How blockchain technology helps to build credibility and neutrality for AI

published for DAO UTokyo https://www.iii.u-tokyo.ac.jp/event/230921event

---

Abstract:
The integration of blockchain technology with artificial intelligence (AI) presents a novel approach to addressing the inherent issues of transparency, centralization, and bias that plague current AI systems. This paper explores the synergy between blockchain's immutable and transparent nature with AI's need for vast, unbiased data sets, offering a path toward decentralized AI systems. We discuss the potential for blockchain to democratize AI development, ensure data authenticity, and promote neutral, bias-free AI decision-making processes.

Key Points:

1. Transparency and Accountability:
AI's current opacity is a significant hurdle. Blockchain can inject transparency by providing a verifiable record of AI decisions and data provenance.

2. Decentralized Data Management:
Centralized data repositories lead to monopolies and censorship. Blockchain enables a decentralized data infrastructure, enhancing data security and accessibility.

3. Bias Reduction:
Ensuring diversity in data annotation and model training is crucial. Decentralized platforms can facilitate broader participation, reducing systemic biases.

4. Innovative Solutions and Case Studies:
Decentralized AI models, like Numer.ai, and platforms for decentralized annotation showcase the practical application of these principles. The EU's initiatives for machine-readable data mining opt-outs and the Cross-Platform Origin of Content Framework are highlighted as regulatory advancements supporting this cause.

5. Web3 Crowdsourcing for AI:
Exploring methods such as peer prediction, staking, and prediction markets, we propose Web3-based incentives for crowdsourcing honest data, ensuring a diverse and equitable AI development landscape.

Conclusion:
Blockchain's inherent properties offer a foundation for building credible and neutral AI systems. By fostering transparency, decentralization, and diversity, blockchain technology not only addresses AI's current limitations but also aligns AI development with ethical standards and societal needs.

Transcript

1. Taka (Shunsuke Takagi), DAO UTokyo How blockchain technology helps to build credibility and neutrality for AI
2. What's the next billion on crypto? →my idea: with AI? to build credibility and neutrality for AI
3. • AI is opaque, crypto brings transparency • AI needs data, and blockchains are good for storing and tracking data blockchain + AI intersections
4. • Centralization sometimes makes(causes) censorship • Who annotated the data? • (human/cognitive) biases • Annotation work is $2/h by Kenyan Workers? current AI program(problems?): monopoly & opaque(intransparency)
5. Criteria for “Decentralized AI” to accomplishing credibility and neutrality • Forkable (OSS)? • ⚠ But, vulnerability to adversarial machine learning attacks? • Is data collecting licensed? • Does the annotation process have a diversity of labor? (For avoid bias)
6. Some good news for AI building
7. Machine-readable opt-outs from text data mining Now drafting by EU legislative term
8. Linking digital accounts to content Cross-Platform Origin of Content Framework makes it easy to verify for Digital content, like “robot.txt”.
9. Linking digital accounts to content Cross-Platform Origin of Content Framework makes it easy to verify for Digital content, like “robot.txt”.
10. Linking digital accounts to content Cross-Platform Origin of Content Framework makes it easy to verify for Digital content, like “robot.txt”.
11. Linking digital accounts to content Cross-Platform Origin of Content Framework makes it easy to verify for Digital content, like “robot.txt”.
12. Decentralized Annotation OpenAssistant collected volunteers’ annotation. (And completed!)
13. Decentralized Annotation OpenAssistant collected volunteers’ annotation. (And completed!)
14. Decentralized Annotation OpenAssistant solicited volunteer annotations. (And completed!)
15. Decentralized AI Model Numer.ai is the AI hedge fund which build by Decentralized Kaggle competition
16. Decentralized fine-tuning “Democratic Fine-tuning with a Moral Graph” (DFTmg), is a winner of the OpenAI democratic process grant
17. Decentralized fine-tuning “Democratic Fine-tuning with a Moral Graph” (DFTmg), is a winner of the OpenAI democratic process grant
18. Decentralized Annotation OpenAssistant solicited volunteer annotations. (And completed!)
19. Decentralized Annotation OpenAssistant solicited volunteer annotations. (And completed!)
20. Decentralized fine-tuning “Democratic Fine-tuning with a Moral Graph” (DFTmg), is a winner of the OpenAI democratic process grant
21. Decentralized Annotation OpenAssistant solicited volunteer annotations. (And completed!)
22. Decentralized Annotation OpenAssistant solicited volunteer annotations. (And completed!)
23. Crowdsourcing is so-so working (Almost AI depends on Wikipedia)
24. How to make a creditable neutral crowdsourcing process? → crypto can be the incentive as a money
25. DeCartography: Decentralized Annotation platform via web3 crowdsourcing
28. potential methods for Web3 crowdsourcing
29. Peer Prediction Method elicit truthful (i.e., honest) responses from a set of respondents when they report. Crypto will be an incentive?
30. Staking Skin in the game
31. Prediction Market polymakert, omen…etc is working on crypto!
32. 1person, 1ID Bright ID, Worldcoin, Web of Trust, Gitcoin Passport