Why is the integration of blockchain and AI considered a natural necessity?
TechFlow Selected TechFlow Selected
Why is the integration of blockchain and AI considered a natural necessity?
From a trend perspective, AI has a natural demand for blockchain, as AI needs blockchain to provide genuine resilience for its development.
Navigating Web3 tides with focused insightsBlockchain is also one of the most important trends
AI has gained far more traction than blockchain this year. However, the crypto world shouldn't be discouraged. How should we understand the future opportunities in blockchain? Let me share some thoughts:
-
Blockchain is one of the most significant trends in human history. The evolution from Web2's information internet to Web3's value internet is equally essential for advancing productivity. In just over a decade, it still has decades of development ahead. Its underlying impact ranks second only to AI among current technologies.
-
There is a genuine need for convergence between AI and blockchain, although progress may not come quickly.
Today, let's focus on the second point: the demand for integration between AI and blockchain.
How blockchain can help AI
Computing
As everyone knows, AI requires massive computing power. There is clear demand for utilizing idle computational resources to support AI workloads. However, currently, training AI models involves intensive computation and remains very expensive. In terms of general-purpose AI computing, blockchain’s ability to contribute is still limited.
Three main criticisms stand out: first, the need for specialized GPU hardware; second, data transmission latency; third, proving the completion of decentralized computing tasks.
-
As mentioned above, AI training involves large-scale intensive computation. LLMs have billions or more parameters, requiring enormous FLOPs during training. Only specialized hardware—such as AI GPUs with dedicated components like Tensor Processing Units—can deliver optimal performance. Moreover, for best results, all GPUs should ideally be homogeneous, enabling synchronized data exchange and continuous computation. In a decentralized network, this imposes requirements on participants’ GPUs. Higher requirements mean higher barriers to entry, which undermines decentralization and limits the use of idle computing capacity.
-
AI GPUs require constant data exchange. Network latency makes distributed computing less viable for AI model training.
-
Efficient and low-cost solutions are needed to verify task completion in a decentralized manner.
These challenges explain why integrating decentralized computing with AI remains difficult today—the primary reason why AI-blockchain synergy hasn’t yet taken off. That said, from BlueFox Notes’ perspective, as more players explore this space, these obstacles will gradually be overcome, though it will likely take considerable time.
Now let’s discuss areas where solutions may emerge incrementally. If the crypto space struggles to enter general-purpose AI now, perhaps it can start with niche AI domains—especially those closely tied to current AI computing demands. Two key aspects stand out: first, inference tasks dominate current AI computing needs; second, certain fine-tuning and inference tasks require fewer resources, making them feasible candidates for decentralized computing. These suggest potential opportunities for decentralized compute networks.
Specific fields such as law, medicine, finance, education, and data analytics might benefit early from specialized decentralized computing networks. As noted earlier, the challenge isn’t performing the computation itself, but rather verifying its completion in a decentralized way. Some projects, such as Gensyn and Together, are already working on this problem.
Gensyn integrates academic research including probabilistic learning proofs and graph-based precise localization protocols, while also drawing from Truebit’s incentive and check-and-balance model. Gensyn breaks down the entire process into eight stages: task submission, parsing, training, proof generation, proof verification, challenge, arbitration, and settlement. "Probabilistic learning proofs" establish baseline threshold distances, providing verifiers with a reference standard. "Graph-based precise localization" monitors how verifiers execute validation. Truebit’s game-theoretic model ensures rational behavior among participants. For full details, refer to Gensyn’s whitepaper. It’s worth noting that off-chain computing projects similar to Truebit could evolve in this direction and potentially unlock new business opportunities—if their teams choose to pursue them.
Compared to decentralized computing, AI model sharing and AI data sharing offer faster paths to real-world adoption. The following two areas may represent early breakthrough points for AI-blockchain integration: decentralized model sharing and decentralized data sharing.
Models
Token incentives can encourage model sharing, leading to better models. These models could even be deployed on-chain and jointly trained by any participant, accelerating development. Additionally, as AI models grow increasingly complex, trust in inference becomes critical—this is where on-chain verifiable inference comes into play.
In the domain of model fine-tuning and inference, projects like Giza, ChainML, Bittensor, and Modulus Lab are actively exploring. Giza offers an on-chain model marketplace, deploying lightweight models directly on-chain for on-chain inference, allowing model owners to earn fees whenever their models are used.
Modulus introduces the concept of zkML, arguing that running inference models fully on-chain is impractical due to cost. Their solution runs inference off-chain, then generates zkSNARK proofs that are submitted to the chain, where smart contracts utilize them.
Data
Token economics can incentivize users to provide feedback on models and collect higher-quality data. Distributed data collection can yield high-quality datasets—particularly valuable in specialized domains—and significantly advance AI development. This approach can also integrate zero-knowledge (ZK) technology, preserving privacy without exposing sensitive underlying data. The main challenge lies in proving the quality of the data itself.
Combining high-quality data with decentralized AI models opens up exciting possibilities for AI advancement.
Anti-forgery
Since the rise of deep learning models, distinguishing between real and AI-generated images, audio, and video has become increasingly difficult. In the age of AI-generated content, authenticity and tamper resistance are growing concerns. Blockchain serves as a crucial technological tool to address these issues.
Cryptographic identity and digital signatures ensure content authenticity, preventing forgery. This becomes especially critical as AI tools are increasingly misused. Cryptography thus becomes a vital defense against fabricated content. In an era where fakes can easily pass as real, cryptographic techniques are essential for distinguishing truth from deception.
Moreover, blockchain can assist in establishing ownership rights. For instance, when viewing a piece of digital art, it may be impossible to visually distinguish whether it was AI-generated or minted as an NFT. This is where blockchain proves indispensable.
More resilient AI
By integrating with blockchain, AI gains support across computing, models, data, bandwidth, and storage—leveraging decentralized infrastructure to enhance self-evolution capabilities. Furthermore, blockchain’s strengths in encrypted payments and value transfer can further empower AI’s evolutionary trajectory.
Once a robust blockchain infrastructure matures, AI will gain greater autonomy and self-improvement capabilities. In other words, a more decentralized AI aligns with AI’s own developmental needs. Leveraging blockchain’s distributed nature to advance AI reflects an intrinsic requirement of AI itself.
For AI itself, being monopolized solely by tech giants like Microsoft or Google would hinder its long-term evolution. AI has a natural need for decentralized growth—a necessity for building a more resilient system. The combined power of AI and blockchain may exceed what people currently imagine.
How AI can advance blockchain
Integration of AI and on-chain data
By applying AI to analyze dynamic on-chain data, predictive capabilities become possible—for example, in investment research. One particularly exciting prospect is embedding AI within smart contracts to enable autonomous, dynamic decision-making. For instance, DeFi protocols could adjust automatically based on real-time market conditions. Dynamic—not static—smart contracts open up entirely new application scenarios and user demands.
Advancements in AI can unlock new possibilities for crypto applications.
AI brings new potential to DeFi, Web3 gaming, Web3 social platforms, and broader Web3 applications (transportation, accommodation, travel, etc.). For example, AI-powered Web3 games could give birth to unprecedented gameplay experiences; AI combined with IoT and crypto payments could create smarter, self-operating networks.
The importance of ZKP
To ensure both privacy and completeness in computational tasks, ZKPs must be integrated to form verifiable proofs of work. Once ZKP technology matures, it will enable AI operations on-chain, offering privacy-preserving and verifiable machine learning.
Overall, blockchain can provide a collaborative framework via decentralized protocols for compute, data, and models—ultimately fostering AI development. Many technical details still need refinement, such as efficiently verifying contributors’ inputs (whether compute, data, or models). Only when these can be achieved at low cost can blockchain truly support AI; otherwise, the vision remains theoretical.
That said, in terms of long-term trends, AI has an inherent need for blockchain—to build true resilience into its own development.
At the same time, AI can drive the evolution of blockchain applications. Whether in DeFi, gaming, or other sectors, we may see the emergence of more intelligent crypto-native applications. This could become the next major narrative—even if it doesn’t fully materialize in the next cycle, the opportunity may arise in the one after.
The points above are partial and incomplete, subject to change over time. Additional insights are welcome in the comments. (This article was written two months ago but forgotten; publishing now.)
Risk warning: All analysis here represents a narrow view of technology and markets, and may not be accurate. Please maintain independent judgment and practice proper risk management.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
蓝狐笔记
