afaqs! news bureau
Digital

Should publishers be worried - or excited - about AI?

There is increasing speculation about how artificial intelligence might affect publishing. Four digital minds examine the issue.

As far as topics with respect to digital media go, probably the most fascinating is artificial intelligence (AI). Talking about AI, the first thing that comes to mind is the Cambridge Analytical scandal. It revealed how Russian AI-powered powered fake news had the power to steer the US 2016 presidential campaign. So, AI is the stuff of thrillers. But there is more to AI than Cold War-ish games.

Digipub World 2018's session discussion on the impact of AI on publishing saw Chirdeep Shetty, CEO, Quintype, Rohit Prakash, Engineering Lead, PocketPills and Sandeep Amar, Founder, Inaaj not only simplifying AI but also explaining what AI can do - and is doing for - publishing and publishers and editors. The session was moderated by Colin Morrison, Flashes & Flames.

Should publishers be worried - or excited - about AI?

The panelists (from left) Sandeep Amar, Chirdeep Shetty, Rohit Prakash with moderator Colin Morrison at Digipub World
Click on the image to enlarge

To start at the beginning, what is AI actually? Is it more than traditional coding? Shetty starts with an example. "If you were trying to recognise cat images you would start with edge detection (an image processing technique for finding the boundaries of objects within images) looking at ears, eyes, round face, a furry ball. That was explicitly coded by engineers. AI converts that kind of recognition into a mathematical equation that the machine will learn and get better by itself. For this to happen you need to have a large data set and huge computing power. What you do is give millions of images of animals and in the output say that these are images that have cats. And the machine itself will start learning and getting better. That is how AI is different from traditional forms of coding," he explained.

Prakash cut through the glossary jungle - four entirely different terms like AI, machine learning, deep learning and big data - that is thrown about by most people who evenly faintly know of AI. "Artificial Intelligence," he clarified, "is something - anything - that mimics human intelligence to solve problems using computers. It can be intelligence guesswork, some kind of statistical inference or, as Chirdeep said, we can learn from lots of data and try to mimic what a human would do."

Music-lover Amar first came into contact with AI when listening to Pandora Radio Service. Powered by the Music Genome Project, Pandora is a music streaming and automated music recommendation internet radio service that tries to guess what song, a user would like to listen to. This was based on the listener's earlier preferences. Listen to four songs and Pandora would throw up a fifth option it thinks the listener would like to hear. The Music Genome Project was a Ph. D project started in 2004 by three people.

"They were doing it with 'machine learning algorithms' which are pretty generic," said Amar, adding, "but they were failing in it because they were attributing the songs based on some kind of software. And they assigned attributes (like it is 20 per cent jazz or 30 per cent blues and 10 per cent some other genre) to songs. Their conclusion was that manual attributes with machine learning was the best way to arrive at the outcome." However, at times, the outcome has to be refined again and again.

That AI is here to stay is not under dispute. Nor the fact that it is making its way into many areas. Take entertainment. Netflix uses AI to great effect with its personalisation for viewers. Shetty quoted Jeff Bezos to underline the importance of personalisation. If Amazon has 4.5 million customers, they need 4.5 million home pages. Not in terms of layout but in terms of content. Each person needs to see different content. Shetty felt that the publishing world has started reading from that. Since e-com has always been at the bleeding edge of AI "we are seeing that taking shape in media and entertainment as well. Something that Netflix is doing well is that it has personalisation not only with content and content recommendations, but even personalisation-based jacket images. Let's say you have been watching Sacred Games. Now, I may have Saif Ali Khan on the jacket I get to see on screen, while you could have Nawazuddin Siddiqui. It is very personalised to each person's preferences. Personalisation of content means what to recommend, in what form to recommend it (whether long-form content or short-form content) and at what time it should be delivered to you."

Morrison wanted to know if "all of us recognise the fact that what we are mostly talking about is a way of personalising media. Is that the focus really of AI - that my media is what I want?"

Shetty felt that AI can do a lot more than that on data gathering - like fact-checking for fake news. Amar's felt that AI has been glorified a little too much. "The personalisation on Netflix is pretty bad. It is more of collaborative filtering. For instance, if I like three particular movies and Chirdeep and Colin like three but there are two that are common to all of us. Netflix would try to show the third movie (that is not common) to me. The problem is that I like John Hughes movies, but Netflix does not have those. So how can it show me a Sixteen Candles?"

He recounts his experience with Outbrain, a content discovery platform. "When I met their guys in New York, they had 200 data scientists and they said they could customise ads. And that we could put their solutions under our article. But what appeared were the same ads - Alia Bhat in a pink bikini from Shandaar - time and again. It happened even the article was a political one, say, about Mr Modi. They only had a finite number of ads."

"Does that not say then that we need AI at the back end as well as the front end," queried Morrison.

According to Amar, publishing businesses have not matured in terms of using AI. "They are about breaking news," he said. So, while the algorithm may be good, the data is not good enough.

But writing is all about writing. Can AI create content? "Publishing is probably a two-sided platform. There is content creation and there is content consumption. We have covered content consumption through discovery and recommendation," said Prakash, and went on to add, "but AI can also help create and augment to create better content. Say, you are writing a news article. With AI, the computer can detect grammatical errors and prompt you then and there. We can also have checks for plagiarism in place. It actually helps you create better content."

There have been instances of AI being used by newspapers. Five years ago, when Bezos took over the Washington Post for $250 million, it became evident that change was in the air. In 2016, taking a leaf out of the Amazon book, the Post used AI-based algorithms to write articles. Powered by Heliograf, a technology developed in-house, the Post carried 850 AI-powered articles - this included 500 articles around the election that generated more than 500,000 clicks - last year (source: Digiday). Associated Press' strategy manager and AI co-lead has been quoted in media saying that "in the case of automated financial news coverage by AP, the error rate in the copy decreased even as the volume of the output increased more than tenfold." AP went on to announce that AI freed up 20 per cent of a reporter's time spent covering corporate earnings.

Shetty has another example, this time from the FIFA Football World Cup. "Fox Sports had access to the entire footage of all the World Cups from the 1950s. It partnered with IBM Watson to start analysing the entire set of footage, to identify where goals are scored in each of these matches. They could then identify where goals were scored, when and how penalty kicks were awarded, where and when there were red cards. A human could have done it but it could have taken weeks," he explained. Does that make AI as intelligent as an average human?

Chirdeep quoted Benedict Evans to counter that. Evans, a partner at Andreessen Horowitz (the Silicon Valley-based venture capital firm), had said that "today, with machine learning, the computer will match a 10-year-old and, perhaps, a 15-year-old. It might never get to the intern. But what would you do if you had a million 15-year-olds to look at your data? What calls would you listen to, what images would you look at, and what file transfers or credit card payments would you inspect?"

Chirdeep pointed out that these young men and women could have been used to scan and make contextual summaries of, say the Panama Papers, which a media organisation had got hold of. One of audience had a query. "Are publishers using AI to create better content and create efficient digital publishing?" According to Shetty, it is bit of both - experimental and proper use. "Many start-ups are doing something with content gathering," he said.

Also, can publishing learn about how to use AI from e-com? Prakash agreed that there were parallels. "If you talk about collaborative filtering, it can be used at any number of places. You could find out similarity scores between different things - between movies or between two people. But the exact application would depend on the use case," he concluded.

Replacing a human with AI - unlike the automated cars business - in digital publishing could remain a dream. In the former, it will be more of a case of AI assisting journalists and writers. It can make their jobs easier but will not be able to take over what they do. So, white collar workers, after all, will be safe especially because there is a case of creativity.

Could this report have been written using AI instead of a reporter spending a couple of hours deciphering the discussions on stage? Take a guess.

Digipub World 2018 was partnered by - Timesnownews.com (platinum partner); Akamai Technologies and Facebook (silver partners); and Freshworks, Vidooly, comScore, Quintype, Times Internet, and 24 Frames Digital (bronze partners).

Have news to share? Write to us atnewsteam@afaqs.com