has been crucial.”
With no existing ‘defamation dataset’ they knew of, CaliberAI built one from scratch, guided by the experience of Brady’s father and the advice of Dr Carl Vogel, associate professor of computational linguistics at Trinity College Dublin.
“We’ve worked hard to bake in compliance with existing regulatory standards like GDPR too, as well as keep an eye to others coming down the line, like the Digital Services Act,” Brady said.
As of early October, CaliberAI had just hit the minimum viable product (MVP) stage. The company secured five official trial partners – a mix of publishers, law firms and public relations firms – as well as more than 50 people in more than 30 news organisations testing the product on a voluntary basis. News organisations involved in the trial include Mediahuis, the Guardian, The Financial Times and The New York Times.
“We’ve recently collated their feedback and will be incorporating this as we progress from the MVP,” Brady added.
The target market for CaliberAI
With regard to the company’s target market, Brady said that, in Ireland alone, in excess of €30m was spent by NewsBrands members defending defamation action between 2010 and 2015.
“This is likely an underestimate, as it only counts what happens from the point of proceedings onward. Legal and other pre-trial consultancy and editorial fees, as well as loss of advertising revenue and the wider effect of reputational damage aren’t in that figure. We know of one large Irish publisher that has set aside €60m to address accumulated costs of this kind,” he added.
“Content also doesn’t have to be defamatory to cost, with toxic content increasingly likely to cause a reputational hit that will affect the bottom line,” Brady expanded. “Law firms and public relations companies are also very interested in what we’re doing. However, where we see a really big opportunity is in relation to social media and other internet publishing companies. The law in relation to legal liability for digital publishing is in the process of being significantly altered in the United States, European Union and elsewhere. In short, they’ll soon be operating in a much more accountable and costly environment.”
In the short-term, CaliberAI will continue to build technology to identify defamation as a core part of its product offering, while hate speech is part of the company’s long-term plans.
“Just about every government and regulatory authority around agrees that, in addition to strengthening liability for illegal material, there’s a need to address material that is legal but harmful or hateful,” said Brady. “The cause of such speech is multi-faceted, but as polarisation and extremism have chipped away at the centre, there’s a need to find a floor here.
“Going forward we will produce a body of classifications that will have been generated by the most experienced team around, when it comes to editing, ethics, law and tech. There’s a delicate balance to be struck here between freedom of expression and censorship, but we think CaliberAI meets this perfectly. It’s a safety net. An editor and lawyer in one, on your shoulder. The user can still ultimately choose to post whatever she or he wants though. It’s not a censorship machine.”
‘Accountability is coming’
In terms of investment, Brady said the company is almost two-thirds of the way towards its goal, thanks in some part to its investment from Enterprise Ireland as a high-potential start-up. The company has also availed of the Enterprise Ireland Commercialisation Fund.
“Enterprise Ireland, Trinity’s Technology Transfer Office and the Adapt centre have been outstanding in that way, and I couldn’t recommend them enough,” said Brady.
“Another €300,000 would allow us to ramp up the data annotation, deploy new product features, expand existing ones and strengthen our sales function,” Brady explained, but now the major hurdle is getting investors to see the opportunity CaliberAI offers.
“The main challenge has been getting VCs and some customers to take a longer-term, holistic view of the problem itself. Because public speech has been so consequence-free for so long, many don’t seem to understand that this time it’s different, and accountability is coming.”
Want stories like this and more direct to your inbox? Sign up for Tech Trends, Silicon Republic’s weekly digest of need-to-know tech news.