2023 is the year of artificial intelligence. The surge in interest in AI technology this year has led to a $4 trillion rally in tech stocks and new applications for AI in major industries cropping up around the globe. Since 2013 (a full decade before the current buzz around AI) our vision has remained consistent: to help publishers leverage artificial intelligence and automation to make significant productivity gains and increase performance. 

Having innovated and operated at the intersection of tech and publishing for years, our experts are well-placed to explain what generative AI really means, share their vision of the future of the publishing industry, and outline the risks and opportunities that AI presents.

In part 1 of our great Q&A on generative AI, we sat down with Simon Tanné, our Head of Science since 2014 and Dr. Marc Fletcher, CTO since 2013. 

First things first: We hear many different terms associated with AI. Can you give a brief overview of what is meant by “automation” and “generative AI”? Are they different?

Picture of Simon Tanne, Head of Data Science at Echobox
Simon Tanné, Head of Data Science

Simon (Head of Data Science): Let’s start with generative AI. In simple terms, generative AI is a type of artificial intelligence that can create new content. The AI uses patterns and structures it has learned from the data it has been trained on to produce new and unique outputs. These outputs can be anything from text to images or music, and more.

A prime example of generative AI for text is GPT4, which most of us are familiar with as it powers ChatGPT. The model has been trained on a vast amount of textual data, such as online articles, and learned the patterns and structures of language. When prompted with clear instructions, the AI can generate text that is articulate, relevant and applicable to the topic at hand. 

Marc (CTO): Exactly. To put it a bit differently, generative AIs are statistical models that use the history of what they have seen previously to respond to inputs known as “prompts”. Generative AI models now have vastly increased degrees of freedom over previous generations of AI. That’s precisely why we’ve come to know them as “generative”. 

However impressive the outputs can be, for seemingly tiny amounts of input, there’s still no thinking involved; it’s just all statistics. You can see this effect demonstrated quite humorously if you ask a generative AI to multiply two large numbers together.

Simon: Automation, on the other hand, is technology that performs tasks without human intervention. By leveraging generative AI alongside other automation capabilities, Echobox can fully automate the process of sharing on social media, freeing up time for other important tasks and improving the effectiveness of social media campaigns.

Businesses of all types and sizes are recognizing that AI can offer a competitive advantage and a true edge to their offering. – Simon Tanné, Head of Data Science

Marc: Automation existed long before generative AI, but the two can be combined. Researchers and companies like Echobox have a new tool to help save others significant amounts of time in everyday tasks by automating certain activities and facilitating workflows.

How has generative AI evolved over the past few years? How do you explain the recent increase in investments and publicity surrounding it?

Photo of Marc Fletcher, CTO of Echobox
Marc Fletcher, CTO

Marc: Generative AI has evolved along a relatively predictable path as models became larger and the process required to convert text, images and sounds into trainable data became more advanced. Before the GPT3 and GPT3.5 models were popularized by ChatGPT, GPT2 (released in 2019) was already capable of generating sensible text responses most of the time. The tipping point for generative AI was reached with the release of GPT3.5 in March 2022. The primary factor in its success and adoption was that grammatical errors were mostly eliminated in its responses, which appealed to businesses. 

Simon: Agreed. First and foremost, like Marc mentioned, generative AI is more effective. It has evolved from producing basic outputs to creating realistic, coherent text due to advancements in machine learning techniques, increased computational power and larger data sets. 

Secondly, many have understood the wide-ranging potential applications across various sectors. From creative industries to healthcare, businesses of all types and sizes are recognizing that AI can offer a competitive advantage and a true edge to their offering. For example, you can now generate creative ideas for advertising copy in seconds, where it would have taken a few hours before. 

Finally, these advancements in the technology and its adoption by businesses have rendered AI technology more accessible and democratized.

Marc: There still remains much to be desired as even the most recent incarnations of generative AIs are known to “hallucinate”. Hallucinations refer to cases in which the AI confidently presents an answer, when it is in fact false. Yet I can’t help but feel like we’ve reached a milestone in the history of AI this year. Following the success and media exposure received by the GPT3.5 model, there has been a sea change in the public perception and a new rush in AI investments. I hope this will uncover ever more innovative ways to use AI technology.

It’s crucial that we invest and promote digital literacy. – Simon Tanné, Head of Data Science

Developments in generative AI seem to have accelerated rapidly in the last year, with GPT3, ChatGPT and GPT4 all being released, as well as DALL-E 2 being made available for public use. 

What limitations or risks do you see with this rapid evolution of generative AI? 

Simon: AI models can sometimes produce unexpected or inappropriate outputs. This poses some problems when it comes to reliability and control. It may also be the case that it’s hard to understand why AI made a specific decision, raising transparency and accountability issues. This can make it challenging to come up with an optimal prompt for a given generative task, as it’s unclear which part of the prompt rendered the model’s output less accurate or coherent. 

There are ethical and social issues, as well. Certain individuals or businesses may misuse realistic AI-generated content — deep fakes and AI-generated fake news are some common examples. It’s also important to consider user dependency on generative AI. This phenomenon could lead to diminished human skills and blind trust in AI-generated content. It’s crucial that we invest and promote digital literacy. 

Finally, from a business perspective, generative AI is extremely resource intensive. Training these models requires significant computational resources, which raises significant barriers for smaller entities who want to venture into this field. These technologies consume a lot of energy, and also raise environmental concerns. 

Marc: Completely agree. I think we’ve now reached a plateau of generative AI models in terms of output quality. The innovation that follows will be focused on use cases and optimisation. For example, we might see the same generative models being more easily run on consumer hardware rather than requiring large cloud infrastructure. Looking beyond this, while generative AI models remain statistical in nature, they will likely never be able to overcome the challenges they currently face in terms of accuracy and bias. Solving these challenges will most probably require a completely different technological foundation which doesn’t yet exist.

There has been a sea change in the public perception and a new rush in AI investments. I hope this will uncover ever more innovative ways to use AI technology. – Marc Fletcher, CTO

This Q&A is not over! Keep your eyes peeled for part 2, where we’ll delve further into publishing, Echobox and recommendations for publishers wishing to use artificial intelligence. 

In the meantime, if you wish to learn more about Echobox’s artificial intelligence, you can sign up to our newsletter or follow us on LinkedIn, Facebook, and Twitter. We’ve also got great white papers on our latest research on AI for publishers.