Decentralized AI startups are raising significant amounts of cash, creating some interesting new tools, products and networks, and they’re even making AI systems more intelligent.
What’s more, all of this is being done for the greater good, to try and weaken the grip of the biggest technology companies on one of the most innovative and potentially game-changing technologies to emerge this century.
What is decentralized AI?
Decentralized AI is all about open-source models and giving users control over the data, ensuring both privacy and unrestricted access to the most groundbreaking new AI tools. It also aims to solve questions around intellectual property and the unrestricted use of copyrighted data for training AI, and even the shortage of AI hardware such as GPUs.
Similar to the concept of Web3, the idea of decentralized AI is not always so easy to define. But just like Web3, the future of decentralized AI will be built on the blockchain technology that underpins every single cryptocurrency. And it involves much more than just open-source models, which alone, aren’t enough to democratize access to AI.
The number of open-source AI models has exploded, with more than 600,000 listed on the popular Hugging Face website, but it’s notable that many of them were created by big tech companies – such as Meta Platform’s Llama 3.
The many facets of decentralized AI
Proponents of decentralized AI want to build more transparent networks based on blockchain to ensure they are more ethical, open and tamper-proof than the black box AI systems developed by the likes of OpenAI and Google.
They argue passionately that decentralized AI will provide a more efficient way for industry participants to share data and collaborate, so the latest technologies, ranging from AI models like chatbots to the underlying infrastructure, will be accessible to everyone.
Blockchain is the perfect vehicle for creating a decentralized network of GPU resources, such as The Render Network, which provides a way for people to sell the idle GPU capacity in their laptop or PC to AI developers via a peer-to-peer marketplace.
However, decentralized AI means much more than just open models and more affordable computing resources. It also means giving users more control over their personal data. An example of this is Personal AI, which has created an AI assistant that’s trained on individuals’ personal data but retains close control of that information. It will tap into more powerful models such as ChatGPT when it’s required to complete a task, but because the user’s information is secured via blockchain, those third-party models cannot access it, unless the user agrees to it.
A similar concept has been spawned by Vana, which has created a tool for Reddit users to pool their data and sell it to developers that want to use it to train AI models. It takes advantage of laws in the EU and in California, which allow Reddit users to request a copy of all data the company has about them. The idea is to collect all of this data to build up a kind of “treasury” and then collectively license the information to AI developers. Users can participate in Vana’s governance, voting on proposed deals with AI developers, with the weight of their vote determined by the amount of data they have provided to the treasury.
Decentralized AI infrastructures have also been proposed that could make such AI systems even more “intelligent” than their centralized counterparts. The AI network startup Qubic recently announced an ambitious project known as “Aigarth”, which aims to build upon the potential of artificial neural networks, which have excelled at specific tasks but have so far failed to exhibit “general intelligence” or demonstrate they can become “smarter” over time.
Qubic’s goal is to create a system that cannot only perform specific tasks, but also evolve and learn autonomously, mimicking the natural evolution of intelligence in nature. It wants to replicate the evolutionary process of intelligence within an AI network. To do this, it needs to fulfill three key steps, the first of which is to acquire more computing power via the Qubic network, and then develop simple ANNs capable of performing basic functions that serve as prerequisites for reasoning. Finally, it will then try to demonstrate how those ANNs can “self-improve” over time.
Qubic admits that the Aigarth project is extremely ambitious and that its “feasibility is uncertain”, but believes it’s necessary to try anyway. If it fulfills its vision, it could pave the way for the creation of more powerful decentralized AI networks that everyone can access.
Blockchain will be key
Most of the innovators in the decentralized AI industry are still taking baby steps and it’s clear that the entire concept remains embryonic. There’s no guarantee any of them will succeed, and it’s quite possible the movement could fade away into obscurity, similar to how decentralized social media platforms never quite made it out the door.
One challenge decentralized AI needs to overcome is the skepticism over blockchain itself, which remains closely associated with crypto and that industry’s rebellious aims to give people self-sovereignty and financial freedom. The problem is that, for every passionate crypto enthusiast, there are just as many critics who swear blind that digital money has “no value” and proudly resist it. They also tend to shun anything related to blockchain.
Then again, the reliance on blockchain can also be a good thing. Decentralized AI’s association with crypto is justified, as it shares many of the same goals as the crypto industry. That should help it attract plenty of passionate developers who are committed to upholding the ethical standards, privacy, security and transparency that’s sorely missing from the centralized AI landscape.