What is Decentralized AI?

Artificial intelligence (AI) has made remarkable strides in recent years, transforming industries and reshaping our daily lives. However, large tech companies dominate today’s AI landscape, controlling vast amounts of data, computing resources, and sophisticated AI models. These dynamics have led to concerns about privacy, security, fairness, and the concentration of power in the hands of a few entities.

Decentralized AI is a fundamentally different approach to developing, deploying, and managing AI systems that distribute key components across a network of participants rather than centralizing them under the control of a single entity.

Let’s explore the critical components of decentralized AI, how they compare to their centralized counterparts, and the benefits and challenges ahead.

Federated Learning

Centralized AI collects data from various sources—such as their own user base or public databases—and centralizes it within a single location. When training a model, the entire dataset is accessible to powerful servers or clusters that use it for training. And this creates issues with privacy and security that are starting to surface today.

Large tech companies are incentivized to vacuum up as much data as possible, which puts them at odds with privacy-conscious consumers. At the same time, companies that want to leverage centralized AI may need to send their data to these services, creating potential issues with privacy and data security. 

Decentralized AI embraces the idea of federated learning, which leaves each participant in control over their local data. Using this concept, a central server sends the current model to a subset of participants who train it on their data before sending only the model updates back to the server to improve the global model.

Collective Computing

Conventional AI companies spend vast resources on computational power to train their latest models. According to HSBC, the cost of training ChatGPT-5 could range from $1.7 billion to $2.5 billion! The race to build better models has led to a sharp increase in demand for GPUs and other hardware (and a boost to NVIDIA’s stock price!).

These high costs create a pivotal barrier to entry for governments, universities, or researchers looking to train their own competitive AI models. While Facebook provides its base Llama model for free, users must rely on the company’s initial training parameters and other settings when using the model for their own purposes.

SETI@home pioneered the idea of collective computing to search the cosmos and later by Folding@home to discover new therapeutics. Decentralized AI aims to leverage the same principles to harness the power of personal computers, smartphones, and other consumer devices to perform AI tasks that Centralized AI uses data centers for.

Collaborative Governance

Centralized AI creates a significant governance issue. For example, large tech companies controlling AI may sell access to anyone they choose to maximize profits. On the other hand, they can refuse to provide access to anyone they choose. The result is a gatekeeper that’s not accountable to anyone but their shareholders.

In addition, there is little insight into how and what data the models train on, which could lead to biases that are difficult for end users to assess. Many of these biases are already surfacing, such as the Correctional Offender Management Profiling for Alternative Sanctions (COMPAS) bias toward saying black defendants were more likely to re-offend.

Decentralized AI aims to leverage collaborative governance like Decentralized Autonomous Organizations (DAOs). These organizations use governance tokens to empower community members to make decisions about the project and its goals. The source code behind the organization is open for anyone to read or improve.

Decentralized AI Projects

Decentralized AI remains in its infancy relative to centralized AI projects like OpenAI or Llama. However, a growing number of projects are popping up that leverage federated learning, collective computing, and collaborative governance. These projects could build upon centralized AI’s foundations to provide users with alternative options.

Some of the most popular projects include:

• Gensyn connects all the world’s computers into a single network, enabling programmatic machine learning training at a low cost and on a large scale. The organization outlines its approach in its Litepaper.
  
• OORT is a decentralized, verifiable cloud computing platform that utilizes global resources, from data centers to smartphones, to enable trustworthy AI applications. Notably, the project is already live with customers.
  
• Bittensor is pioneering the decentralized production of AI by producing competitive digital commodities, such as machine intelligence, storage space, computing power, or protein folding.

Not surprisingly, many of these projects intertwine with the crypto and blockchain ecosystems. Blockchains provide a decentralized way to organize information, while cryptocurrencies make it easy to create selfish incentives to reach collective goals. It’s a perfect fit for the requirements of decentralized AIs to be successful.

Interestingly, centralized AI could also provide a boost. While companies don’t share the internal workings of their models, the models can help train other models to quickly bring them up to speed. So-called model distillation involves training smaller models to mimic larger models’ behavior to capture functionality at a fraction of the cost.

Challenges Ahead

The most significant benefit of centralized systems is their ability to move quickly. For example, a central bank can instantly print money to control inflation, whereas cryptocurrency ecosystems must rely on decentralized networks to mint tokens. The same is true with centralized AI, where companies can simply raise money and start computing.

Decentralized AI faces a few unique challenges, including:

• Economies of Scale. Federated learning and collective computing require a certain scale to be successful. After all, a limited dataset or a few smartphones aren’t enough to compete with OpenAI’s massive data centers. Attracting participants is most difficult in the early stages.
  
• Technical Challenges. Coordinating computations across numerous devices can be slower than centralized processing while handling large-scale distributed systems involves a lot of added complexity. Meanwhile, the nature of open-source development means a reliance on early volunteers.
  
• Data Quality. Many centralized AI companies spend a lot of their time improving data quality. But, when relying on diverse, uncontrolled data sources, it could be challenging to ensure data quality. Non-IID (non-independent and identically distributed) data can also affect a model’s performance.
  
• Security Concerns. Decentralization improves privacy but could introduce security issues. For instance, open-source projects involve stakeholders, and a failure to see vulnerabilities could result in the project inserting them into the codebase. It’s also much more difficult to limit access to bad actors.

Of course, these challenges only scratch the surface. There are also potential resource and infrastructure challenges, social and adoption challenges, and several possible regulatory and legal challenges. Many of these challenges, namely regulatory and infrastructure challenges, could mirror those of the crypto ecosystem.

The Bottom Line

Decentralized AI could bypass traditional gatekeepers behind today’s centralized AI and revolutionize the industry. By paving the way for more democratic and trustworthy AI ecosystems, they could address many valid concerns behind centralized AI.

If you use decentralized AI platforms that involve tokens, you may need to report any gains or losses to the IRS. ZenLedger can help you aggregate transactions across wallets and exchanges, compute your capital gain or loss, generate the paperwork you need to file each year for the IRS, and avoid costly audits!

Get started today for free!

This material has been prepared for informational purposes only and should not be interpreted as professional advice. Please seek independent legal, financial, tax, or other advice specific to your particular situation.