Ce diaporama a bien été signalé.
Nous utilisons votre profil LinkedIn et vos données d’activité pour vous proposer des publicités personnalisées et pertinentes. Vous pouvez changer vos préférences de publicités à tout moment.

Decentralized Markets for Data and Artificial Intelligence

1 117 vues

Publié le

Society is becoming increasingly reliant on data, especially with the advent of AI. However, a small handful of organizations with both massive data assets and AI capabilities have become powerful with control that is a danger to a free and open society.

With the help of blockchain, tokenomics and privacy-by-design, Ocean Protocol aims to unlock data, for more equitable outcomes for users of data, using a thoughtful application of both technology and governance.

Publié dans : Internet
  • Soyez le premier à commenter

Decentralized Markets for Data and Artificial Intelligence

  1. 1. A Roadmap for Public Utility Networks Dimitri De Jonghe @DimitriDeJonghe Research, co-founder - Ocean Protocol Co-founder BigchainDB, IPDB, ascribe, Interledger, Spherity
  2. 2. Communication Economy Community Knowledge Energy & Materials optimists realists pessimists enablers Computing Intelligence Public funding For paradigm shift Moore’s law Roadmap for Public Utilities
  3. 3. Our Economy = Trusted Institutions $ ping www.google.com < PING 172.217.17.36
  4. 4. Our Economy = Trusted Institutions $ ping www.google.com < PING 172.217.17.36 Trust Tax
  5. 5. But… Transparency?
  6. 6. Centralized Economies from Silo-ing Resources I’ll only open my data set if I gain insights and/or profit from it!
  7. 7. But… Private Data?
  8. 8. But… Provenance & Attribution
  9. 9. Here’s your personal data We “forgot” an ownership layer...
  10. 10. The tragedy of the commons is an economic theory of a situation within a shared-resource system where individual users acting independently according to their own self-interest behave contrary to the common good of all users by depleting or spoiling that resource through their collective action.
  11. 11. Users Applications B2C Service $$$ Services Consumer goods E-commerce … ConsequencesAppusage Datagathering&serviceprovision Datatrading&selling ▪ Lack of data control and multiple versions of “you” on different databases ▪ No right to be forgotten ▪ No user share in ad revenue Individual How to farm Digital Me (d-Me)?
  12. 12. Where’s my personal data?
  13. 13. Do you need the middlemen? Can you take control? Is there an incentive to collaborate?
  14. 14. 2008 - Bitcoin founder(s) “unknown” - 2017: 107B$ Market Cap, 27TWh, 7 transactions/sec
  15. 15. Public Protocol “decentralized” Inter- mediary Trusted Intermediaries “centralized”
  16. 16. Adding a layer of digital Trust Application Application Information Trust by Consensus Information Hardware Hardware Internet of information / web 2.0 Internet of value / blockchain / web3.0
  17. 17. Consensus as a Service Coins & wallets > transactions > blocks > mining > chain
  18. 18. The longest chain wins… or forks
  19. 19. Do you need the middlemen? Peer to peer - decentralized Can you take control? Self-sovereign identities Is there an incentive to collaborate? Mechanism design for tokens
  20. 20. Decentralized Communities from Pooling Resources
  21. 21. Blockchain Superpower: Get people to do stuff By rewarding with tokens
  22. 22. Bitcoin goal: maximize security of network Token rewards if: run compute to secure network
  23. 23. Economic Incentive for Bitcoin Objective: Maximize security of network • Where “security” = compute power • Therefore, super expensive to roll back changes to the transaction log E(Ri ) α Hi * T E() = expected value # tokens (BTC) dispensed each block block rewards hash power of actor = contribution to “security”
  24. 24. Bitcoin goal: maximize security of network Token rewards if: run compute to secure network TerraHashes / sec
  25. 25. Ethereum: The World Computer
  26. 26. “Be your own bank” “Value store for e-gold” “Tokenize Networks” “An ICO launch platform” & cryptokitties ETH: Blockchain 2.0BTC: Blockchain 1.0
  27. 27. Filecoin: A Decentralized Storage NetworkFilecoin - a Decentralized File Storage Network
  28. 28. Decentralize the Reward Function (& Tokenize) Proof Stake Reputation Attention ...
  29. 29. Public Utility Networks => Blockchain 3.0? Public Permissionless, rent-free Token is the reward for good behavior > self-sustaining Utility The service delivered is useful work Proof of X drives down margin cost to produce X Network Communities: Horizontal > vertical Building blocks are co-owned, vision is shared
  30. 30. Proof of ... Resource / Service / Utility Block/Client Reward (aka Token) Proof / Stake Token Hashing Power Transactions Business Logic Compute Storage Knowledge Privacy AI/ML training/testing Existence Identity Burn Roaming Bandwidth ... PROTOCOL
  31. 31. From Community to Utility Pooling Prediction/Knowledge High Perf. Compute Data & IoT
  32. 32. From Community to Utility Prediction Markets Dec. Compute Data & IoT MarketsKnowledge Pools Com pute Pools DataPools
  33. 33. Decentralizing Data?
  34. 34. Decentralized Me (de-Me): Personal & Private Data
  35. 35. ?DATA Blockchain Self-sovereignty Attribution Commons Data + Ownership[Blockchain] = self-sovereignty
  36. 36. But… Provenance & Attribution
  37. 37. Provenance of TITLE (IP) Provenance of COPIES https://medium.com/ipdb-blog/forever-isnt-free-the-cost-of-storage-on-a-blockchain-database-59003f63e01
  38. 38. centralized application stack FILE SYSTEM e.g. S3, HDFS APPLICATION PROCESSING e.g. EC2, Azure DATABASE e.g. MySQL, MongoDB Bitcoin Blockchain? PLATFORM e.g. AWS, Google App Engine, Heroku CONNECTNETWORKS e.g.TCP/IP HARDWARE
  39. 39. Towards a decentralized application stack FILE SYSTEM e.g. S3, HDFS IPFS, SWARM APPLICATION PROCESSING e.g. EC2, Azure, Ethereum, Hyperledger, Tendermint, Lisk, Corda DATABASE e.g. MySQL, MongoDB BigchainDB PLATFORM e.g. AWS, Google App Engine, Heroku, Eris/Monax, BlockApps CONNECTNETWORKS e.g.TCP/IP,InterledgerILP e-Cash/e-Gold Bitcoin, zCash, Ripple, Blockstream, Multichain HARDWARE IoTA, Riddle & Code, Chronicled, Sawtooth Lake
  40. 40. Immutability Decentralized Control Native Assets Scalable Queryability Operationalized Traditional Databases Traditional blockchains BigchainDB “Big Data” + “Blockchain” - a blockchain database
  41. 41. Blockchain Consensus Database Consensus IMPLEMENT A 2 PHASE CONSENSUS FEDERATION Single Database or MongoDB RethinkDB Database Options Big Data Substrate + Federated Consensus / BFT
  42. 42. But… Personal & Private Data?
  43. 43. Interledger Vertical: Identity Value proposition: Sovereign personal data GDPR, Right to be forgotten, ... ©ITU/L.Berney, (CC BY 2.0)
  44. 44. Users Applications 3th party services Consumer goods E-commerce Automotive Consequences ▪ Users and devices control their data and access can be provided and revoked ▪ Only one single version of your data ▪ Users get share of revenue generated from their data Private Data Access? Data … Requestpermission Share$$$ Bring YOUR DATA to the Service Public Claims Access? Data / Permission @BigchainDB @GETJolocom
  45. 45. Interledger Interledger A protocol for connecting Payment networks Blockchains Banks Mobile money The unbanked
  46. 46. Interledger Vertical: Government Land Registry/Real Estate Value proposition: Low-cost registry Less risk of corruption Low-friction capital markets ©ITU/L.Berney, (CC BY
  47. 47. Digital Twin / Twin of Things - Spherity
  48. 48. Blockchain: a trust network for supply chains Powered by IoT 57 Recordkeeping paper, sms, email, calls, … Inefficiency change records, own versions/formats, disagreements
  49. 49. 10 Tag unique ID to asset Info added to Digital Twin Data on Blockchain Physical path of asset Path of Digital Twin Factory Logistics Retail Customer End of life Info added to Digital Twin Info added to Digital Twin Set Digital Twin inactive Traceability across the supply chain with Digital Twin
  50. 50. Decentralized Database Assets / Permissioning / Provenance Secured Block-chai ned Objects Digital Twin Platform DecentralApplications DecentralApplications Authorised Access Write and Query Data Authorised Access Send and receive data Dash- boards User Enterprise Apps Read and write transaction data Immutable data transaction history of Digital Twins 1) DSaaS = Decentral Software-as-a-Service 6 How The Digital Twin Works
  51. 51. Decentral App Service Open Platform Technology Ecosystem 3 2 1 third party apps DT driven Apps DT Apps Community Apps API API infrastructure and protocols creation of unique identity Car Pass Object Tagging Supply Chain App App App App App … innogy Apps VSE “Design Netz” Innogy renewables Urban Concepts Digital Twin Platform Others 60 Digital Twin (DT) Platform and decentral App services
  52. 52. Innovation as a Luxury 61
  53. 53. DATABASE IPDB CLIENT SIDE APP BROWSER/JS OR MOBILE APP A digital twin for Cars
  54. 54. Provenance + Fraud detection
  55. 55. Decentralized AI
  56. 56. 1 ZB1 16 ZB2 2010 2016 2025 160+ ZB2 1) Source: Apixel 2) Source: Storagenewsletter Data is growing exponentially
  57. 57. 1 ZB1 16 ZB2 2010 2016 2025 160+ ZB2 1) Source: Apixel 2) Source: Storagenewsletter 3) McKinsey % of data analyzed3 But only a small amount is analyzed and shared
  58. 58. Market value of data in 2030 GDP of Germany in 2016 Up to 3.8 trn. USD1 3.5 trn. USD2 1) Source: Ocean Protocol analysis 2) Source: Worldbank Unlocking data will open up Trillions in value
  59. 59. https://en.wikipedia.org/wiki/File:HTM_Hierarchy_example.png Deep Learning: Neural Networks * Moore’s Law ≈1950s algorithms on 1000x+ more storage & compute
  60. 60. Deep learning models with >> capacity Error 5% .. 0.01% Models with limited capacity Error 25% .. 5% Another 1000x more data Deep Learning Loves Data
  61. 61. ▪ AI needs data ▪ Without data, AI models are not accurate ▪ 150 fundamental use cases across all industry sectors identified ▪ AI advances 6x faster, if data is available AI is starving for data
  62. 62. Have lotsa data (enterprises) Have lotsa AI (AI startups)$$
  63. 63. Data is siloed. Humans are farmed. AI monopolies threaten our future. Let’s change the rules of the game with incentives. Democratize access to Data & AI! Self-sovereignty, attribution & privacy is core Dimitri De Jonghe @DimitriDeJonghe Vision & Mission
  64. 64. Have lotsa AI (AI startups) Have lotsa data (enterprises)
  65. 65. Have lotsa AI (AI startups) Have lotsa data (enterprises) Market Market Market Market Market Market Market Market Ocean Protocol A new data economy
  66. 66. Blockchain-Based Machine Learning Marketplaces https://medium.com/@FEhrsam/blockchain-based-machine-learning-marketplaces-cb2d4dae2c17 Open markets + incentives + attribution impacts the intelligence supply chain at large scale Data providers: Data pooling AI/ML experts: Model Competition Infrastructure: Ad hoc/Flexible Model users: Commons, Enterprise, Personal PRIVACY...
  67. 67. Decentralized AI: Companies can deploy models to be trained or used in the field without risking their intelligence being stolen. Protected Consumer Privacy: the previous application opens up the possibility that consumers could simply hold onto their data, and "opt in" to different models being trained on their lives, instead of sending their data somewhere else. Companies have less of an excuse if their IP isn't at risk via decentralization. Data is power and it needs to go back to the people. Controlled Superintelligence: The network can become as smart as it wants, but unless it has the secret key, all it can do is predict jibberish. Privacy-preserving Decentralized AI http://iamtrask.github.io/2017/03/17/safe-ai/
  68. 68. Client contracts metadata block N-3 proofs accounts Client Service API Keepers Node N2 N4 VerifierService API Service API Proof ask bid access authorize verify access challenge 1. Contract setup 2. Access Control 3. Verification verify proof CONTRACT 0x12345... Lock: Consumer.fee Execute: Proof.valid Abort: Timeout response contracts metadata block N-2 proofs accounts contracts metadata block N-1 proofs accounts contracts metadata block N proofs accounts
  69. 69. CONTRACT 0x12345... Lock: Consumer.fee Execute: Proof.valid Abort: Timeout Data Source: ● Tokenized access control ● Curated Decentralized Model ● Community owned ● Benefits for the commons Service (Data + Proof) Reward (Token)
  70. 70. CONTRACT Lock Execute Abort CONTRACT Lock Execute Abort Data Service: ● Royalty Attribution ● Contract-driven supply chain ● Algorithms, labels, visualizations, ... % Royalty Reward (Token) Reward (Token)
  71. 71. CONTRACT Lock Execute Abort CONTRACT Lock Execute Abort CONTRACT Lock Execute Abort CONTRACT Lock Execute Abort CONTRACT Lock Execute Abort CONTRACT Lock Execute Abort CONTRACT Lock Execute Abort CONTRACT Lock Execute Abort CONTRACT Lock Execute Abort CONTRACT Lock Execute Abort Autonomous Machines Personalized Recommenders Global Monitor PRIVACY
  72. 72. Federated Learning + Privacy https://research.googleblog.com/2017/04/federated-learning-collaborative.html https://github.com/OpenMined/Docs http://iamtrask.github.io/2017/03/17/safe-ai/ OpenMined Partial Homomorphic Encryption
  73. 73. Service Integrity: Crypto-Proofs of Service
  74. 74. Verifiable Computation & Execution procedure circuit procedure circuit transcript computation + setup input execution output query transcript response verify response more query - response verifier prover Or Secure enclaves: - SGX - ARM Trustzone
  75. 75. Protocol realizes Function Multi-party Computation
  76. 76. Proxy re-encryption
  77. 77. Secure infrastucture: [virtual] HSMs Public storage & compute Network glue for scale Public data & intelligence marketplaces Decentralized Intelligence Ecosystem
  78. 78. Community Governance “Make opinion scarce” Crypto-economic Primitives Token-curated registries, curation markets, stake machines https://medium.com/@DimitriDeJonghe/curated-governance-with-stake-machines-8ae290a709b4
  79. 79. What unlocking AI data & services unlocks
  80. 80. Self-driving cars: fewer accidents, more mobility
  81. 81. >100x more data for health care research Personalized Care, Medicine, Prevention and Coaching
  82. 82. Universal Recommender: Push => Pull Marketing
  83. 83. Dimitri De Jonghe @DimitriDeJonghe

×