Home INNOVATION Emerging Ideas Generativity: Driving the Promise of Generative AI

Generativity: Driving the Promise of Generative AI

July 18, 2023

By Llewellyn D.W. Thomas and Richard Tee

Generative fit is why Stable Diffusion, DALL-E, and ChatGPT provide APIs and extensive documentation; it makes it easier for developers to integrate the underlying technology into innovative new products and services. In this article, we discuss how managers can use generativity to transform their organisations.

Generative AI is increasingly prominent in the consumerisation of artificial intelligence. Currently popularized by services such as ChatGPT¹, Stable Diffusion², DALL-E³, and others, generative AI are computer services that can use existing content like text, audio files, or images to create new plausible content. These text-to-text and text-to-image AI services are improving at a very fast rate, while newly emerging ones from Make-a-Video⁴ can generate videos from a text prompt and Phenaki5 can generate video from a still image and a prompt. Even more exciting, Nvidia’s Magix3D⁶ can be used to create 3D models from text descriptions.

Underlying the promise and excitement of the business applications of generative AI is its generativity. Paraphrasing one of the more popular definitions⁷, generativity is a technology’s capacity to enable innovation by large, varied, and uncoordinated audiences. In the context of generative AI, this means that ChatGPT, DALL-E, Stable Diffusion, and others are generative in two senses. On the one hand, they are generative in the sense that the underlying technologies can be combined with others to create new innovations (what is called combinatorial innovation). The current hype around GPT-4’s performance⁸ and its implementation in Microsoft Bing⁹ and Microsoft Office¹⁰ is one. Lensa is another good example of a third party taking the Stable Diffusion technology and creating a new service. And on the other hand, they are generative in the sense that they create new content (such as new text, images, and models) that can be used as media input for innovation. For instance, Magic3D will allow anyone to create 3D models without the need for special training, thus enabling new content for much more varied and speedier video game and virtual reality (VR) development. Generative AI has been used to create a never-ending Seinfeld¹¹ spin-off, while ChatGPT is able to provide text output at a level that approaches that of a top MBA student¹².

Generativity is not a new idea, dating back over 70 years in the social sciences. More recently, generativity has been used to understand how innovation occurs in digital contexts¹³. Perhaps the most famous generative technology is the internet, where digital technologies such as TCP/IP, HTML, and GUI-based web browsers led to an explosion of innovation and creativity, the effects of which we are still living today. Other highly generative technologies are the operating systems that enable mobile phones, which have spawned millions of apps in both the iOS and Android app store ecosystems almost entirely created by dispersed, independent developers. Indeed, much of the generativity of AI is itself enabled by the capacity of the internet and mobile phone operating systems to support the launch and use of new technologies such as generative AI.

Generativity is not a new idea, dating back over 70 years in the social sciences. More recently, generativity has been used to understand how innovation occurs in digital contexts

But how do technologies such as the internet and AI become generative? Our recent research has shown that the generative potential of a technology is directly related to the fit between the technology and the community that uses it. By fit, we mean how the characteristics of the user community relate to the capabilities of the technology. For instance, in the case of generative AI, this means the fit between the people using the technology and how the technology’s capabilities are exposed to them. If it is complex to use, then only more technically minded people will use it; if it is easy to use, then it is accessible to a much broader community.

Generative fit is why Stable Diffusion, DALL-E, and ChatGPT provide Application Programming Interfaces (APIs) and extensive documentation; it makes it easier for developers to integrate the underlying technology into innovative new products and services. We saw the same with mobile phone operating systems. In the case of the iPhone, Apple made it easy for developers from the existing Mac developer community to participate on the (then new) smartphone platform by leveraging functions, tools, methods, and identities that were attractive and familiar. In contrast, when Microsoft launched the Windows 8 mobile platform, there was a poor alignment of the technology with the existing Windows developer community, and generativity did not emerge.

Beyond (re)combining the underlying technologies into new product and service innovations, what makes generative AI special is that the new content created by generative AI supercharges its generativity. Well-designed user interfaces on consumer generative AI services make it as easy to use for the widest possible user community. This drives user adoption and the creation of a huge variety of novel content that can be recombined to create new media products. This is why Nvidia’s Magix3D is so exciting. Furthermore, this broad user adoption also drives popular interest and raises the profile of the technology, familiarising the broader community with the potential for the technology. As emphasised by our research, generativity occurs when a community is varied (as different perspectives can lead to innovative combinations), independent (as individuals need to be able to innovate in their own way), and interconnected (so that best practices and learnings can be shared).

However, it is not just the fit between technology and community that is important to enabling generativity but also the feedback from the outcomes of the generative process. This is particularly true for Stable Diffusion, where not only can the user see their image being slowly created following the text prompt, but they can also view the images created by other users at the same time and in the past. Not only do these images feedback into the technological infrastructure that enables the technology but these images also feedback to the community who are creating new images. We see even more complex feedback with mobile phone operating systems. For one, the constant flow of novel apps is cumulative, as continual reinterpretations, expansions, and refinements of the platform capabilities lead to more generativity. The experience of the community in creating the apps also feeds back, with the app developers becoming more engaged as well as more skilled in generating new apps. There is also governance feedback as tension between the mobile phone operating system platform owner and the app developers shapes the nature and design of governance.

However, as we have seen with the internet, while generativity can lead to an explosion in creativity and economic value, there is also a dark side. Just as the unrestrained innovation that typified the internet included viruses and cyberattacks, so has generative AI led to concerns of deep-fakes and concerns, both ethical and legal, as to creative ownership¹⁴. However, such concerns about generative technologies are not new or unexpected; it is the community that drives the innovations that are created by the technology and its outputs. For instance, the internet, probably one of the most generative technologies of the modern era, as well as fundamentally transforming the modern economy, has also enabled spam, copyright theft, rapidly scalable disinformation campaigns, and more.

What is key is ensuring that the innovations that are unleashed by such generative technologies are applied in the most socially beneficial way possible. So how can managers leverage generative AI, and generative technologies in general? We suggest there are six key actions managers can take (see associated sidebars).

Understand which aspect of generative AI is of interest to your organisation. Generative AI has implications for marketing¹⁵, code generation¹⁶, conversation17 and knowledge management. But are you interested in the underlying technologies, or are you interested in the new content that can be generated? The benefits of automated content generation, improved content quality, increased content variety, and personalised content18 can be delivered by both approaches. The answer to this key question has fundamental implications for how your organisation will approach generative AI. If you are interested in the underlying technologies, then you will need to work with your IT and perhaps new product development functions. In contrast, if you are interested in the new content, then it may be more straightforward to use existing services.
Since the trajectory of generative technologies is very hard to predict with any degree of certainty, managers should actively evaluate opportunities or challenges that emerge over time. It can be easy to dismiss early versions, especially when they seem rudimentary, but unpolished outputs and the underlying technology can improve much more quickly than anticipated. Witness the recent improvements to GPT-4. Appreciate the potential for disruption!
Consider which community the generative AI interacts with, and how it fits the requirements of the community. If you are seeking to leverage the technology, finding this fit is both exciting and daunting, given the potential scope of your community (which besides users may also consist of developers, advertisers, and other stakeholders) and the competing demands they might impose. If you are only seeking to use the content, understanding the community is important to understand how others could potentially (mis)use it, with the accompanying risks.
Leverage your existing infrastructure to bootstrap a community if you wish to leverage generative AI technology. For instance, witness how Mid Journey used the Discord social media platform as the only interface that allows users to connect to its technology. The Microsoft Windows Phone 8 example here is instructive; while there was already an existing community of Windows Phone developers, Microsoft was unable to leverage them to jumpstart generativity for the new operating system.
Monitor the resulting co-evolution that will occur between the innovative output, the community, and the technology, regardless of whether you are leveraging the technology or the content. Without careful governance of the output of generative technology, its successful development may be hindered. The key here is the amplification of positive feedback, such as spontaneous innovation by previously unknown users, while mitigating negative feedback, such as IP infringements and spam.
Finally, seize opportunities that emerge as the technology evolves. For instance, responding to the rapid uptake of Stable Diffusion, Apple made changes to its operating system and key machine learning components that allowed for significant performance improvements on Apple’s proprietary architecture. Microsoft has rapidly adopted GPT-4 and ChatGPT. It is not just tech giants like Apple and Microsoft that can take advantage of generative AI, app makers (such as Lensa) and other complementors can leverage the technology and introduce new features.

Generative AI is only the most recent digital technology to gain prominence due to the power of generativity. By leveraging technology’s capacity to enable innovation by large, varied, and uncoordinated audiences, managers can not only harness the creativity that comes from creating new content, but also leverage the underlying technologies. Managers that can adapt their existing organisation to reap the benefits of generativity are much better positioned to launch new products and services, as well as revolutionise marketing, R&D, and customer service in their organisation.

This article was originally published on May 19, 2023.

Go to top

About the Authors

Dr. Llewellyn Thomas is an Associate Professor at IESE Business School in Barcelona and a Visiting Professor at Imperial College Business School. He holds a PhD from Imperial College London, an MBA (with Distinction) from Cass Business School, UK, and both LLB (Hons) and BA(Hons) from the University of Sydney, Australia.

Dr. Richard Tee is a senior lecturer at Surrey Business School and research fellow at CoDE (Centre of Digital Economy) and the Surrey Institute for People-Centred AI. He has a PhD from Imperial College London and undergraduate degrees from the University of Amsterdam and the University of Maastricht (cum laude).