Does the marriage of AI and blockchain technology represent the next technological revolution, or will this blend of technologies be handcuffed into developmental purgatory?
Our last article explored how the blend of these two technologies has the potential to ignite the next great industrial revolution in everything from healthcare to global supply chains. While much of the popular discussion focuses on the potential benefits of merging AI with blockchain technology, it is equally important to acknowledge and address the very real and challenging obstacles builders and entrepreneurs must overcome.
In this article, we'll examine how the convergence of AI and blockchain technology must address the issues of data quality and integrity, scalability, and energy consumption before it can truly change the world.
Data has always been a critical resource for businesses to leverage. It allows internal analysis of a company's inner workings and enables them to respond with more effective policies. Data can provide an exact diagnosis of a company’s health by sifting through the details and context of the information within the company.
Yet, data in the digital age has become exponentially more valuable. When the internet began maturing in the late nineties and early two-thousands, more of the population started creating digital profiles with valuable footprints. These footprints contained customer preferences that would allow a direct link to the individual lives of those users. The harvesting and marketing of this data laid the foundations for some of the most powerful centralized tech monopolies on the planet. To name a few, Google, Facebook, Amazon, Apple, and Netflix.
Every search request, comment, like, or click of a button has become a point of value coveted by these companies. The staggering volume of data being produced has only compounded the importance of this valuable resource. For context, in 2010, the total global volume of data that was generated was estimated at around two zettabytes. By the end of 2024, it is estimated that that number will rise to one hundred and forty-seven zettabytes, with ninety percent of all data ever produced having been generated in the last two years alone. Social media algorithms, the smartphone revolution, and user-generated content have accelerated this trend, while the introduction of frontier AI models has further fuelled this developmental surge.
This massive treasure trove of data is precisely what has allowed for the accelerated development of AI. AI models require large data caches to analyze in their training phase to produce accurate prediction outputs.
Despite the rise of these centralized tech monopolies creating immensely valuable data silos, these data centers remain largely isolated from each other. This creates a problem with training data's development and overall quality. Larger tech companies quickly introduced restrictions on AI developers' access to certain platforms for data scraping. Most evident was when Elon Musk introduced tweet restrictions to Twitter (X) in February 2024. The gatelocking of user data highlights just how valuable data is in the era of modern AI.
Ironically, AI development's hypergrowth and competitive nature later forced other platforms into collaborating and open-sourcing their base models. This was the case in July 2024 when Meta open-sourced Llama 3.1.
Blockchain developers seeking to create new AI protocols are now facing a similar dilemma. Most layer-one blockchains remain siloed off from other blockchain projects. Users often prefer one protocol for one use case, like trading, and others for borrowing and lending for example. Additionally, users also may adopt differing approaches to engagement across multiple projects, preferring to keep higher volatility projects separated from the rest of their portfolio. While interoperability protocols like Cosmos are trying to solve this problem and create a more comprehensive picture, the issue of fragmented data sets remains. Without the whole picture of user data, the quality of AI protocols can remain fractured and incomplete.
Another vector of attack to be mitigated is how easily the integrity of the data can often be compromised or manipulated. This can be done maliciously through data poisoning or trojan attacks, where false or misleading information is introduced into the raw training data, producing misinformation and hallucinations in outputs. Within blockchain AI protocols, the raw data a model uses to train could be targeted or manipulated just as easily. Smaller isolated Layer 1 blockchains with less transaction volume and larger centralized exchanges that can mask trading volumes are especially vulnerable to this kind of attack. This is partially what led to the collapse of the FTX exchange when they obfuscated and mixed user funds with Alameda Research.
Raw data mishandling can skew AI model outputs, leading to inaccuracies when weights and biases are improperly allocated. Famously, an early version of Google’s Gemini generative AI tool produced images of the founding fathers as people of color. The issue occurred because the Gemini team trained the model to always respond with a diverse cross-section of society regardless of the context. AI models reflect the accuracy of the data they are trained on and the weights imposed upon them. Obfuscating either one can lead to a lack of trust in the model or create misinformation in the results. Blockchain’s integration with AI’s raw data sets could help produce more accurate and transparent results with immutable, incorruptible data.
AI development is not only technologically challenging but also incredibly expensive and difficult to scale. Developer competition remains fierce, and chip manufacturing has struggled to keep pace with demand. It requires immense start-up costs and the ability to attract a limited number of knowledgeable developers that exist within this emerging and hypergrowth industry.
The previously mentioned data silos present another difficult hurdle for the scalability of blockchain AI protocols. Combined with an ever-expanding amount of data that must first be vetted, cleaned, organized, and labeled, the development cost can appear insurmountable for start-ups.
Yet, integrating blockchain technology with AI across Decentralized Physical Infrastructure Networks (DePIN) represents a potentially novel solution to the issue of AI development. By creating modular multi-AI agent systems, a company like Fetch.ai can help reduce the cost of AI deployment. The AIOZ and Koii networks are two DePIN blockchain projects focusing on decentralized computation. They could help lower the computational cost of AI and blockchain protocols by decentralizing and distributing the energy load across unused space. Blockchain technology also enables an immutable record of provenance where a data source originated, helping further address the issue of data poisoning and trojan attacks by identifying them at the source.
The blockchain community, and especially those involved with chains that utilize energy-intensive consensus mechanisms like Proof-of-Work (PoW), are all too familiar with the issue of energy consumption. The energy-intensive nature of AI development is finding itself in a similar crosshairs with the mainstream media and a global audience. In January 2024, the International Energy Agency (IEA) released its forecasting report for global energy usage. At the time, the agency concluded that the data centers used for AI development and cryptocurrency mining accounted for 2% of global energy usage. The second part of the forecast concluded that this number would double by 2026.
Energy consumption has previously been used as an avenue to attack PoW protocols for being wasteful. The concern with energy consumption in AI data centers is that despite the potential benefits AI development is creating, the energy cost is simply too large to ignore. The claim is that AI development presents an energy infrastructure and grid systemic risk that must be addressed. This issue will certainly start garnering more attention as the energy demands for AI protocols is forecasted to double in the next two years. Energy grids will need to expand aggressively over this period to adapt to this new demand.
Fortunately for AI projects, blockchain technology and PoW protocols have a decade and a half of dealing with these kinds of energy branding attacks. They have shown that PoW protocols like Bitcoin can be used for energy grid enhancements by accessing trapped and lost energy in everything from flared gas in oil fields to methane gas mitigation at municipal waste sites. The former CEO of the Texas energy grid ERCOT has often spoken positively about Bitcoin’s ability to accelerate a green energy transition and how it helps stabilize the energy grid through its demand response. There is no reason why AI data centers couldn’t be combined with DePIN architecture for similar results.
By learning from blockchain’s adaptive strategies in energy management, AI projects can mitigate public concern and align with sustainable practices, ensuring that their growth doesn't come at the expense of energy stability.
The early reporting on AI’s immense energy consumption does seem to be more positive than what the PoW protocols have previously seen. The World Economic Forum (WEF) has published articles recognizing how the new energy demand is driving grid adoption and a green energy transition rather than the combative approach they had previously taken against blockchain development. Yet, how this transition occurs in the short and medium term may still be a point of contention with the general public if it starts encroaching on the everyday usage of energy for people.
We are living in a period of great technological disruption. With disruption comes challenges that must be overcome. While much of the excitement around the convergence of AI and blockchain technology sounds hyperbolic, the real tangible impacts it is starting to have on the world are undeniable.The challenges of energy usage, scalability, and trust in the raw data fed into these models is not trivial.
An immense amount of commitment from developers and entrepreneurs is needed to navigate these roadblocks and arrive at innovative solutions. However, if we can take any solace, it is that the blockchain community has been developing novel solutions in everything from zero-knowledge protocols to Layer 2 scaling solutions and more energy-efficient consensus mechanisms for the past fifteen years. If anyone is up to the task, it is the builders.