Disclosure: The opinions and opinions expressed here belong to the authors solely and do not represent the views or opinions of the crypto.news editorial.
All artificial intelligence is rage. However, there is a significant flaw under the hype surrounding decentralized AI (DEAI). There is a lack of diverse, safe and verifiable data. On-chain datasets are too limited to train truly powerful models. This risks passing the future of AI to a centralized giant.
DEAI’s Promise – Democratized, transparent and robust AI depends on filling this data gap. Skillful encryption provides the root.
The beauty of traditional AI lies in its gluttony. The more data you devour, the smarter you become. But this advantage is also the heels of Achilles. Centralized AI models are trained with data that are often harvested without explicit consent, raising troublesome questions of privacy and control.
Built on Blockchain’s principles of decentralization and transparency, DEAI offers an attractive alternative. However, most data-on-chain comes from financial transactions or debt. Small language models in particular require more accurate data for fine-tuning. This makes DEAI models hungry for the rich and diverse datasets needed to be refined to the competitive level that is expected of the latest models.
Such datasets are available outside of Web3, with piles and general crawlings each containing data from billions of unique sources. Like the amount of data, the depth of existing validated Web2 data sources has enabled centralized AI providers to refine GPT widely and quickly.
Reproducing the same level of data-on-chain is not feasible on competitive timescales. Also, some AI companies are violating data creators who accuse them of stealing the type of subtle data we’ll explain here, but there is another way to get more data on-chain .
Bridge building
This is where encryption comes in. Zero knowledge proof already making waves in blockchain scalability and privacy offers a powerful solution. In particular, two techniques, zero-knowledge fully approximate encryption (ZKFHE) and zero-knowledge TLS (ZKTLS) hold the key to unlock DEAI’s Web2 data.
ZKFHE allows you to perform calculations without decrypting encrypted data. Imagine training an AI model for sensitive medical records without releasing raw patient data. This is Zkfhe’s power. The DEAI model can now be trained from a vast, privacy-protected dataset, greatly expanding the training possibilities.
ZKTLS extends this principle to Internet communications. This allows users to prove ownership of certain data from the website without revealing the underlying information. This is important for integrating the rich data that exists in Web2 silos into the DEAI system. For example, a decentralized credit scoring model can leverage ZKTLS to access certified financial data from traditional institutions without compromising confidentiality.
Benefits, DEAI?
The meaning is profound. By combining ZKFHE and ZKTLS, DEAI can take advantage of the vastness of Web2’s data while maintaining the central tenets of privacy and decentralization. This could level the arena, allowing DEAI to compete and perhaps even outweigh the intensive AI.
Consider developing large-scale language models dominated by currently well-funded, high-tech giants. These models require a huge amount of textual data for training. By leveraging ZKTL, DEAI developers can provide privacy to publicly available web data and create more democratic and transparent LLMs.
Of course, there are challenges. The implementation of ZKFHE and ZKTLS is computationally intensive and requires significant advances in hardware and software. Standardization and interoperability are also important for widespread adoption. However, the potential rewards are immeasurable.
In the race for AI hegemony, data is the ultimate fuel. By employing encryption solutions such as ZKFHE and ZKTLS, DEAI has access to the fuel it needs to run. This doesn’t just build smarter AI. It is about building a more democratic and equitable future of AI.
