Decentralized AI: Investment thesis and top tokens
The key to avoid extinction and the top projects by layers
The fusion of artificial intelligence (AI) with blockchain technology to create Decentralized AI presents a groundbreaking shift. This integration promises to redefine the landscapes of data privacy, security, and the democratization of AI technologies, heralding a new age of innovation and accessibility.
By distributing AI’s computational and developmental processes across decentralized networks, DeAI not only democratizes innovation but also enhances data security and operational transparency. This paradigm shift, underpinned by cryptography, crypto economic primitives and smart contract functionalities, promises a future where AI technologies are developed in a more open, collaborative, and ethical manner. One of the most significant advantages of DeAI is its ability to transcend geographical and regulatory boundaries.
Why Decentralized AI Matters
- Data Privacy and Security: In a world increasingly concerned with data misuse, decentralized AI offers a new paradigm. By leveraging blockchain’s inherent security and encryption protocols, it ensures that individuals retain ownership and control over their data, even as it’s used to train AI models. Decentralized AI also provides checks against mass surveillance and manipulation by governments or corporations.
- Democratization of AI: Centralized AI systems often concentrate power in the hands of a few, limiting access to these advanced tools. Decentralized AI, on the other hand, makes AI more accessible to smaller entities and individuals, fostering innovation and leveling the playing field. The core of decentralized AI is moving away from centralized data storage and processing to a distributed model. This is often achieved using blockchain technology, which ensures data integrity, traceability, and security.
- Incentivization Models: Decentralized AI projects often use tokens for transactions within their ecosystems. Tokens can incentivize data sharing, reward contributions to the AI network (like training models or providing computational power), and facilitate governance mechanisms.
- Transparency and Trust: The blockchain’s ledger ensures that all transactions and data used in training AI models are immutable and traceable. This transparency builds trust among users and participants, a crucial factor in AI’s adoption and ethical deployment.
- Smart Contracts: In a decentralized AI system, smart contracts automate agreements and transactions, enabling trustless interactions between parties without the need for intermediaries. This is crucial for AI tasks such as accessing datasets, compensating data providers, and ensuring compliance with data usage agreements.
Challenges and Considerations
However, technologically speaking, training data is not necessarily faster in a decentralized architecture, the cost of moving data across long distances may offset some of the benefit of a closed-loop server farm optimized for training and operating AI models. Therefore, we have yet to see if DeAI will stand a chance against its centralized counterpart. After all, there is a cost to decentralized architecture in terms of computational efficiency.
Artificial Intelligence cannot exist without data, in fact AI is not true intelligence, it is simply statistical models that can extract valuable insights from data, and in some cases predict future outcomes through probabilistic inference. Without data, AI cannot exist, and more importantly, not all datasets are created equal. AI models can take years and several hundreds of millions of dollars of investments before producing decent results, so it is not magic.
First, let’s describe a simplified data pipeline and process in bringing AI models to commercial viability: 1) defining the goal and selecting an AI model for the task, 2) data collection, 3) data processing (labeling, cleaning, etc), 4) training the model, 5) testing & refinements, 6) optimizing the model for operational efficiency, 7) deployment and live interactions with users, 8) additional data collection with user interactions and periodic retraining after eliminating data bias, etc. Each of these steps may require different tech stacks optimized for the task at the end, running everything on a decentralized architecture is likely not going to be viable at the current time point, which means there are plenty of commercial opportunities for web3/AI entrepreneurs.
Transparency is the another factor, foundation model architectures involve millions of interconnected neurons across several layers, making it impractical to understand using traditional monitoring practices. Nobody really understands what happens inside GPT-4, and OpenAI has no incentives to be more transparent in that area. Decentralized AI networks could enable open testing benchmarks and guardrails that provide visibility into the functioning of foundation models without requiring trust in a specific provider.
Much of today’s AI exists in centralized black boxes owned by a few influential organizations. This concentration of control counters the otherwise democratizing potential of AI and hands over outsized influence on society, finance and creativity to a handful of unchecked entities.
While open source models are a great advancement, they are often built in silos rather than collaboratively. To effectively decentralize AI, open source developers need to coordinate to build machine learning models that can learn from each other over time. This collaborative approach across decentralized teams is key to creating AI systems that can rival centralized alternatives.
Promising progress
To decentralize AI, we must rethink the fundamental layers that comprise the AI stack. This includes components like computing power, data, model training, fine-tuning and inference. Merely using open source models is not enough if other parts of the stack, such as the entities providing compute for training or inference, remain centralized.
Decentralized computing can be incredibly relevant during pre-training (the stage in which a model is trained on large volumes of unlabeled and labeled datasets) and fine-tuning (the phase in which a model is “retrained” on domain-specific datasets to optimize its performance on different tasks) and may be less relevant during inference (the stage in which a model outputs predictions based on specific inputs). Foundation models notoriously require large cycles of GPU compute, which are typically executed in centralized data centers. The notion of a decentralized GPU compute network in which different parties can supply compute for the pre-training and finetuning of models could help remove the control that large cloud providers have over the creation of foundation models.
Data decentralization could play an incredibly important role during the pre-training and fine-tuning phases. Currently, there is very little transparency around the concrete composition of datasets used to pretrain and finetune foundation models. A decentralized data network could incentivize different parties to supply datasets with appropriate disclosures and track their usage in pretraining and fine-tuning foundation models.
Many phases during the lifecycle of foundation models require validations, often in the form of human intervention. Notably, techniques such as reinforcement learning with human feedback (RLHF) enable the transition from GPT-3 to ChatGPT by having humans validate the outputs of the model to provide better alignment with human interests. This level of validation is particularly relevant during the fine-tuning phases, and currently, there is very little transparency around it. A decentralized network of human and AI validators that perform specific tasks, whose results are immediately traceable, could be a significant improvement in this area.
Decentralizing the evaluation of foundation models for different tasks is an incredibly important task to increase transparency in the space. This dimension is particularly relevant during the inference phase.
Finally, the most obvious area of decentralization. Using foundation models today requires trust in infrastructures controlled by a centralized party. Providing a network in which inference workloads can be distributed across different parties is quite an interesting challenge that can bring a tremendous amount of value to the adoption of foundation models.
Now, either we sacrifice decentralization to use cutting-edge proprietary AI, or forgo access to the most powerful technology available by locking into strictly decentralized alternatives, which currently lag behind centralized models in capabilities.
To break this tradeoff, we need coordination between the disparate participants across the layers of the stack. The end goal is a collaborative substrate of artificial intelligence where decentralized infrastructure can plug in and optimally leverage AI capabilities.
We’ve made a lot of progress in the area of coding techniques that make it possible to build AI systems that don’t rely on centralized control:
- Homomorphic Encryption: This is a way of coding information so that certain calculations can be done on the coded data directly, without needing to decode it first. After the calculations, the results are still in a coded form. This means you can work with the data without ever seeing the sensitive information inside.
- GAN Cryptography: This approach was started by Google and is detailed in a study called “Learning to Protect Communications with Adversarial Neural Cryptography” from late 2016. It’s a way to keep data safe and private when sending it between different places. It uses a type of AI that learns how to encrypt data so that only the intended recipient can understand it, keeping everyone else out.
- Secured Multi-Party Computations (SMPC): This is a building block for creating new types of blockchain technology. It allows a group to calculate something together, using private information, without actually sharing that private information with each other. This method is key for making AI models that can learn from data without the data being exposed to anyone who shouldn’t see it.
Also, Explainable AI (XAI) and open-source models are emerging as promising avenues to address the black box issue in decentralized AI.
Investment thesis by layers
These are the most promising projects on different layers for AI decentralization.
Decentralized Compute and AI Inference Platforms
Overview: This category merges entities offering distributed computing resources with those specializing in decentralized AI inference and computational processes. It includes organizations at the forefront of AI training advancements as well as those tackling issues of privacy and censorship in AI model inference.
Projects: Render, BitTensor, Hyperbolic, Inference Labs, Lumino, Morpheus.
AI Data and Model Provenance
Overview: This focuses on the decentralization, ownership, governance, and transparency surrounding AI content. The category covers initiatives dedicated to ethical AI development, data sovereignty, and the tracing of content origins.
Projects: Rainfall, Numbers, Grass.
On-Chain AI Agents and Security
Overview: This area is dedicated to developing AI agents embedded within blockchain technology, which offer payment solutions and mitigate platform risks. It also includes efforts towards on-chain verification of AI models.
Projects: Fetch.ai, Modulus Labs, Delysium, Agent Protocol, Morpheus.
AI-Driven Blockchain Marketplaces and Learning Platforms
Overview: This category is about establishing decentralized marketplaces for AI algorithms and datasets, along with federated learning projects. It encourages cooperative AI development while prioritizing user privacy.
Projects: SingularityNET, ManyAI, Numerai, BitTensor.
Model Verification
Overview: These projects leverage cryptographic technologies (e.g., zero-knowledge proofs) to confirm the accuracy of claims made by models on the blockchain.
Projects: Modulus, Giza, EZKL
Top 3 tokens
BitTensor
Bittensor (TAO) is a novel protocol that combines blockchain technology with decentralized machine learning. It’s designed to democratize AI by allowing for the exchange of machine learning predictions and capabilities among network participants in a peer-to-peer manner. The platform’s goal is to enable collaboration and sharing of machine learning services and models, leveraging Proof of Intelligence as its consensus mechanism alongside a Decentralized Mixture of Experts (MoE) model among other technological innovations.
BitTensor has been used to develop and deploy Corcel, a multi-purpose AI application that has a: 1) Chat, 2) Image Generation, 3) Image Processing, 4) Image Search, 5) translation service. Each of these are built on different AI algorithms. Corcel Studio for example is a competitor to the centralized AI service MidJourney, which has made some considerable breakthroughs using a stable diffusion algorithm for generating high quality images based on textual inputs.
SingularityNET
SingularityNET is a platform aimed at creating a decentralized, open-source framework for artificial general intelligence (AGI). It leverages a network of AI agents that can outsource tasks to each other, exchange data, negotiate payments, and improve their reputation within the system. The marketplace is designed to facilitate the search, trial, and integration of a growing library of AI algorithms, allowing users to integrate these services into their applications. It features an AI Marketplace and AI Publisher for this purpose.
The AGIX token is integral to SingularityNET’s ecosystem, serving various purposes such as purchasing AI services, participating in network governance, and staking for network consensus. AGIX tokens can be used within the AI marketplace, allowing users to access a range of AI-powered services. SingularityNET has recently made significant strides, including the transition to become cross-chain compatible, which is a part of its Phase 2 plan, greatly supported by the community. This move is aimed at enhancing the platform’s interoperability and scalability by incorporating blockchain technology like Cardano, thereby broadening the reach and utility of AGIX tokens.
Fetch.ai
Fetch.ai is known for its decentralized network of autonomous agents that can perform various tasks across different industries by utilizing AI and ML. The FET token, integral to the platform, is used for transactions, accessing machine learning utilities, staking for network validation, and more.
Fetch.ai’s technology includes several innovative components like the Digital Twin Framework, Open Economic Framework, Digital Twin Metropolis, and its blockchain. These elements work together to support the creation, deployment, and training of digital twins, fostering an environment where digital entities can securely and efficiently interact and transact. The platform has targeted a wide range of use cases, from DeFi to transportation, energy, and travel, demonstrating its versatile application across sectors.
Fetch.ai’s roadmap highlights its commitment to continual improvement and expansion, with notable developments such as the Capricorn mainnet upgrade enhancing interoperability and the public beta of Resonate, a blockchain-based social platform for NFT trust-centric sharing. It operates on a Useful Proof-of-Work consensus mechanism and places a strong emphasis on creating a dynamic digital economy where data can be efficiently and securely exchanged. Autonomous Economic Agents (AEAs), crucial to Fetch.ai’s ecosystem, can represent a variety of entities and perform an array of economic activities autonomously. These AEAs, alongside the Open Economic Framework and Fetch Smart Ledger, form a sophisticated system that facilitates the exchange and monetization of data.
I will be performing a complete valuation of one of these tokens in the following articles. Which one is the most promising investment? Any other interesting projects I should be looking at? Send me your comments and don’t forget to follow!
