Is Decentralised AI an Achievable Reality or an Insurmountable Challenge?

Note: Views presented in this article represents the perspective and opinions of the author and do not necessarily represent Coinlive or of its official policies.

While some extol the virtues of decentralised artificial intelligence (AI), touting its potential to reshape the future of technology, critics warn of its inherent risks.

They argue that the lack of centralised control leaves decentralised AI vulnerable to manipulation and exploitation.

Knowing How Centralised AI and Decentralised AI Work

Centralised AI and decentralised AI refer to two different approaches in the realm of AI based on how data processing and decision-making are organised.

In centralised AI systems, all data processing and decision-making occur in one central location or server, like a powerful computer server owned by a company or organisation.

This central server collects data from various sources and analyses it to make decisions or provide insights.

It is like having one big brain making all the decisions.

As for decentralised AI systems, data processing and decision-making are distributed across multiple devices or nodes rather than being controlled by one central authority.

Each device or node in the network has its own computing power and contributes to the overall processing and decision-making process.

It is like having many smaller brains working together to make decisions.

Source: Coinmotion on decentralisation and centralisation

Centralised AI Offers Efficiency and Ease of Management but Risk Exploitation

A pivotal debate in this domain centres on the advantages and drawbacks of centralised versus decentralised control over AI systems, especially concerning the protection of public interests on a large scale.

Centralised control of AI infrastructure offers advantages in terms of advanced utilisation and efficiency.

In this model, a single entity, often a government or large corporation, holds sway over the development, deployment, and operation of AI systems.

This centralised approach facilitates streamlined decision-making processes, swift adoption of new technologies, and the aggregation of vast data sets for analysis and optimisation.

By tapping into centralised AI systems, brands can gain insights into consumer behavior, tailor marketing strategies with precision, and automate various operational aspects to enhance efficiency and competitiveness.

However, the concentration of power inherent in centralised AI control raises significant concerns about potential misuse and exploitation.

Centralised AI systems wield immense influence over individuals' lives, potentially shaping their choices and behaviours in ways that may not align with their best interests.

Moreover, the lack of transparency and accountability in centralised control models can exacerbate issues of bias, discrimination, and privacy violations, further eroding citizen rights and autonomy.

Below are the notable concerns:

Data Privacy: Centralised AI systems often amass vast troves of personal data, posing significant privacy risks. When a single entity controls this data, individuals' personal information becomes susceptible to misuse or exploitation.

Control: Concentrated power can foster opacity and lack of accountability in AI decision-making. If a small group or organisation dictates AI algorithms, it may manipulate outcomes to its advantage, potentially infringing on the rights and freedoms of others.

Bias and Discrimination: Centralised AI systems may inadvertently perpetuate or intentionally exacerbate bias and discrimination, especially if the training data is skewed or biased. This can have detrimental effects on marginalised communities.

Security Risks: A centralised AI system becomes a prime target for cyberattacks. Breaches can result in extensive damage, including data breaches, manipulation of AI algorithms, and other malicious activities.

Innovation Suppression: A centralised AI landscape may impede innovation as smaller players and startups encounter barriers to entry. Innovation is crucial for addressing ethical and safety concerns in AI.

Centralised AI Dominates Current Landscape

In terms of market share, centralised AI dominates the current landscape, with established tech giants and organisations deploying centralised AI solutions extensively.

These systems have a significant user base, ranging from enterprises to individual consumers, and have demonstrated their effectiveness in various applications, from recommendation engines to natural language processing.

Technologically, centralised AI systems have reached a high level of maturity, benefiting from years of research and development.

They often boast sophisticated algorithms, vast datasets, and robust infrastructure, enabling them to deliver accurate predictions and insights consistently.

For example, banks and financial institutions use centralised AI systems to analyse vast amounts of data to assess credit risks, investment opportunities, and market trends.

Credit card companies employ centralised AI algorithms to detect fraudulent transactions and identify patterns indicative of fraudulent activities.

When it comes to the art and music scene, some art studios and digital artists use centralised AI algorithms to generate artwork automatically based on predefined styles, themes, or input images.

While streaming platforms like Spotify employ centralised AI algorithms to analyse user preferences and behaviour to recommend personalised playlists and music recommendations.

Ben Ratliff, a music critic and author of "Every Song Ever: Twenty Ways to Listen in an Age of Musical Plenty", added:

“Spotify is good at catching onto popular sensibilities and creating a soundtrack for them.”

However, despite their maturity, centralised AI systems face challenges related to data privacy, scalability, and single points of failure, which can limit their effectiveness in certain scenarios.

Decentralised AI Advocates for Autonomy and Privacy

Decentralised AI offers a contrasting approach, prioritising individual autonomy and collective empowerment.

In this model, AI systems are distributed across a network of nodes, with no single entity exercising overarching control.

This distributed architecture mitigates the risks associated with centralised control, as there is no single point of failure or manipulation.

Decentralised AI systems operate on principles of transparency, accountability, and democratic governance, ensuring inclusive decision-making processes that represent diverse interests and perspectives.

By dispersing control among multiple actors, decentralised AI limits the potential for abuse and oppression, safeguarding citizen rights and promoting equitable access to AI technologies.

It plays a vital role in safeguarding against dystopian AI scenarios for several reasons:

Data Ownership and Privacy: In a decentralised system, individuals retain greater control over their data, deciding how, when, and with whom it is shared. This reduces the risk of data breaches and privacy violations.

Transparency and Accountability: Decentralisation promotes transparency by involving multiple stakeholders in AI development and deployment. This makes it harder to manipulate AI systems for unethical purposes and increases accountability.

Reducing Bias: With diverse stakeholders involved in development, decentralised AI can be more inclusive and less biased. This helps address fairness and discrimination issues.

Improved Security: Decentralised systems are more resilient to cyberattacks due to distributed data and decision-making. This makes it harder for adversaries to compromise the entire system.

Fostering Innovation: Decentralisation encourages innovation by allowing a broader range of participants to contribute to AI research and development. Startups, individuals, and smaller entities can play a significant role, making the field more dynamic and responsive to societal needs.

For instance, decentralised finance (DeFi) platforms leverage decentralised AI algorithms to automate lending, borrowing, and trading operations without intermediaries, offering users more control over their finances.

As for the art and music scene, decentralised art marketplaces like OpenSea allow artists to sell their digital artworks directly to buyers without intermediaries, utilising decentralised AI algorithms for authentication and provenance verification.

Some blockchain-based platforms also use decentralised AI algorithms to track and distribute royalties to musicians, ensuring transparent and fair compensation for their work.

Another use case is federated learning which represents an approach where model training occurs directly on decentralised devices, preserving user data on individual devices rather than consolidating it centrally.

A prime example of this methodology is Google's Federated Learning of Cohorts (FLoC).

Allison Duettmann, CEO of the Foresight Institute and a featured speaker at Consensus 2024, highlights three pivotal areas where blockchain can revolutionise AI systems.

When asked what she sees as the most promising, legitimate ways that blockchain can help with the development of safer or better AI, she explained:

“I guess I would say there are three buckets. One would be the security side of things, where the Web3/crypto space has a lot of experience with failing fast if you don't have secure systems built. Because many of the systems in the permissionless or immutable spaces — if you don't build them correctly — there may be a million-dollar bounty attached to them. So I think this spirit of building safety and having an almost unintentional vetting process is an important mindset, as we build up AI systems. The second one is on the side of, how can we leverage technologies that the crypto space is also working on, to build AI systems that we would want from a human perspective? [The third bucket] The governance side of AI. So, you can possibly use cryptographic technologies to prove that you’re following certain safety benchmarks or standards. Rather than having to share all of the information that is possibly proprietary, you can only share information that is needed to prove you’re hitting certain safety benchmarks."

Arif Khan, CEO of Alethea and also a featured speaker at Consensus 2024's AI Stage, asserts that decentralised AI transcends mere hype and abstract concepts; it represents a pivotal shift toward the future.

He emphasizes its tangible benefits, suggesting that it has the potential to significantly enhance our daily lives.

Decentralised AI Adoption Growing but Lacks Maturity and Faces Multiple Hurdles

Adoption of decentralised AI solutions is growing steadily but faces hurdles related to interoperability, scalability, and regulatory uncertainty, as well as fewer established players and a smaller user base.

Without centralised coordination, achieving consensus and implementing collective decisions can be complex and time-consuming, hindering efficiency.

Moreover, the lack of centralised authority may lead to fragmentation and divergent goals among nodes, potentially impeding cohesive AI strategies and initiatives.

Interoperability in particular, stands as a significant hurdle in the path of decentralised AI, given its reliance on diverse platforms and technologies.

Without seamless compatibility, the expansive potential of AI remaining unexplored.

Moreover, the uncertain regulatory environment poses challenges, as governments globally strive to adapt to the swift evolution of AI technology.

This regulatory lag could result in disjointed frameworks or, even worse, a lack of effective oversight.

Additionally, security emerges as a critical concern, considering the distributed nature of decentralised systems.

While this architecture offers resilience, it also exposes them to potential cyber threats, underscoring the urgent need to safeguard their integrity.

Jesus Rodriguez, CEO of IntoTheBlock, contends that AI tends to naturally centralise over time, posing a challenge to decentralisation initiatives.

One major challenge of decentralised AI is maintaining synchronisation among decentralised nodes in the AI network.

In a decentralised system, multiple nodes independently process and analyse data, making it crucial to synchronise their actions to ensure consistency and accuracy.

However, coordinating the activities of disparate nodes in real-time can be complex, especially when dealing with large volumes of data and dynamic environments.

Another challenge is ensuring data consistency across decentralised nodes.

In a distributed AI system, each node may have its own dataset, leading to discrepancies and inconsistencies in the data used for training and inference.

This can result in biased or inaccurate AI models, compromising the reliability and effectiveness of AI applications.

Potential Solutions to Address These Challenges Have Limitations

One approach is to implement consensus mechanisms that enable decentralised nodes to agree on the state of the network and coordinate their actions effectively.

Consensus algorithms such as proof-of-work and proof-of-stake are commonly used to achieve synchronisation and data consistency in decentralised systems.

Additionally, techniques like federated learning and blockchain-based data sharing can help improve data consistency by allowing decentralised nodes to collaboratively train AI models on distributed datasets while preserving data privacy and security.

However, these solutions have limitations that impose constraints on decentralised AI applications.

Consensus mechanisms can introduce latency and overhead, slowing down the overall performance of AI systems.

Federated learning techniques may suffer from communication bottlenecks and scalability issues, particularly in large-scale distributed environments.

Moreover, blockchain-based data sharing can be resource-intensive and may not scale well with increasing network size and complexity.

Path to Decentralised AI Possible but an Uphill Battle

Despite the growing popularity of decentralised systems like blockchain, there remains a significant gap in understanding among the general public.

Many individuals are unfamiliar with the fundamental concepts of decentralisation and how it differs from traditional centralised models.

This lack of awareness often leads to scepticism and mistrust, hindering the widespread adoption of decentralised technologies.

Currently, the level of public acceptance varies widely depending on factors such as geographic location, socioeconomic status, and familiarity with technology.

In regions where internet access and digital literacy are high, there tends to be greater acceptance of decentralised technology.

However, in areas with limited access to technology or where centralised systems dominate, there may be more resistance to embracing decentralised solutions.

Thus, enhancing transparency and participation is key to building trust in decentralised technology.

By providing clear and accessible information about how decentralised systems work and their potential benefits, developers and advocates can demystify the technology and address misconceptions.

This includes educating the public about concepts like blockchain consensus mechanisms, data encryption, and decentralised governance structures.

Moreover, fostering greater participation in decentralised networks can help empower users and demonstrate the democratic nature of these systems.

By involving stakeholders in decision-making processes and giving them a voice in the direction of development, decentralised projects can instill confidence and build stronger community support.

Open-source development, community forums, and decentralised governance mechanisms all contribute to this goal.

In theory, decentralised AI may seem pretty straightforward but in practice, it presents significant challenges.

As AI continues to evolve, it tends to centralise, making decentralisation a formidable endeavour.

However, the widespread adoption of open-source generative AI models is crucial for advancing decentralised AI infrastructures.

Currently, the focus of decentralised AI leans towards inference rather than pre-training or fine-tuning, given the state of generative AI technology.

For decentralised AI to become feasible, Web3 infrastructures must scale significantly, while foundation models must become more compact and adaptable to decentralised environments.

And given the current landscape, achieving this goal poses considerable challenges.