A Framework for Considering the Effects of New Technologies

14th of May, 2023

Since I first discovered how useful large language models, like ChatGPT, are when I stumbled upon OpenAI's GPT-3 Playground at the beginning of my final year of university, I have been adding ideas for applications of this technology to a growing note on my phone called ‘Today I saw the Future’.

After a short time, I realised it was quite easy to think of applications for this technology that followed the structure: idea = existing product or service + AI, which can be generalised to idea = existing product or service + new technology but difficult to think of ideas that are the ‘next step further’. The problem is that ideas that follow this structure are a dime a dozen.¹ They are useful, the world will be better off with them existing, and many people will turn them into compelling use cases that find product/market fit either through new products or as features of existing products, but they are not where a large share of the value (using the term broadly) of a new technology lies.

When going through the list of ideas, I kept thinking that many of them were “first order effects” of a new technology and that it would be useful to have a framework for thinking about the innovations that would arise from new technologies that went beyond the immediate consequences.

First Order Effects

The direct and immediate consequences of a new technology, which typically involve the integration of the technology into existing products or services, improvements in functionality, and the creation of value for users, companies, and society as a whole.

The majority of early applications of a new technology are these first order effects. They are clear applications of the technology and usually don't require societal or cultural changes to be adopted. Some examples of early (and quite compelling) applications of LLMs include:

Writing assistants and extensions
Document and meeting summarisers
Image and video generators
Transcription, speech, and translation

Incumbents may be better positioned to win market share of the first order effects because they own the dominant products and services that the new technology would be integrated into which means a) they have an existing userbase to test, iterate and deploy what in many cases is a new feature, and b) are not having to build the entire product or service from scratch in addition to the integration of the new technology.

The tables of generative AI start ups in the early days of ChatGPT and GPT-4 show the large number of companies working on products that fit the formula idea = existing product or service + AI and how quickly this landscape becomes both competitive and saturated. Many of these early, first order copmanies may find that they are unable to compete with incumbents with industry industry leading products and services who then move fast to integrate generative AI.

GIF of the number of generative AI companies increasing rapidly over time

Second Order Effects

The broader, indirect consequences that emerge as a new technology becomes widespread and integrated into various aspects of life and serves as a platform for other applications. Second order effects involve changes in systems, industries, or market structures, and can lead to the creation of new opportunities, applications, and challenges that were not immediately anticipated.

To show this through an example, let's look at the first and second order effects of how generative AI could impact the media industry.

First Order Effects: The tools to generate image, video, text, and audio, coupled with the ability to summarise, manipulate and edit these media types at scale are built into content creation software.

Second Order Effects: This prerequisite will lead to the creation of new forms of media, new ways of consuming media, and new ways of interacting with media. These could be:

Dynamic Storytelling: Generative AI could revolutionize media by enabling tailored and dynamic storytelling experiences. The same story could be experienced in different formats (short or long movies, episodic series) and from different perspectives. Creators could create a seed story and then use generative AI to grow and adapt the story (as demoed in Unreal Engine 5.2)
Empowering Creators: The requirements for creating high-quality content could be lowered, allowing more people with fewer resources but high levels of creativity to create content and distribute content (a common thread through nearly all computing innovations, à la Moore's Law). Blockbusters and AAA games will still exist, but the long tail of content will be far more diverse and interesting. Reductions in the cost of creating content will also make it much easier to green-light new projects.
Interactive and Personalized Experiences: AI could make media consumption more interactive and personalized, such as allowing viewers to interact with different characters or choose their preferred actors.
Multidimensional Content: Future content could be more multidimensional, with the same story being consumed in different formats and perspectives. It would be far easier to turn a game into a movie, or a movie into a book, especially for smaller teams of creators.
Business Model Shift: The adoption of generative AI would favour an AI-equipped platform-centric model over publisher-centric models, a shift that has already been happening as platforms would own the user data (and possibly the generative models) which would be used to create and tailor content.

Third Order Effects

The long-term, transformative changes to society, culture, and human behaviour that result from the sustained influence and advancements of a new technology. These effects encompass shifts in values, norms, and expectations, as well as the emergence of new capabilities, and applications that are difficult to predict and may have far-reaching implications for the future.

This definition is comprised of two parts:

An extension of the second order definition (Given the technology has affected society, how does this transformation change us?)
The technology becoming drastically more advanced and mature.

Under this definition, the third order effects are largely influenced by how the technology affects broader society (the second order effects) which makes them hard to predict. This is especially true when the new technology first takes off and the first order effects are playing out.

Given these classifications, where should founders, investors, and incumbents focus their attention? Building products that capitalise on the first order effects of new technologies could suit teams especially skilled at execution and with an ability to move quickly. However, those looking to make a significant impact on the world and be poised to create and capture large amounts of value should focus on the second order effects of new technologies. This will likely require greater levels of creativity, innovation, and strategic foresight, as the outcomes are not as immediately obvious and may require more significant change or risk. As for third order effects, given that in the early stages of a new technological paradigm, the third order effects are either a) not yet apparent or b) too far in the future that the timing may not be right, they are best left as moonshot bets or research projects while the field is still in its infancy.

¹It is especially true that execution matters far more than the idea for applications that are first order effects. ↩