Decentralized Artificial Intelligence (DAI)

What is Decentralized AI (DAI)?

A decentralized artificial intelligence (DAI) system is a type of artificial intelligence (AI) solution that uses blockchain technology to distribute, process, and store data across a network of nodes.

Decentralized AI systems allow users to make use of pre-trained AI models on their local devices so they can benefit from AI-generated insights without handing over their data to a centralized authority.

As part of a decentralized approach, the user can use a prebuilt AI model to process data that are stored on their device and send the results to a third party without sharing their underlying personal data.

What are the Core Components of Decentralized AI?

As AI researcher Professor Longbing Cao explained:

“DeAI refers to the AI thinking, methodologies, technologies, systems, and services for developing, managing, and deploying decentralized intelligence in decentralized settings.”

This includes “storing, updating, sharing, and exchanging decentralized intelligence between decentralized agents, nodes, or devices; and integrating decentralized intelligence from local agents and across decentralized ecosystems (with their services, and environments) for higher level intelligence and intelligent problem-solving.”

At a high level, decentralized AI systems are built with a number of core components, including AI platforms or decentralized apps (dApps), blockchain-distributed ledgers, smart contracts, federated learning, and homomorphic encryption technologies.

Blockchain’s distributed ledgers enable AI developers to distribute pre-built machine learning (ML) models to user’s devices. Then, these devices can act as autonomous agents carrying out AI inference tasks locally, either as an independent entity or as part of a coordinated connected network.

Likewise, federated learning and homomorphic encryption help to isolate and maintain the privacy of data processing activities on the user’s device so that their information can’t be viewed by unauthorized third parties.

The end result is a decentralized AI system that enables:

• Users to provide their data to AI training models without disclosing it to a third party;
• Processing and decision-making that operates independently of a centralized authority;
• Developers to distribute pre-built training models across a network of nodes;
• More transparency over an AI model’s processing activity.

What is the Point of Decentralized AI?

Traditionally, the development of AI systems has remained siloed among a handful of technology vendors like Google and OpenAI, who have had the financial resources necessary to develop the infrastructure and resources necessary to build and process large datasets.

However, the centralization of AI development in the industry has meant that organizations need to have significant funding to be able to develop and process the data necessary to compete in the market.

Likewise, it’s also incentivized vendors to pursue a black box approach to AI development, giving users and regulators little to no transparency over how an organization’s AI models operate and make decisions. This makes it difficult to identify inaccuracies, bias, prejudice, and misinformation.

Decentralized AI applications address these shortcomings by providing a solution to move AI development away from centralized providers and toward smaller researchers who innovate as part of an open-source community.

At the same time, users can unlock the benefits of AI-driven decision-making locally without needing to share their personal data with third parties.

Federated Learning vs. Decentralised AI

Federated learning is the name given to an approach where two or more AI models are trained on different computers, using a decentralized dataset. Under a federated learning methodology, machine-learning models are trained on data stored within a user device without that data being shared with the upstream provider.

While this sounds similar to decentralized AI, there is a key difference. Under federated learning, an organization has centralized control over the AI model used to process the datasets, while under a decentralized AI system, there is no central entity in charge of processing the data.

Thus federated learning is typically used by organizations looking to build a centralized AI model that makes decisions based on data that has been processed on a decentralized basis (usually to maintain user privacy), whereas decentralized AI solutions have no central authority in charge of the underlying model that processes the data.

As Patricia Thaine, co-founder and CEO of Private AI, explained to Techopedia, “Federated learning tends to have a centralized model that gets updated based on the learnings of distributed models. A decentralized system would have multiple nodes that come to a consensus, with no central model as an authority.”

“I’d think the two might get confused because they are both distributed forms of AI. In general, depending on the problem, these can be very difficult to get right, with mainly large companies like FAANG, Microsoft, or similar having the expertise and resources to do the necessary R&D for reliable at-scale deployment.”

Benefits of Decentralized AI

Using a decentralized AI architecture offers some key benefits to both AI developers and users alike. Some of these are:

• Users can benefit from AI-based decision-making without sharing their data;
• More transparency and accountability over how AI-based decisions are made;
• Independent researchers have more opportunities to contribute to AI development;
• Blockchain technology provides new opportunities for encryption;
• Decentralization unlocks new opportunities for integrations with Web3 and the metaverse.

Democratizing AI Development

While decentralized AI is still in its infancy, it has the potential to democratize AI development, providing more opportunities for open-source model developers to interact with users independent of a centralized authority.

If enough vendors support decentralized AI models, this could significantly reduce the amount of control that proprietary model developers have in the market and increase transparency over AI development as a whole.