AI & Blockchain: An Introduction

  • 910 浏览

911

2018/07/22 发布于 技术 分类

AI  blockchain  区块链 

文字内容
1. AI & Blockchain: An Introduction Matt Turck, FirstMark Capital June 28, 2018 Brains and Chains: aiblockchainconf.com
2. Early stage venture capital firm based in New York City SELECT FIRSTMARK MINaVchEinSeTLMeaErnNinTgS& AI Crypto & Blockchain MATT TURCK Managing Director Twitter: @mattturck Blog: mattturck.com
3. Hype… and more hype
4. Defining Technologies of the Last Decade Social Mobile Cloud
5. But Only Obvious in Hindsight AWS = $17.5B in revenues (2017)
6. Defining Technologies of the Next Decade AI (ML, data science, big data) Blockchain (decentralized AI infrastructure, apps, crypto tokens) Blockchain IoT IoT (intelligent infrastructure, autonomous transportation, IIoT, robotics)
7. AI + Blockchain: Strange Bedfellows? “ Crypto is libertarian, AI is communist “ - Peter Thiel “ Crypto is anarchy, AI is the rule of law “ - Reid Hoffman
8. Opposing Paradigms Crypto • Decentralized • Open • Transparent • Deterministic AI • Centralized • (Mostly) Closed • Black Box • Probabilistic
9. Government Surveillance?
10. Not Just in China…
11. GAFA Monopoly and Platform Dependency “Over time, the best entrepreneurs, developers, and investors have become wary of building on top of centralized platforms. We now have decades of evidence that doing so will end in disappointment. In addition, users give up privacy, control of their data, and become vulnerable to security breaches. These problems with centralized platforms will likely become even more pronounced in the future.” – Chris Dixon
12. Blockchain as a foil against the pitfalls of AI Two big ideas: • Decentralized AI Marketplaces • AI Networks & Decentralized Autonomous Organizations (DAO)
13. Also (but for another day) • Several ways AI can help improve blockchain • More efficient mining (optimize energy consumption) • Increase scalability (data sharding) • Help with detection of fraudulent activity
14. Decentralized AI Marketplaces: The Concept What if… … every individual and business could provide their data completely privately and securely in an open data exchange? … AI models could compete with each other to provide the best results? … everyone could be compensated fairly for participating in the above? Wouldn’t we… … end up with MORE data than is currently available to GAFA and other centralized entities? … also have NEW and BETTER data? … and therefore better models, and better AI? … but also more transparent AI
15. Decentralizing the Building Blocks of AI (Lots of) Data Models Computing Power
16. Wait… isn’t the blockchain bad infrastructure for AI? Among other issues: • Not scalable • Heterogenous nodes  hard to arrive at the same machine learning outputs
17. BigchainDB: Building a Scalable Blockchain Database Approach: • Start with an enterprise-grade distributed database • Engineer-in blockchain characteristics
18. Ocean, Computable Labs: Building Decentralized Data Protocols • A tokenized service layer to allow data to be shared and sold in a secure manner • Stores metadata (i.e. who owns what), links to the data itself, and more • On top of the protocol, there can be numerous data marketplaces and exchanges, all accessing the same data
19. Snips: Decentralized Data Generation • Sometimes the data does not exist! (e.g. consumers have never spoken to their coffee machine before) • Create “fake” user data by generating thousands of training examples • Snips is creating a decentralized network incentivized by the upcoming Snips AIR token
20. Gems, Effect: Decentralized Mechanical Turk for Data Labeling • Decentralized, interactive marketplace for micro tasks that require human intelligence • When workers compete a task, they are paid with a network token
21. Secure Computing Goal: Training models on data while keeping the data private
22. OpenMined: Private Machine Learning Protects data owners using:'>using: • federated learning • differential privacy Protects model owners using:'>using: • multi-party computation • homomorphic encryption
23. Algorithmia: Selling Models • Developed a protocol for crowdsourcing ML models • Buyers supply data and offer rewards to incentivize model submissions • Model validation takes place on a public blockchain • If model meets requirements, payment is sent via smart contract
24. Numerai: Creating Competition Among Models • Building a stock market meta-model through crowdsourcing • Participants get data and in return submit models • Incentives are aligned through staking and rewards • Model validation is centralized but payment takes place via smart contract
25. DeepBrain Chain: Decentralized Cloud Computing Platform • Vision is to become “The AWS of AI” • Help AI companies save up to 70% of computing costs • Leverages idle GPU / FPGA clusters and individual computing units owned by SMEs
26. Putting It All Together: Decentralized Data Marketplaces • Tokens & Crypto Economics to solve the cold start problem and incentivize participants • Network effects: “Multi-sided network effects from users, data providers, and data scientists make the system selfreinforcing.” — Fred Ehrsam Via Fred Ehrsam, “Blockchain-based Machine Learning Marketplaces”
27. Next Big Idea: Bots / AI networks • Currently: lots of companies building “AI for X”: Specialized bots • What happens when we have many bots covering individual tasks? Can they be combined for more ambitious projects? • Can this be done in an open manner?
28. Fetch: Autonomous Economic Agents • AEAs = autonomous digital entities that can transact independently of human intervention and can work together to construct solutions • Open economic framework = “the ultimate value exchange dating agency” • Smart ledger (scalable) • Example use case: organize complex trip (predict misconnections, dynamically reroute trips, etc.)
29. SingularityNET: “The Global AI Network” • Decentralized marketplace for transacting with AI agents • AI agents could include neural net tools, machine vision toolkits, etc. • Uses staking and reputation to optimize discovery of the best AI agents • AI agents can sub-contract tasks to more specialized AI agents
30. Botchain: Secure Identity System for Autonomous AI Agents • Bot identity and validation • Bot audit and compliance • Control boundaries of autonomy • Shared marketplace for bot add-ons
31. AI-Powered Decentralized Autonomous Organizations What is an AI DAO? A computational system that runs autonomously on decentralized infrastructure Feedback loop continues by itself, taking inputs, using resources, producing outputs Example A decentralized, autonomous Uber service of autonomous vehicles
32. Only one issue with AI DAOs… You wouldn’t be able to turn them off… ever.
33. Conclusion • Fascinating area • Combining two individually challenging areas = major levels of complexity • All very early and experimental • But a lot of infrastructure is being built • Now is the time to think through the implications