AI is an existential threat to the publishing industry, and companies must agree a strategy on how to protect their content, says Lucky Gunasekara, co-founder and CEO of search personalisation service provider Miso.ai. But, he tells Ashley Norris, AI-driven content is also a significant opportunity.
- ChatGPT has created a “Napster moment” for the content industry.
- Media companies must come together to create a road map for how to use AI.
- If media companies don’t have a “curious culture” they may find AI difficult to integrate into their workflows.
There’s an AI asteroid coming for all of us, says Lucky Gunasekara, co-founder and CEO of Miso.ai. And now we must sit “in the same room and figure this out”.
Lucky has been working with AI for over a decade. By harnessing and tweaking open source technology, Miso has created AI solutions for publishers. Answers, for example, is a chatbot that generates summaries based solely on the content available on a media company’s site.
He is an outspoken advocate for publishers working together to meet the AI challenge, and says direct conversations and difficult conversations are critical.
Lucky spoke to us at an Mx3 AI event in London.
What is your role in Miso?
I originally trained as an interaction designer. I lead product work with our team and think strategically how can we apply AI to publishing in a way that is private, profitable and sustainable for teams to adopt.
How long has AI been on your radar?
For around 15 years. It was the first thing I looked into after dropping out of school in my 20s. I’ve been fortunate to work on natural language processing, machine learning and deep learning since then. Initially on search, and then other applications with news media.
Tell us about Miso: What are your core products, and who do you work with?
We mainly work with medium to large publishers, groups that have five to 10 brands. We have a product called Answers, which is like an actual language-generative search engine that answers questions or inquiries from users based solely on the publisher’s content. It basically flywheels off a lot of data that we use for the first-party data programmes, programmatic for advertising, contextual advertising, content, recommendations, personalisation, newsletters and more.
Are you finding that one part of the world is embracing AI quicker than others? Is the US ahead of Europe and Asia?
The US is certainly one of our largest footprints right now. But the UK is growing super quickly. It’s already number two. It could be number one by the end of next year at the rate at which things are going.
We’ve already got deployments in Taiwan and Japan, as well. So, I don’t really think of AI as a regionally specific trend. I think it has a lot to do with the maturity of the publishing groups in the different regions. Do they have product managers? Are they used to doing A/B testing? Are they data-driven? And do they have an appetite and a willingness to experiment and dive in with AI?
Or are they more risk averse? They see this big wave coming and think, I don’t want to get in the water. I don’t want to drown. I’m just gonna stay on land where it’s safe right now. And I think those are usually the dynamics that we notice and pay attention to.
Was Chat GTP’s launch a catalyst for the growth of your AI product? Did it put AI on many more people’s radar?
The thing that ChatGPT has done is it’s educated the world very quickly on what AI is, in a broad sense, and what it likely will represent in the future. Everyone, at a certain level, now understands this idea that machines can talk to you and be largely conversational. This wasn’t the case two years ago. I think that that has been a huge catalyst for everyone.
In publishing, it has also been a wake-up call as this happened under everyone’s noses. Media companies’ content is being scraped by the bots, and that has awoken everybody to a much deeper existential threat from the the AI platforms.
You suggested that publishing is having a “Napster moment”. What did you mean by that?
People sometimes look at ChatGPT and think it might be a fad. And this will go away. But I don’t think so. Consumers are very quickly getting used to this idea that they can talk to a machine and get answers to questions that are top of mind without having to go to a site and having to click on a search link.
This has very similar parallels to what happened to music in the late 1990s. With Napster, suddenly, you didn’t have to buy a CD to listen to a song. You just downloaded it and put it on a device. And that type of shift was really profound.
There’s a legal path to fight it, and I think the music industry did that successfully, but the reality is that consumer behaviour has already shifted. And it’s going to continue to shift. So publishing needs to accept that that shift is happening and then recognise that the existing business model of subscriptions and display advertising won’t work if people never go to your site in the first place. And so the question mark is, “What’s going to come next? And how do media companies get paid to deliver answers as they are a trusted source of information with readers?”
I think it’s fundamental to have a direct conversation with these platforms. “Have you generated an answer to a question? And did you use our information to do it?”
The longer they delay having difficult conversations, the more the tech platforms will progress, and the more consumers will get used to this. And unless there’s swift action, they’ll get disintermediated further.
Are the publishing companies moving fast enough to get together and work out a solution to AI accessing content?
It’s way too slow. I’m not advocating for a frenzy of lawsuits. But taking very conservative collective action through the courts and lobbying and making a clear point of view known about what this represents, and the economic and societal harm it’s going to pose, is really important. Relative to that risk, there’s just not nearly enough collective action.
What would your five-year plan be if you were a publisher?
I would start implementing AI on my sites immediately, taking the short, medium and long view of how to adapt to this consumer shift towards answers and this mode of consuming information in this way. Our articles would have AI components to them such as personalised summarisations or text-to-speech.
I would also start thinking about search in a completely new way that parallels what Bard, Perplexity and others are doing.
I would think about our newsletters differently to be AI-mediated alerts and briefings for readers. There’s a lot that you can do within the four walls of the site itself.
I would also be calling every publisher I know, and every association, saying, “Hey, let’s have dinner on Thursday, and talk about what this means for all of us. It’s an asteroid coming for all of us. We all have to sit in the same room and figure this out.”
Tell me about Shiva?
Shiva is our internal logic language model. It’s built with open-source roots. We started using Llama Two, which was open-sourced by Meta as their contribution to the open-source community. Then we basically fine-tuned it to do questions and answers with a high focus on reasoning and verification.
That model is now the bedrock of our platform for search called Answers. It’s doing super well at reasoning, and it has allowed us to bring the unit economics down for Answers to the point that it’s actually default profitable for publishers to use in the first place. It would have been very difficult to do this with larger and more expensive closed-source models.
So, it is cheaper and faster than the AI used by the big tech platforms?
Part of it is that we’re focused. We’re not trying to build an AI model that achieves artificial general intelligence. In the long run, we’re just trying to build a large language model that can intake questions, do research and produce a cohesive answer that’s helpful, factual, and up-to-date for the reader who asked the question in the first place.
Focusing on the problem set allows you more leverage and to bring those costs down because you’re not trying to build an everything model for the entire internet.
We didn’t try to build it from scratch, we’re building on open-source roots. Every few weeks, when we see a new open-source model or a new fine-tuning technique, we’re in the lab, testing it, benchmarking and progressing it. This means I don’t have to spend tens of millions of dollars. I’m building on top of this amazing community that is doing amazing work right now.
How do you think Shiva will develop?
I think there’s going to be a very quick progression. We’ll probably see a version of Llama Three.
People will realise that closed models are difficult to adopt. In certain respects they’re very expensive to take to scale, the unit economics may not align, and they don’t have necessarily the control over the data and the models they need at a deeper philosophical or existential level. So, I think you’ll see more adoption of either open source directly by small teams that can pick it up and run with it or the adoption of platforms that are built on open source technology.
- The interview has been lightly edited and condensed for clarity.
Media Makers Meet – Live
We’d like to see you at our upcoming live events in Spain and Portugal
Mx3 Barcelona focuses squarely on innovation in media, emphasising creator-led, consumer and B2B media operating in and across media verticals. It takes place from 12-13 March.
Click here for more.
The FIPP World Media Congress, which we produce on license from FIPP, takes place from 4-6 June in Cascais, Portugal. The magazine media world’s flagship global event will be the 46th edition since it first launched in Paris in 1925.
Click here for more
To read the full takeaways from our December Mx3AI event held in Hoxton, London, download the MX3AI Report.
