Increasingly depicted as the next industrial revolution of our time, AI and its multiple subsectors (NLP, machine learning, …) have occupy the center stage since ChatGPT groundbreaking launch in 2022 and sparked since then an incredible run for tech stocks at the forefront of this sector as shown by the YoY performance of the magnificent seven.
Yet, the narrative around AI should not only be merely resume to that of a quick financial return opportunity for market investors due to its far reaching consequences on our current value creation model thanks to the nature of the new underlying disrupting technology fuelling this emerging sector:
- Machine intelligence to optimise, augment and automate decision making
- Digital productivity tools at the operator level to increase the productivity of each economical agent
- Fully automated process and robotics enabling the development of scalable network of automated services both online and IRL (Tesla taxirobot, …)
- Systemic full-scale digitisation of planning and management process across all sectors not requiring a human-presence.
Nonetheless, despite all those lengthy promises of better days to come once AI will have been integrated broadly across our global economical system, limitations are already starting to emerge due to the current oligopolistic nature of the AI market. Indeed, from the increasing data and model centralisation around a few leading tech corporations, to the complete opacity surrounding data management and the impossibility to create a thriving competitive AI market due to anti-competitive behaviour from existing AI tech leading firms (GPU pre-ordering, …), concerns are rightfully starting to emerge both from private and public actors relatively to the threat posed by a centralised and corporate owned AI.
Full decentralisation: A potential remedy
While open-sourcing AI development initiatives such as Llama are welcomed initiative enabling broader technical algorithmic transparency, most of those initiatives ultimately rely on some form of centralised entity at the infrastructure and core development level due to the inherent lack of economical incentives faced by most decentralised network in the Web2 space. Consequently, more than a trendy meme, Web3 AI appears like an interesting alternative framework enabling to empower decentralised network with enough CAPEX to sustain the development of AI models thanks to purpose-built tokenomic models rewarding all active participants from the users to the ML engineering team.
Web3 AI: A new tech stack model
Similarly to the general Web2 framework, the Web3 AI tech stack is relying on a development funnel based on the three main AI technical layers that are Infrastructure, Middleware and Application. However, in line with the broader blockchain ethos, the Web3 AI tech stack is leveraging an alternative architecture composed of community-owned layers developed in parallel instead of a pyramidal framework owned by a single network agent.
Infrastructure Layer
Representing the core of the stack, this layer encompasses components such blockchain protocols or cloud infrastructures and can further be split down four sub-layers
Middleware Layer
Layer acting as an intermediary facilitating seamless communication and data management between the underlying infrastructure layer and application. It encompasses a broad array of tools such as APIs, SDK designed to streamline the development process and enhance the functionality of AI applications within the Web3 ecosystem.
Example of platforms: Fetch.AI, DAIN, Chain ML, Oraichain …
Application Layer
Representing the most-user facing aspect of the tech-stack, the application layer offers a diverse array of services and applications (AI driven marketplace, AI personal agents, …) leveraging the infrastructure provided by the underlying layers.
Example of projects: 3commas, Numerai, 0x0, Paal AI
The benefits of Web3 AI
Through its inherent structure and technological primitive extracting from the blockchain space (DLT, smart contracts, …) this emerging Web3 AI framework appears custom built to tackle the current issue faced by centralised counterparts relatively to cost, revenue distribution and transparency. Indeed, thanks to the leveraging of fully decentralised network with built-in encryption features, Web3 AI protocols are able to offer services at a far a lesser costs while achieving a fairer revenue distribution and full transparency. Additionally, through this cross-integration between AI and blockchain technology, this alternative AI framework is unlocking a number of potentially groundbreaking use-cases among which:
- Personal agents trained on user owned data
- AI agents banking enabling a massive leap in their abilities and use-cases
- Decentralised GPU marketplace unlocking a global untapped market of underutilized GPU power
- Decentralised models build on decentralised and verifiable data set
- Access to non-poised & verifiable data
- Trustless verification of AI models
- …
If so beneficial, why did we not witnessed already a ChatGPT moment in Web3 AI?
Mainly, due to its decentralised nature…
Indeed, while being ultimately its main advantage, the decentralisation aspect of the Web3 AI framework is also a definitive limitation in the short term in terms of development speed by comparison with hyper-focus centralised team. Furthermore, while having the potential to act as a key catalyst in the expansion of AI agents to a broad range of sectors by enabling them to perform value transfer autonomously, the Web3 AI tech stack also implies a broad acceptance of crypto payment throughout the web which for now is not the case.
Conclusion
While maybe the only path towards the development of a truly decentralisation AI sector in the long term, the Web3 AI framework does not currently appears ready to replace the current centralised AI model and will need to overcome a few technical and regulatory hurdles before being able to do so in the future.
