In a recent article, I explored potential opportunities at the intersection of ChatGPT and Web3 technologies. The hype behind generative artificial intelligence (AI) and technologies such as ChatGPT and GPT-4 is warranted, and Web3 hasn’t been exempt from it. In recent weeks, we have seen AI-related crypto tokens rally to historical highs, and even new venture funds created to invest at the intersection of generative AI and Web3.
While the possibilities of combining ChatGPT type technologies with Web3 infrastructure can make our minds fly, the Web3 community should face the reality that the majority of the value in generative AI is being captured in traditional Web2 infrastructures. Extrapolating that idea a bit further takes us down the path of a controversial theory, which is nonetheless worth exploring: that ChatGPT momentum can have a negative and lasting impact on Web3.
Jesus Rodriguez, a speaker at CoinDesk's Consensus festival in April, is the CEO of IntoTheBlock.
The core idea behind the potential negative impact of generative AI in the Web3 space is relatively simple. Generative AI has the potential to change every aspect of how software and content are developed and consumed, from the infrastructure to the application layer. These days we are seeing every major technology and content provider incorporating generative AI into their platforms. If the core of that revolution is taking place outside Web3, it is likely to have an impact on the innovation, talent and funding gap between Web2 and Web3 technologies. Furthermore, if not addressed quickly, this gap is likely to continue expanding at a multi-exponential growth rate. The solutions to this problem are certainly far from trivial, but there are some first-principles ideas that can be explored to start addressing that gap.
That the generative AI movement is taking place in Web2 shouldn’t come as a surprise if we factor in that for a decade Web3 hasn’t created any meaningful infrastructure or technologies to support machine learning (ML). Web3 stacks evolved around foundational components such as decentralized computation, storage, identity and messaging, but there has been little attention paid to the ML space. Not surprisingly, all ML breakthroughs such as the transformer architectures and pretrained models had no footprint in blockchains or Web3 infrastructures. When the release of models such as ChatGPT, GPT-4 or stable diffusion signaled that generative AI could hit escape velocity, the Web3 movement found itself having no relevant foundation to support the new generative AI revolution. This problem is even worse when we evaluate the rate of progress of generative AI technologies.
Multi-exponential growth and massive technical gap
Gaps in generative AI capabilities between the Web3 and Web2 world are widening fast. Trends such as cloud or mobile computing evolve at a linear or polynomial rate, in which a new release improves upon previous with new features and capabilities. Generative AI grows at a multi-exponential rate.
Models such as ChatGPT or GPT-4 use a baseline of data and infrastructure, which represents a high bar for startups trying to recreate those capabilities. Additionally, those models become exponentially better as more people use them, and they collect more data that can be used to pretrain future versions. At this point, the gap can become so big that it is unbridgeable.
Read more: Jesus Rodriguez - A Pragmatic View of ChatGPT in a Web3 World
At the moment, Web3 infrastructures do not possess the compute, data or data science framework foundations to embrace generative AI. Decentralized applications (dapp) can certainly incorporate generative AI capabilities by interacting with models via Web2 APIs, but the idea of Web3 native generative AI seems a bit challenging at the moment. As generative AI continues to evolve rapidly, the challenges for Web3 can become apparent across different dimensions.
Let’s take a look at different levels of the stack.
Cloud platforms such as AWS, Azure and Google Cloud are rapidly incorporating generative AI capabilities in areas such as natural language, images, video and several others. The computation and data requirements of generative AI models currently seem beyond the capabilities of Web3 infrastructures. Consequently, the new generation of generative AI applications will fundamentally be powered by Web2 cloud platforms with very little footprint in Web3 infrastructures. If generative AI delivers on its promise, this means that Web3 platforms can fall incredibly behind in terms of adoption.
As Web2 platforms incorporate generative AI capabilities, this will power a new generation of applications that will incorporate generative AI as a first-class citizen. These new-generation applications will take place disproportionately in Web2 as Web3 stacks are not equipped to power generative AI capabilities. Sure, we will see dapps incorporating features powered by models such as ChatGPT, but obviously, those features will be completely off-chain.
For years, crypto and Web3 technologies were regarded as the next major trend to modernize fintech. Undoubtedly that focus has shifted towards generative AI. Most fintech platforms are more concerned about not being disrupted by leaner alternatives powered by models like ChatGPT than to build digital currency rails.
The level of innovation around generative AI technologies and the popularity of technologies such as ChatGPT is certainly contagious and is attracting developers looking to build next-generation applications. The explosion in generative AI technologies coincides with a brutal downturn in the crypto space. With the combination of these two events, the Web3 space might be at risk of experiencing a developer talent drain going into the generative AI space.
Venture capital investments are another area that is likely to shift from Web3 to generative AI. The 2021 bull run brought record levels of VC investments into Web3 companies, and movements such as decentralized finance (DeFi) and non-fungible tokens (NFT) finally showed practical applications about the promise of Web3. The downturn of 2022 combined with the explosion in the generative AI space has shifted the flow of VC funds into the generative AI space, which also contributes to attracting the top talent in the tech industry.
The lack of strong machine learning foundations has precluded Web3 from participating in the first wave of generative AI innovation, but this could still be addressed. Given the current state of technology and challenges, there are two clear areas in which generative AI can really benefit from the native capabilities of Web3 architectures.
- Decentralized generative AI: There are enough concerns about knowledge centralization and control over large generative AI models that it creates an opening for decentralized alternatives. Even though the decentralized AI trend never reached meaningful adoption, generative AI is resurfacing the dialogue around the value proposition of decentralization to mitigate control, bias, fairness and other desired features of these models.
- Proof of knowledge: Some of the biggest pushbacks against the adoption of generative AI come from the potential for generating toxic, racist, biased content, as well as their propensity to hallucinate or "make stuff up." From that perspective, enforcing verifiable traceability mechanisms across the pretraining, fine-tuning, and use of generative AI models such as ChatGPT is a super important capability for its adoption in mission-critical scenarios. This is one of the scenarios in which blockchain runtimes are incredibly well situated to inject accountability into generative AI models.
These scenarios combine the strengths of Web3 and generative AI platforms. While Web3 was not well situated to embrace the first wave of the generative AI revolution, it can still contribute meaningfully to its future. Releases such as ChatGPT should definitely be a wakeup call for the Web3 community that decentralization is not enough, and we need to build the technological foundations to embrace future waves of innovation.