How can AI create marketing content for you now quickly and cost effectively?

Hello and welcome to this episode of Speak PR. I am really delighted to have Stelios Kampakis with me today. And Stelios is, or should I say Dr. Stelios, is on a mission to educate the public about the power of data science, artificial intelligence, and blockchain. He's a member of the Royal Statistical Society, an honorary research fellow at the UCL Centre for Blockchain, and a data science advisor for the London Business School. So he's also the CEO of Tesseract. Wow, Stelios, that's a lot. I'm so delighted you've had time to come on the call with us.

Yeah, happy to be here.

Now, Stelios, as you know, SpeakPR is all about helping business owners to get noticed, and especially using tools like, like AI. Tell us, what's your experience, how AI can help business owners in their marketing?

So I think AI can help in multiple ways in marketing, and we only recently started to discover some of these ways. I think that, I mean, there are many use cases, the obvious one being around analytics and trying to answer some traditional marketing problems from improved ways to do A/B testing, to understanding attribution in a better way. But then we can move on to some fancier ways, or fancy maybe is not the right word, like content generation, for example, where I think it's one of the— this is one of the hottest areas of application that we're seeing some tools coming out. Over the last couple of years or so, and I predict we're going to see some very interesting things happening in the next few years. By content generation, I'm referring to anything from articles to titles to images, even to videos. I saw a company the other day where you write the product, it's basically you provide a script and then it creates a video of that script. Very, very interesting and impressive. I mean, No one knows where we'll be in the next 5 to 10 years, really.

What do you think, though, is the sort of the challenge for the smaller business that wants to use AI? Because it's— the danger is it's quite complicated, isn't it? I mean, people have got to understand how to deploy it. How do you think they can do that?

Yeah, I mean, it really depends. As in, so you see, a large part of my work over the last few years has been around explaining AI to decision makers, especially startups, but also some bigger organizations, and helping them understand how they can implement AI in the best possible way, how they can get the most value out of AI and data science, how they can design a data strategy, etc. But that being said, I think that many of the tools for marketers, they are more focused and easier to use. So the idea behind these tools is that you can use them for very specific purposes, right? So you can use them for, let's say again, write titles for articles. So in this case, there's less risk. In terms of how you're going to implement this, because the use case is very, very clear.

If you have AI writing, is it really doing plagiarism? Is it going out and finding other articles and just reassembling bits and pieces? Or is it creating fresh content that you're not at risk of, for example, being sued for copyright?

We can create fresh content as well. That's the whole idea behind generative models. So I don't think there's much risk around this. I guess it's still— I wouldn't say it's early days. We've advanced a lot in the sense that we now have algorithms which write pretty realistic content. Obviously, there are many improvements to be made, but I don't expect copyright to be such a big issue. I expect realism and writing relevant content to be a bigger issue.

How do you get the AI writer to create things that have a personality? Because companies have got their own style and have got their own set of vocabulary, for example, that they use. So how can you teach— can you teach the AI tools to kind of write like you would like to write? Otherwise they're all going to kind of sound the same.

Yes, absolutely you can. I mean, that's actually not very difficult to do because in natural language generation we have these big models like GPT-3 these days, and GPT-2 is available, like open source. And you can simply fine-tune this model, as we say, on some texts. And what the model does is it learns how to speak in the language of those texts. And this means that if you— let's say, to give you a bit of a fancy example, if you want to make the network speak like Shakespeare, then you can just train it on some of the works of Shakespeare and it starts speaking in this manner. So that's an entirely solvable problem. You're not going to get an army of bots that basically are imitating each other.

Can you give us an example of a tool, a content generation platform that you could recommend that people could, if you like, train to write Like them, if you like.

I can tell you that the state of the art in natural language generation is GPT-3, which is a model developed by OpenAI. It's not open sourced, but if you type GPT-2 in Python, then you're going to find some implementations in Python. And that's fairly easy to use. This model is pre-trained, so it knows English basically, and then you can fine-tune it on texts. And I did a demonstration for a friend of of mine and I've written up a book called The Decision Maker's Handbook to Data Science. And I fed my book into this algorithm and then it started speaking in this manner. And it came up with some sentences which were pretty realistic. So the whole piece of text that it produced, it read a bit like something which, let's say, a high school student would write about computer science. It was factually correct. There were no deep insights in there. But it had really uncovered all the relationships between terms, like that AI is a subfield of computer science and databases are also something related to these topics and so on and so forth. It was pretty interesting.

Is AI writing really only in English? Because obviously the Chinese are also pursuing AI. You've got Spanish. All the Indian languages. How does that work if you're a multi— multinational company?

Well, you can have a neural network that can be trained on any language or sometimes even multiple languages simultaneously. So that's not an issue. Obviously, training a network requires significant resources. So if you want to train a network on a language that, you know, it hasn't seen before, you'll need to spend time training it, but also collecting a dataset. And this, this may be a bit challenging at times, but I think for languages spoken by many people, that's, you know, it's not very difficult to create something for those languages, not at all.

What about the different formats? For example, we talk about writing. Well, of course you've got long-form articles. What about things like headlines or even advertising copy? Because there's humor, for example, and nuance in advertising that isn't in a long-form article? Can you use the same AI engine or do you use different ones?

I mean, that's a bit of a technical matter, right? So you need to make sure that— I mean, yeah, these are really technicalities. I mean, you can use a similar technology, but then when it comes down to a use case, it needs to be fine-tuned. So the technology is there, I guess it's just not refined enough to be used, like, in general, right? You need to fine-tune it. You need to— also, these models are quite big. So they might not necessarily be easy to serve to, you know, to users. So, but these are like really technical matters which are going to be solved sooner or later. So I think it's only a few years before AI content generation becomes widely available. That's not to say that humans don't have a role, obviously, you know, because AI learns from humans. So I think I don't think that AI will be able to come up with very novel ideas, at least for now. It might in the near future. But in the very near future, I would expect that we will see some very good content generation services based on AI for content which is largely vanilla. You know, maybe you want simply to if you have 10 articles talking about one topic, you just want to produce 10 more articles, you know, of the same topic. And maybe you also want a human to edit this after an AI writes it, but it essentially becomes a very good productivity tool.

And what about things like technical writing? Because if you're writing, for example, brochures or handbooks, can AI help with that? Because that's a Big body of work as well for most companies.

That's not the best approach because when we're talking about handbooks or guides or textbooks, we're talking about facts and very systematized knowledge that has been put into boxes. And it's— and AI, I mean, it is very good in capturing statistical relationships between different things, And I believe that it's not— I mean, if you try to create a guide for that, it might come up with some funny results. That's not what you're after, right? Because when you write a guide, you want to be very precise. Whereas when we're talking about AI, it produces content which looks as if it has been created by a human, But I guess most people would agree that doesn't really understand what this content means.

What about then the intellectual property rights? If you use AI to create an article, technically you haven't written it yourself. Do you own the copyright, or can you copyright content that wasn't generated by you but just as a function of you pressing the button?

Good question. I'm not sure, that's more of a legal question, but that's a great point. I'd say why not? I mean, if you generate it via AI and then you publish a book, I mean, what does it matter?

You mentioned that AI is great at reaching out across various bodies of information and kind of compiling it and synthesizing it and generating some narrative. Where do you think this is going for Sort of relating text and audio and video. Are these 3 data formats all entirely separate as far as AI is concerned, or could you be creating an article that could then be creating, for example, a video from the same content?

Yeah, yeah, yeah, this is what many researchers are trying to do. They try to caption videos and yeah, obviously, yeah, you want to connect all these different, let's say, creative parts. Yeah, absolutely.

Have you got any examples of of smaller companies using AI, um, from your own work, um, and maybe from your own center that you could share with the listeners where you've gone, okay, we've taken a problem, used AI, and that's either, you know, replaced people that couldn't be hired for that job or was much more cost effective than it could have been done with, with people?

I think there are plenty of examples, right? Not necessarily in marketing, but like in all industries. I guess especially now with automation, it's not just about efficiency. With the pandemic still going on, it's also about using technology in, you know, in those domains where humans might not be able to sometimes work. Because of a virus, right? And we've seen some even pretty imaginative cases and things like robotic barman, for example.

What's your sort of future view, Stelios, on the impact of AI on marketing?

Well, I discussed this with speakers of the AI and Marketing Meetup, which were organized last week, and I think everyone agreed that we'll see lower barriers to entry into the game for smaller agencies and smaller companies. As AI tools for marketing become cheaper, it will be easier for a smaller company to do the job that the bigger company with 10 employees just for the marketing department is doing. And this, I think, is going to level the playing field. Obviously, bigger companies can use these tools, but I mean, you can only serve so much content, right? So the problem that smaller companies have is that probably they can't generate enough content because you have people trying to do too many things at once. But with these tools, we will see that the playing field will be leveled. And we'll see more and more and better and better tools. But I guess the first step is to create cheaper tools, because now some of the solutions out there, they're meant for big companies. And I think eventually they will get cheaper and cheaper.

Where do you think the tools are coming from, Stelios? Because The, you know, the goal of China, for example, is to be an AI leader. There's some great products like Prowly coming out of Poland. There's product coming out of Ukraine. Do you see it as being a global innovation, or do you see certain pockets being more innovative than others?

I think that we see lots of innovation taking place in many countries. The world leader is the US. China is a close second. Then outside of these 2 countries, I think the UK is very strong in AI, Israel as well, Canada. And we're really witnessing— what we're really witnessing is a global arms race around AI and data science. And what we're going to see over the next few years, I think, is multiple different pockets of innovation. Right, as more countries realize the hidden value in these technologies.

Stelios, if people want to find out about you, and obviously you have your own Tesseract Academy, tell us just a little bit about the academy that you run and how you help companies to understand and adopt AI.

So the goal of the Tesseract Academy is to help decision makers. By decision maker, I mean anyone from an executive to, to a bigger company, to an entrepreneur, or a manager, to better understand how they can use, how they can implement data science without really having to go through all the technicalities and, and the details which can sometimes seem obscure or esoteric. And through the Tesseract Academy, I've worked on topics like data strategy, on scoping out AI projects, and really worked with with some small companies to big companies like the US Navy, Vodafone, British Land. And that's a very recurring theme in my work, not only with the TestCert Academy, but also with some of the universities I work with, explaining AI to the stakeholders, helping them understand how AI and data science can be used and also related technologies like blockchain and help them achieve their goals. Because quite often people lose sight of the big picture, they make hiring decisions which are not good. They might not have the right plan in place. They might not know how they should build the right culture for data science adoption. And really, these are the things that the decision maker needs to deal with early on so they make sure that they get the most value out of their data. I think these days, as data science and AI become more and more widespread, it's also a question as to whether you can do this better than the competition because everyone will start doing it. And sooner or later, you have to You have to make the step basically and implement AI if you haven't already in your organization. So this is what the DataScient Academy is doing. This is the work that I have been doing as well over the last few years, helping decision makers get the most out of this technology in the simplest way possible, in the most efficient way possible.

Stelios Kampakos, where can people find out about you?

Yeah, so people can find out about me through my personal website, thedatascientist.com. And if they want to learn more about data strategy, then they can visit the site of Tesseract Academy. It's tesseract.academy. You can also find the website of the Tesseract Academy from my personal website, thedatascientist.com.

Stelios, thank you so much for joining me today on the Speak PR Podcast.

Yeah, thanks, Jim. It has been my pleasure. Very interesting conversation.

We've been talking with Stelios Kampakis, and I'll put those links in the show notes. So thanks so much for listening to this episode of Speak PR. We've been talking about the power of AI and technology-assisted marketing. So until we meet again, I wish you the best of health, most importantly a profitable business, and that if you're a small company, look at the, the resources that AI offers you because it really does give you a huge amount of tools really at a fraction of the cost of traditional toolsets. Thank you so much for listening.