To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system,...

40
Turing Plan To Develop a Blockchain-based Computing System for Decentralized Artificial Intelligence TuringPlan Foundation United Kingdom Technical Wihtepaper Version 1.1 28/03/2018

Transcript of To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system,...

Page 1: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

Turing Plan

To Develop a Blockchain-basedComputing System for Decentralized

Artificial Intelligence

TuringPlan FoundationUnited Kingdom

Technical Wihtepaper

Version 1.1

28/03/2018

Page 2: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

“A computer would deserve to be called intelligent if it could deceive a

human into believing that it was human.” –Alan Turing

Page 3: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

Abstract

With the emergence of deep learning, the past five years have witnessed boom-ing development of artificial intelligence (AI). The success of modern AI canbe largely attributed to today’s super computing power. Despite that comput-ing power is developing rapidly, most of the large-scale computing systems arecentralized (i.e. owned by private enterprises or institutes), which has largelyhindered the innovation in AI technologies. To address the issues of central-ized AI, we propose to develop a Blockchain-based computing system, whichis able to connect ANY possible computing unit on the internet (even includ-ing PCs). As a core of the proposed system, Blockchain serves as the mediato connect the computing resources. In this system, AI tasks are divided-and-conquered in a decentralized manner. Every computing node connected tothis system can participate by publishing new tasks or serving as computingresources.

Different from traditional distributed systems, the proposed system not on-ly deals with computing tasks but also supports the consensus-based trans-actions, thus constructing an ecosystem of decentralized AI. Technically, theproposed system runs on a specially tailored public chain based on a two-tiernetwork, where the tier-one subnetwork deals with AI task decomposition andbroadcasting, and the tier-two subnetwork deals with AI subtask processing.In addition, we have designed a new multi-criterion consensus algorithm todetermine which nodes are miners or validators.

The proposed system has deep and wide market potential involving variousapplication areas. In the long run, hopefully, the proposed system will even-tually lead us to the era of artificial general intelligence. In memory of thegreat computer scientist – Alan Turing, we call this project the Turing Plan.Correspondingly, the proposed system is named as the Turing Computing (TC)

System .

Page 4: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

Contents

1 Introduction 1

2 Design Principles 32.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32.2 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.2.1 Task Publisher . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42.2.2 Task Processor . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2.3 Unique Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

3 Technical Details 73.1 Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73.2 Two-tier Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83.3 TNP Token System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103.4 Multi-criterion Consensus Algorithm . . . . . . . . . . . . . . . . . . . . . 113.5 Task Unit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123.6 Turing Virtual Machine . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143.7 Data Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

4 Decentralized AI Paradigms 164.1 Decentralized Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174.2 Decentralized Evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

A People 20A.1 Team Members . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20A.2 Advisors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

B Token Allocation 25

i

Page 5: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

C Market Potential 26C.1 Computer Vision and Pattern Recognition . . . . . . . . . . . . . . . . . . 27

C.1.1 Pixel Restoration for CSI . . . . . . . . . . . . . . . . . . . . . . . 27C.1.2 Fashion Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . 28C.1.3 Machine Translation . . . . . . . . . . . . . . . . . . . . . . . . . 28

C.2 Computer Games, Robots and Self-driving . . . . . . . . . . . . . . . . . . 29C.2.1 AI Game Engines . . . . . . . . . . . . . . . . . . . . . . . . . . . 29C.2.2 Multi-functional Robotics . . . . . . . . . . . . . . . . . . . . . . 29C.2.3 Self-Driving . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

C.3 Trip, Traffic and Shipment Routing . . . . . . . . . . . . . . . . . . . . . . 29C.4 Finance and Investment Strategies . . . . . . . . . . . . . . . . . . . . . . 32C.5 Automotive Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32C.6 Engineering Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

Bibliography 34

ii

Page 6: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

Chapter 1

Introduction

With the emergence of deep learning [1], the past five years have witnessed booming de-velopment of artificial intelligence (AI). Unlike traditional AI (e.g. statistical machinelearning) techniques, deep learning is motivated to simulate how human brain processesinformation via the modelling of artificial neural networks. DL has substantially improvedthe performance of pattern recognition in pictures, text, sounds, and other data to producemore accurate insights and predictions. Undoubtedly, deep learning is leading a revolutionin modern AI.

The success of deep learning can be largely attributed to one important factor – comput-ing power. Taking the AlphaGO as an example, the distributed version had 160,000 gamesand 29 million board/next-move pairs, using 1,202 CPUs and 176 GPUs as the comput-ing power [2]. Even with such large-scale computing resources, it can still take weeks oreven months to train a single deep learning model. Recently, the deep neuroevolution isbelieved to have opened the new era of AI [3–5]. The basic idea is to combine to deeplearning with another classic AI paradigm – the evolutionary computation [6], thus makingthe deep learning models evovable. As a milestone in the AI history, it is the first time thatthe evolutionary computation paradigm has been systematically applied to deep learning;and more importantly, the novel attempt has achieved surprisingly promising performanceon general AI tasks such as reinforcement learning. However, since deep neuroevolutioninvolves thousands or even tens of thousands of deep learning models to train, it demandsimmensely large-scale computing resources to work. For example, if it takes weeks to traina single deep learning model, in the case of deep neuroevolution, it can take up to yearsto complete the evolution if the same amount computing power is given. In the long run,if the evolution can be scaled to massive parallel computing resources, it can experience arenaissance just as deep learning has, further pushing the boundaries of AI development.

Despite that computing power is developing rapidly with the upgrades of computerhardware and architectures, most of the large-scale computing systems are centralized (i.e.

1

Page 7: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

owned by private enterprises or institutes), which have some intrinsic weaknesses or limi-tations. First, centralized computing systems are vulnerable. If a centralized system wereattacked or encountered incidental faults, it could lead to complete failures of AI tasks run-ning on it. Second, centralized computing systems cost high maintenance fees. Runningand upgrading the whole system could cost a big budget, while leaving the idle computingresources unattended is also a waste. Most importantly, centralized computing systems areinaccessible to the public. Nowadays, deploying AI applications or performing AI researchbecomes very expensive, and the majority of small companies or academic institute cannotafford it. As a consequence, the existing AI market is mostly controlled by tech giants likeGoogle, IBM or Microsoft. In the long run, such a centralized pattern may lead to the mo-nopolization of the AI market, causing unfair pricing, a lack of tranidlency, interoperabilityand limited participation of smaller organizations in AI innovation [7].

To address the issues of centralized AI, we propose to develop a worldwide computingsystem which is able to connect ANY possible computing unit on the internet (even includ-ing PCs). As a core of the proposed system, Blockchain serves as the media to connectthe computing resources. In this system, AI tasks are divided-and-conquered in a decen-

tralized manner. Every computing node connected to this system can participate in thedevelopment of the ecosystem by publishing new tasks, providing solutions or serving ascomputing resources. In the long run, hopefully, the proposed system will eventually leadus to the era of artificial general intelligence. In memory of the great computer scientist –Alan Turing, we call this project the Turing Plan. Correspondingly, the proposed system isnamed as the Turing Computing (TC) System .

The rest of this paper is organized as follows. Chapter 2 details the architecture of theproposed TC system, followed by the technical details in Chapter 3. Chapter 4 illustrateshow to deploy decentralized AI paradigms in the proposed TC system.

2

Page 8: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

Chapter 2

Design Principles

2.1 Background

From the data structure point of view, a Blockchain is a set of continuously linked datablocks. The first remarkable application of Blockchain technique can date back to the Bit-coin born in 2008 [8], where Blockchains serve as media for peer-to-peer transactions ofthe digital currency. Bitcoin is known as the first generation of Blockchain, which main-ly deals with transaction recording, namely, the data. Using such a distributed databasetechnology, the transactions between any two parties can be recorded efficiently and in averifiable and permanent way.

While the first-generation Blockchain mainly focuses on the data, the second-generation(aka Blockchain 2.0), has successfully integrated the capability of computing [9]. Specif-ically, Blockchain 2.0 becomes programmable by running the so-called smart contrasts.This is an important breakthrough as it enables value exchange without powerful interme-diaries acting as arbiters of money and information. Since the emergence of Blockchain2.0 in 2014, the related technologies have gone far beyond transactions, opening a new eraof decentralized data storage and information processing.

Ideally, a decentralized network of computing system based on Blockchain can createa shared economy where anyone with a computing unit can lend idle computing powerand make a side income. The peer-to-peer nature of the Blockchain and distributed ledgerswill also help move computation closer to where the data is being generated, and avoidbottleneck round-trips to cloud servers. While not being a computation platform itself,the Blockchain can potentially create a marketplace application that attacks the specificproblem of linking buyers and sellers of computing resources and allowing them to paythemselves in cryptocurrency without needing an intermediary like AWS [10].

3

Page 9: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

2.2 Outline

As shown in Fig. 2.1, the general outline of the proposed TC system consists of three maincomponents: task publishers, Blockchain, and task providers. Since it is a peer-to-peernetwork, any node connected to it can be either a task publisher or task processor, or evenboth of them. As the media to process the peer-to-peer communications between any pairof nodes in the network, a special Blockchain is designed with three spaces: the transactionspace, the solution space, and the algorithm space. In the following, we further providedesign principles for the task publisher and task processor respectively.

Task Processors

Task Publishers

Transaction

Space

Task

Space

Solution

Space

……

Blockchain

……

Figure 2.1: The general outline of the TC system.

2.2.1 Task Publisher

A task publisher is a node which publishes computing tasks in the TC system. First, atask publisher follows standard protocols to decompose a computing task into a group ofsubtasks. Then, the subtasks are packaged and written to the task space of the Blockchain.Meanwhile, it broadcasts a reward of the task using the cryptocurrency in the TC sys-tem. Once the solutions to the subtasks are completely written to the solution space of

4

Page 10: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

the Blockchain, the task publisher can collect the solutions and repack them into the finalsolution. The pseudo code of the above process is summarized in Algorithm 1.

Algorithm 1: Task PublisherInput: T (Task to be processed), R (Reward for processing the task)Output: S (Solution to the given task T )

1 Decompose the task T into subtasks;2 Write the subtasks into the task space of the Blockchain;3 Broadcast the reward R for processing the subtasks;4 while Not All subtasks are Completed do5 Keep checking the solution space of the Blockchain;

6 Read the solutions to the subtasks and repack it into S;7 return S;

According to Algorithm 1, any task publisher has equal rights to publish a task with areward. However, it is not guaranteed that the published task can be solved if the rewardis not properly given. This provides a healthy and sustainable competition environment ofthe TC system.

2.2.2 Task Processor

A task processor is a node which processes computing tasks in the TC system. First, atask processor can freely collect subtasks from the task space Blockchain based on thetask difficulty and reward. Then, the subtasks are processed locally. Once a subtask iscompleted, the task processor will immediately write the solution to the Blockchain if ithas not been solved yet. Once the solution to a subtask is written to the solution spaceof the Blockchain, the reward will be automatically transferred to the account of the taskprocessor. The pseudo code of the above process is summarized in Algorithm 1.

Algorithm 2: Task ProcessorInput: {ti} (Subtask set), D (Data required for processing the subtasks)Output: {si} (Solutions to the given subtasks {ti})

1 for each subtask in {ti} do2 Compute each subtask ti;3 if ti has not been solved by other task processors then4 Write solution si to subtask ti to the solution space on the Blockchain;5 Collect the reward and write the transaction to the transaction space on the

Blockchain;

6 return S;

5

Page 11: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

According to Algorithm 2, any task processor in the TC system has equal chance tocollect a subtask from the Blockchain. However, only the first one to complete it and writeit to the Blockchain has can get the reward. Such a competition mechanism guarantees notonly a high efficiency but also a good robustness, where the failure of any task processorwill not influence the system.

2.3 Unique Features

In summary, the design principles provide the following unique features of the TC system.

• Decentralization: All the participant nodes in the system share equal ownership tothe network, thus avoiding any central points of failure.

• Scalability: The system can be scaled to any number of participant nodes with littleadditional cost or performance deterioration.

• Concurrency: By decomposing the AI tasks into subtasks, the system allows partic-ipant individuals to execute multiple subtasks concurrently.

• Security: Since the system is running on a Blockchain, it is much less likely to beattacked.

• Accessibility: As peer-to-peer nature of the Blockchain eliminate the intermediaries,any pair of users can interact with each other directly.

• Immutability: Nothing on the Blockchain of the system can be changed once con-firmed and written.

6

Page 12: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

Chapter 3

Technical Details

3.1 Architecture

The proposed TC system is built on distributed network and aims to employ idle computingresources on it for AI tasks. To be specific, the AI tasks are decomposed into a numberof subtasks, each of which is solved by one or more nodes in the peer to peer networkinteracted by message passing.

Peer to peer network

Transaction TVM Transaction TVM Transaction TVM

Write Read

Write Read

Write ReadConsensus Algorithm

Consensus Algorithm

Consensus Algorithm

Task publisher Task processor Task publisher Task processor Task publisher Task processor

Service

Engine

Data

Network

Market

Figure 3.1: The main architecture of the TC system.

Generally, there are four roles involved the TC system are:

• Task publisher A task publisher the role responsible for producing AI task withrewards.

7

Page 13: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

• Task processor A task processor is the role processing the task and receiving re-wards.

• Blockchain Miner A Blockchain miner provides consensus algorithm to supporttrust between the participants by building, signing and broadcasting new blocks tothe other Blockchain validators.

• Blockchain Validator A Blockchain validator is the role as a router offering thefunctionality of validating transactions and blocks and gossip legal ones.

As shown in Fig. 3.1, the TC system consists of five layers: market layer, service layer,engine layer, data layer, network layer. The market layer is an open platform that allowsthe task publishers and task processors to access the TC system. The service layer is thesecond tier and serves for compiling and executions of the subtasks. The engine layer isthe middle tier and provides consensus on the order where subtasks are written. Once thesubtasks are confirmed by validators and miners, they will be submitted to the data layer.Finally, the network layer is an indispensable foundation for the whole system to transferdata and value. In the following sections, we present the details of the proposed TC system.

3.2 Two-tier Network

Beyond the existing systems of the Cryptocurrencies such as Bitcoin [11] or Etherum [12],the proposed TC system adopts a novel two-tier network: the task network and Blockchainnetwork, which is as shown in Fig. 3.2. Under this network architecture, the task processorsand task publishers are all serving in the task network. In addition, the task publishersand task processors are also connected to the Blockchain network, making the Blockchainnetwork act as an agent to route the subtask code and solution data. More specifically,the task publishers produce and publish the subtask code to the Blockchain network, andthe task processors collect the subtask code, execute them and then upload the results toBlockchain network. The Blockchain miners and validators are designed to maintain thepublic ledge in the Blockchain network. We summarize the two-tire network workflows asfollows:

The First-tier subnetwork(task network) runs as follows:

Step 1: A task publisher creates a task and decompose the task into some subtasks;

Step 2: The task publisher broadcasts subtasks to the Blockchain by offering a rewardfor each subtask;

8

Page 14: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

Blockchain Network

Task Network

Task Publisher Task Processor Blockchain

Blockchain ValidatorBlockchain Miner

Figure 3.2: The network structure of the TC system.

Step 3: A task processor collects the subtasks from the Blockchain and stores itssolution data back to Blockchain;

Step 4: The task publisher scans the least public ledge to obtain and merge all thesolution data;

Step 5: Once the task publisher confirms the solution data, it will send the reward tothe corresponding task processors.

The second-tier subnetwork(Blockchain network)runs as follows:

Step 1: A miner competes for collecting new transactions into a block using theproposed multi-criterion consensus algorithm.

Step 2: Other validators accept the block if and only if all transactions mined byminers are valid.

Note Miners are a subset of validators since every mining node is also a validationnode.

9

Page 15: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

3.3 TNP Token System

Incentives are a fundamental element of game theory and economics in shaping humanbehavior towards a common good [13]. In the TC system, the incentive is based on a tokensystem, which is designed to encourage all the participants to maintain the protocol andprovide unprecedented security guarantees for the system. To be specific, the only officialcryptocurrentcy in the TC system is defined as the Turing Plan (TNP) tokens, which havethe following two functionalities:

• To Reward the Task Processors: A task processor will take reward from the taskpublisher once completed the assigned subtasks.

• To Reward the Minors: A minor will take percent from the reward given by thetask publisher for by building, signing and broadcasting new blocks, which includethe transaction as well as the computational results returned by the task processors.

To publish a task, a task publisher need to bid for computing power as requested. Aspart of the transaction, the TC system will automatically recommend a bid price accordingto the time complexity (i.e. difficulty) of the task to be processed, which is independentfrom the specific hardware types (e.g. CPU, GPU) of the task processors. This guaranteesthe uniformity and fairness of the transactions. On the other hand, a task processor willalso ask for subtasks with an expected reward, and once they complete the subtasks andsubmit the computational results, the transaction is successfully completed.

By convention, the first miner who commits the block successfully will take a percentfrom the reward given by the task publisher. This encourages the nodes to compete for therole of the minors and provides a way to distribute TNP tokens into circulation initially [8].It is worth noting that, in contrast to the conventional mining systems such as those in BTCor ETH, our mining system do NOT generate additional tokens, which avoids the potentialrisk of currency inflation and thus guarantee the stability of the token system.

For security reasons, a basic design principle to be noted is that the marginal cost of amalicious behavior of network nodes must significantly higher than any possible marginalgain derived from such malicious behavior [14]. As introduced in the following section,the consensus algorithm of the TC system is based on multi-criterion decision making,which involves a node’s transaction history, transaction volume, deposit and historical taskperformance, etc. In this case, if a greedy attacker is able to launch an attack (e.g. tomanipulate transaction), he/she need to spend a lot of money and time to increase his/hercriteria to obtain the trust of other nodes. This forge progress can be easily found by theirauditors (nodes) with a high risk of heavy penalty.

10

Page 16: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

3.4 Multi-criterion Consensus Algorithm

As illustrated in Fig. 3.3, our TC system adopts a multi-criterion consensus algorithm todetermine the potential miners and validators, which falls into the family of DelegatedProof-of-Stake Consensus (DPOS) [15]. To be specific, minors and validators can be thesame nodes in different time zone, while block number and timestamp switch the role ofthe miners and validators.

Internal Transfer

Hybrid Voting PhaseAlgorithm one -

Stateful algorithm

t

v

...

d

securityparameter κ

Consensus PhaseAlgorithm two -

Output Algorithm

Transa1

Task1

...

TransaN

@ verify z

@ b cverify b verify

SecureBlockchainwith taskexecution

result

Figure 3.3: Illustration of the multi-criterion consensus algorithm

The multi-criterion consensus algorithm is parameterized by two sub-algorithms(∏, ζ),

where the stateful algorithm∏

receives a security parameter κ as inputs and maintains alocal state, and the output algorithm ζ(κ, state) outputs an ordered sequence of records m.Both of the security parameter κ and state are produced by the stateful algorithm

∏, and

typically, the record m is an ordered sequence of task code or transactions. The stateful al-gorithm

∏consists of a validity predicate (denoted by

∏V ) that encapsulates the semanticproperties.

∏V returns 1 if and only if the record m is valid for all the validators. The secu-

11

Page 17: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

rity parameter κ is calculated by multiple criteria including transaction time t, transactionvolume v, deposit d, task performance p and task execution times et:

κ = α ∗∏i...n

(tv) + β ∗∏i...n

(d) + γ∑i=0

n ∗ p ∗ et, (3.1)

where n refers to the current round, and the α,the β,and γ meets the following condition:

α + β = 1, γ ∈ [−1.0 v +1.0]. (3.2)

After a consensus is reached in the decentralized network, the blocks will be enforcedto join in the Blockchain cryptographically, where the block and the transactions in thisblock are authenticated by other participants. Assume that the current network relies onN trusted nodes and a majority (namely at least N/2 + 1) of them are assumed honest,all the nodes are allowed to propose blocks at the same rounds but in different priorities.The package priority is in a round-robin fashion among all the trusted nodes. Under thismechanism, despite that forks may occur [16], the fork possibility is restrained by the factthat each non-leader packer proposes a block but delays its block randomly, hence the blockproduced by the leader miner is likely to be received by all the other participants firstly. Ifforks happen, since the block produced by the leader miner have higher scores as given by(3.1) and (3.2), according to the GHOST [12] protocol, forks will be eventually solved.The pseudo code of the above process is summarized in Algorithm 3.

Algorithm 3: Multi-criterion consensus algorithmInput: T (Transaction),N (Current blocks) ,TN (Trusted nodes)Output: B (Blocks)

1 transaction T received from the neighbors or clients;2 for each trusted nodes in p2p network {TNi} do3 give the Current Blocks Number CBN ;4 obtain security parameter κ;5 if current node = the node which holds the security parameter κ then6 collect the transactions and package them into the block N ;7 broadcast the block to all the trusted nodes;

8 return B;

3.5 Task Unit

A task unit (or unit for short) is an autonomous execution unit stored in the Blockchain,encoded as part of an executable code duplicated in various nodes in the TC system. The

12

Page 18: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

task unit is identified by a unique unit address and consists of executable code for AI tasks,where each unit must hold some amount of TNP tokens, has its private storage and isassociated with its predefined executable code. An exemplified task unit is given as follows.

1 /∗ d e f i n e t h e AI t a s k ∗ /v a r t a s k c o d e = {

3 /∗Source code of t h e AI t a s k

5 ∗ /}

7

/∗ s p l i t t a s k i n t o s u b t a s k s wi th t a s k reward f e e ∗ /9 v a r s u b t a s k c o d e = f u n c t i o n ( fee , t a s k ) {

w h i l e ( i < l e n g t h ( t a s k ) ) {11 s u b t a s k c o d e = append ( s u b t a s k c o d e , t a s k [ i ] )

}13 r e t u r n {

s u b t a s k c o d e : s u b t a s k c o d e ,15 s u b t a s k f e e : f e e [ s u b t a s k c o d e ]

}17 } ( ) ;

19 /∗ compi l e s u b t a s k s o u r c e code ∗ /v a r t a s k C o m p i l e d = t u r i n g . compi l e ( s u b t a s k c o d e )

21

23 /∗ de p l oy t h e compi l ed t a s k ∗ /v a r s u b t a s k c o d e = Task . new ( taskCompi led ,

25 {from : a c c o u n t s [ 0 ] ,

27 d a t a : t a s k C o m p i l e d [ 0 ] ,r eward : 1000TPN

29 } ,f u n c t i o n ( e , t a s k ) {

31 /∗ I f no e r r o r o c c u r e d . . . ∗ /i f ( ! e ) {

33 i f ( ! t a s k . a d d r e s s ) {c o n s o l e . l o g ( ” t a s k t r a n s a c t i o n send : T r a n s a c t i o n H a s h : ” +

t a s k . t r a n s a c t i o n H a s h + ” w a i t i n g t o be mined . . . ” ) ;35 } e l s e {

c o n s o l e . l o g ( ” s u b t a s k c o d e mined ! ” + T r a n s a c t i o n H a s h ) ;37 }

39 } e l s e {/∗ I f an e r r o r occured , l o g t h e e r r o r ∗ /

41 c o n s o l e . l o g ( e r r o r ) ;}

43 }} ) ;

:

13

Page 19: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

3.6 Turing Virtual Machine

The code of a task unit must be written in a full Turing-complete language, which is usedto describe the AI task to be processed. To publish an AI task, the task publishers predefinethe task using a conventional programming language (e.g. Python) and publish it alongwith the transaction via the TC system.

Execution results

Passed

Failed

Blocked

Not run

TVM

Task Publisher Task Processor Internal Dataflow Blockchain Dataflow

Figure 3.4: Illustration of the Turing Virtual Machine (TVM).

To process the AI taks, we design a virtual machine as illustrated in Fig. 3.4, knownas the Turing Virtual Machine (TVM). To be specific, the TVM serves as an interfaceconnecting the Blockchain and the the operation system on each task processor, whichworks as follows:

Step 1: Download the source code from the Blockchain;

Step 2: Pass the source code to the operation system to be complied and executed;

Step 3: Upload the executed results to the Blockchain.

It should be noted that, a tailored client software will need to be installed on each taskpublisher or processor. In order to serve as a task processor, the local environment of theoperation system should also be properly configured. For example, if the task is writtenin Python, then only the task processors with Python environment are able to process thecorresponding subtasks.

14

Page 20: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

3.7 Data Management

Since the TC system is a system running on the Blockchain for dealing with AI tasks,the data consists of three parts, the transaction data, solution data and the algorithm data.The original Blockchain transaction data is essential to maintain the ecosystem, while thesolution data and the algorithm data are core parts of the AI tasks. To save disk space, thesolution data and the algorithm data can be removed once the task is finished and collectedby the client.

15

Page 21: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

Chapter 4

Decentralized AI Paradigms

A necessary condition that an AI task can be decentralized is that the task is separable. Astwo fundamental AI paradigms, learning and evolution cover the majority of AI tasks, andfortunately, most of such tasks are naturally separable.

Essentially, many AI tasks can be essentially seen as an optimization task, e.g.:

arg min~x

f(~x),

s.t. x ∈ X ,(4.1)

where X ⊂ RD is the decision space and x = (x1, x2, ..., xD)> ∈ X is the decision vector,

and D is the number of decision variables.A decision variable xi is known as separable iff:

arg min~x

f(~x) =(arg min

xi

f(~x), arg min∀xj ,j 6=i

f(~x)), (4.2)

which means that there does not exist any other decision variable interacting with xi.Based on the definition of separability, a problem f(~x) is known as fully separable iff:

arg min~x

f(~x)=(arg min

x1

f(x1, . . . ), . . . , arg minxD

f(. . . , xD)), (4.3)

where there does not exist any interaction between any pair of decision variables in ~x. Bycontrast, a problem f(~x) is known as fully nonseparable if every pair of decision variablesinteract with each other.

However, if only part of the decision variables are separable while the others are non-separable, the problem is known as partially nonseparable:

arg min~x

f(~x)=(arg min

~x1

f(~x1, . . . ), . . . , arg min~xm

f(. . . , ~xm)), (4.4)

16

Page 22: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

where ~x1, . . . , ~xm are disjoint sub-vectors of ~x, and 2 ≤ m ≤ D. A problem is partially

additively separable if:

f(~x) =m∑i=1

fi(~xi) , (4.5)

where ~xi are mutually exclusive decision vectors of fi, andm is the number of independentcomponents.

In this the following sections, we briefly illustrate how to deploy the two fundamentalAI paradigms, i.e., deep learning [1] and evolutionary computation [6], in our proposed TCsystem.

4.1 Decentralized Learning

As illustrated in Fig. 4.1, a deep neural network (DNN) refers to an artificial neural network(ANN) with multiple hidden layers between the input and output layers [17]. Similar toshallow ANNs, DNNs are used for pattern recognition by modeling non-linear relations,where the modeling process using DNN is known as deep learning.

Decentralization

Figure 4.1: Decentralized deep learning.

The training target of a DNN is to learn parameters θ (i.e. the connections between the

17

Page 23: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

nodes as in Fig. 4.1) by minimizing the error function:

E(X, θ) =1

2N

N∑i=1

(yi − yi)2, (4.6)

where yi and yi are the target value and computed value of the output on a given input xi,and N is the size of training set. To minimize the error function, the most commonly usedlearning algorithm is the Backpropagation (BP) algorithm, which aims to iteratively updatethe weights according to:

θt+1 = θt − α∂E(X, θt)

∂θ, (4.7)

where α is known as the learning rate and θt denotes parameters of neural network initeration t in gradient descent.

In a DNN, since each layer can contain a large number of weight connections and theupdate of each weight in each layer is independent, the BP algorithm needs to be run formany times in each iteration. Using our proposed TC system, the task of weight updatingcan be naturally decomposed into a number of subtasks by running the BP algorithm oneach task processor in parallel.

4.2 Decentralized Evolution

As an important branch of AI, evolutionary computation is a family of algorithms inspiredby natural evolution. Different from deep learning which is focused on pattern recognition,the focus of evolutionary computation is on optimization. Technically, evolutionary com-

Evaluation

ReproductionSelection

Decentralization

Figure 4.2: Decentralized evolutionary computation.

18

Page 24: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

putation is a type of population-based stochastic search methods (known as evolutionaryalgorithms) performing iterative trial-and-errors. As shown in Fig. 4.2, an evolutionary al-gorithm iteratively updates a population of candidate solutions by three main operations:evaluation, reproduction, and selection. In each generation, the new population is producedvia reproduction (i.e. crossover or mutation), and then selected according to the fitness val-ues obtained by evaluation. This process simulates the natural evolution. As a result, thepopulation will gradually evolve to increase in fitness to the given objective function, thusrealizing the purpose of optimization.

In evolutionary computation, the most time consuming component is the evaluation,which needs to calculate the fitness value of each candidate solution by substituting it in-to the given objective function. In practice, running an evolutionary algorithm can costthousands or even tens of thousands of evaluations, while each evaluation can be very timeconsuming, taking up to hours to calculate. In this case, our proposed TC system could eas-ily deal with this task by allocating the evaluations as subtasks to different task processorsin the network, thus saving large amount of computational time.

19

Page 25: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

Appendix A

People

A.1 Team Members

Dr. Ran Cheng

Ran Cheng is a research fellow at the School of Computer Science, University of Birming-ham. Previously, he was a visiting scholar at the Honda Research Institute Europe (HRI-EU). Dr. Cheng received PhD degree in computer science from the University of Surrey,United Kingdom. His main research interests cover various areas of artificial intelligence,including evolutionary computation, machine learning, swarm intelligence, etc.

Dr. Cheng has published over 30 papers, mostly in top AI journals such as IEEE Trans-actions series and major IEEE conference proceedings. He is the founding chair of severalIEEE sponsored events and the Editorial Board Member of two international journals. Dr.Cheng is the recipient of some prestigious academic awards, such as the Association ofBritish Turkish Academics (ABTA) Doctoral Researcher Award, the University of SurreyVice-Chancellors Award, and the IEEE Transactions on Evolutionary Computation Out-standing Paper Award.

Dr. Nabi Omidvar

Nabi Omidvar received the first bachelors (First Class Hons.) degree in computer science,the second bachelors degree in applied mathematics, and the Ph.D. degree in computerscience from RMIT University, Melbourne, VIC, Australia, in 2010, 2014, and 2016, re-spectively. He is a Research Fellow in Evolutionary Computation with the School of Com-puter Science, University of Birmingham, Birmingham, U.K. His current research interestsinclude large-scale global optimization, decomposition methods for optimization, and mul-tiobjective optimization.

20

Page 26: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

Dr. Omidvar was a recipient of the IEEE Transaction on Evolutionary ComputationOutstanding Paper Award for his research on large-scale global optimization in 2017, theAustralian Postgraduate Award in 2010, and the Best Computer Science Honours ThesisAward from the School of Computer Science and IT, RMIT University. He is a member ofIEEE Taskforce on Large-Scale Global Optimization.

Dr. Lijuan Su

Lijuan Su received her Ph.D. degree from Zhejiang University, Hangzhou, China, in 2017.She was a research assistant at Harvard medical school & Massachusetts General Hospital,and a research assistant in the computational neuroscience lab at University of Arizona. Shewas one of the chairs of UV2016 and UV2018, a conference held by MIT Universal VillageProgram. And she was also a member of Brain Sciences Foundation, a nonprofit organiza-tion with aims to improve methods and tools for diagnosis to treat neurological conditions.Her Main research interests include artificial intelligence in healthcare, machine learning,cognitive computing and brain-computer interface.

Dr. Joseph Chrol-Cannon

Joseph Chrol-Cannon is a research fellow in computational neuroscience at the Universityof Surrey, working within the European Human Brain Project initiative, currently investi-gating the application of biologically plausible plasticity to auditory sensory-motor learn-ing. He received his PhD from the University of Surrey in 2015 during which he appliedspiking recurrent neural networks adapted with unsupervised synaptic plasticity to a varietyof time-series data including video and speech. Dr Chrol-Cannons research has combinedunsupervised associative learning with linear regression models, analysed the informationcontent of neural activity patterns and forgetting in synaptic interference. From 2016 to2017, Dr Chrol-Cannon worked as a contributing researcher and developer at Xenesis forthe application of machine learning methods in financial prediction and trading systems.

Dr. Yifang Han

Yifan Han obtained the Ph.D. degree from the University of Oxford, and Master degreefrom the Peking University. He has years of research experience in Blockchain, Internetof things, distributed system architecture, consensus algorithms, and distributed database.Previously, he served with a large central government R & D department and participate inthe national project (classified) security program design. He also has industrial experiencesof serving as an information security development engineer, security architect assistant,

21

Page 27: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

etc. He is an Oracle Certified Database Expert, MongoDb Certified Database Expert, UKNCSC Certified Security Professional.

Dr. Miqing Li

Miqing Li is a research fellow with the University of Birmingham, UK. He received thePh.D. degree in computer science from Brunel University London, UK. His main researchinterests are multi-objective optimisation, evolutionary computation, decision making, andsoftware engineering, on which he has published over 40 research papers.

A.2 Advisors

Prof. Yaochu Jin

Yaochu Jin received the B.Sc., M.Sc., and Ph.D. degrees from Zhejiang University, Hangzhou,China, in 1988, 1991, and 1996 respectively, and the Dr.-Ing. degree from Ruhr UniversityBochum, Germany, in 2001.

He is a Professor of Computational Intelligence, Department of Computing, Universityof Surrey, Guildford, U.K., where he heads the Nature Inspired Computing and Engineer-ing Group. He is also a Finland Distinguished Professor funded by the Finnish Agencyfor Innovation (Tekes). His science-driven research interests lie in the interdisciplinary ar-eas that bridge the gap between computational intelligence, computational neuroscience,and computational systems biology. He is also particularly interested in nature-inspired,real-world driven problem-solving. Dr Jin has (co)edited five books and three conferenceproceedings, authored a monograph, and (co)authored over 200 peer-reviewed journal andconference papers. He has been granted eight US, EU and Japan patents. His current re-search is funded by EC FP7, UK EPSRC and industry. He has delivered 20 invited keynotespeeches at international conferences. He is an Associate Editor or Editorial Board Mem-ber of the IEEE TRANSACTIONS ON CYBERNETICS, IEEE TRANSACTIONS ONNANOBIOSCIENCE, IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE, Evolu-tionary Computation, BioSystems, the International Journal of Fuzzy Systems, Soft Com-puting, Memetic Computing, and Natural Computing.

Dr Jin is an IEEE Distinguished Lecturer and Vice President for Technical Activities ofthe IEEE Computational Intelligence Society. He was the recipient of the Best Paper Awardof the 2010 IEEE Symposium on Computational Intelligence in Bioinformatics and Com-putational Biology and the 2014 IEEE Computational Intelligence Magazine OutstandingPaper Award. He is a Fellow of British Computer Society and Senior Member of IEEE.

22

Page 28: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

Prof. Newton Howard

Newton Howard pursued his interests in his studies and in 2000 while a graduate memberof the Department of Mathematical Sciences at the University of Oxford, he proposedthe Theory of Intention Awareness (IA). In 2002, he received a second doctoral degreein cognitive informatics and mathematics from the prestigious La Sorbonne in France. In2007 he was awarded the habilitation a diriger des recherches (HDR) for his leading workon the Physics of Cognition (PoC) and its applications to complex medical, economical,and security equilibriums.

In 2014, he received his doctorate of philosophy from the University of Oxford for hiswork on neurodegenerative diseases, specifically his “Brain Code” Theorem. His workhas made a significant impact on the design of command and control systems as well asinformation exchange systems used at tactical, operational and strategic levels. As thecreator of IA, Dr. Howard was able to develop operational systems for military and lawenforcement projects. These utilize an intent-centric approach to inform decision-makingand ensure secure information sharing.

His work has brought him into various academic and government projects of significantmagnitude, which focus on science and the technological transfer to industry. While Prof.Howards career formed in military scientific research, in 2002 he founded the Center forAdvanced Defense Studies (CADS) a leading Washington, D.C, national security group.Currently, Dr. Howard serves as the Director of the Board. He also is a national securityadvisor to several U.S. Government organizations.

Prof. Howards several years of working on systems design and dynamic systems analy-sis in military applications, as well as his personal research experiences, led him to studyingthe human brain.

In 2008, Dr. Howard founded the Mind Machine Project at MIT; an interdisciplinaryinitiative to reconcile natural intelligence with machine intelligence, which led to the estab-lishment of the Brain Sciences Foundation (BSF) in 2011. That same year, he published theMood State Indicators (MSI) algorithm which models and explains the mental processesinvolved in human speech and writing to predict emotional states.

His cognitive natural-language approach to systems understanding and design has ledto building more accurate engines for modeling behavioral and cognitive feedback. Due tothis work, in 2012, Dr. Howard became the Director of the Synthetic Intelligence Lab atthe Massachusetts Institute of Technology (MIT) where he focuses on the molecular basisfor human intelligence. This could yield significant benefits and enable the progress inartificial intelligence and neuroscience as a whole.

23

Page 29: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

As Prof. Howard has begun focusing on the development of functional brain and neu-ron interfacing abilities, he particularly concentrated on theoretical mathematical modelsto represent the exchange of information inside the human brain. This work, publishedin 2012, called the Fundamental Code Unit (FCU), has proven applicable in the diagnosisand study of brain disorders and has aided in developing and implementing necessary phar-macological and therapeutic tools for physicians. He has also developed individualizedstrategies to incorporate solutions for psychiatric and brain prosthetic.

Dr. Howard presently serves as Professor of Computational Neurology and FunctionalNeurosurgery at the University of Oxford and Nuffield Department of Surgical Sciencesat John Radcliffe Hospital. Prof. Howard is also the Founder and Director of the OxfordComputational Neuroscience Lab (OxCNL).

24

Page 30: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

Appendix B

Token Allocation

In the TC system, the supply of TNP tokens totals 1 billion (1,000,000,000). Tokens willbe initially distributed as REC-20 compatible tokens, and then converted 1:1 to our nativetokens after the deployment of the stable versions of the TC system.

Figure B.1: Allocation of the 1,000,000,000 TNP tokens.

The specific allocation of the tokens are as follows:

• 30%: 300 million TNP tokens will be distributed in private sales.

• 30%: 300 million TNP tokens will be stored in the Turpingplan Foundation.

• 15%: 150 million TNP tokens will be used for community building.

• 10%: 100 million TNP tokens will be used in charity activities.

• 10%: 100 million TNP tokens will be allocated to the team.

• 5%: 50 million TNP tokens will be allocated to the advisors and early contributors.

25

Page 31: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

Appendix C

Market Potential

Figure C.1: The worldwide AI market from 2016 to 2025.

As shown in Fig. C.1, the size of the artificial intelligence market will show a sharpincrease in the following one decade. In 2018, the global AI market is expected to beworth approximately 4.07 billion U.S. dollars. For example, some current major uses ofAI include image recognition, object identification, detection, and classification, as well asautomated geophysical feature detection. The largest proportion of revenues come fromthe AI for enterprise applications market, where our proposed TC system can bring a greatpotential into play. Some typical industrial application domains are listed as follows.

26

Page 32: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

C.1 Computer Vision and Pattern Recognition

C.1.1 Pixel Restoration for CSI

Figure C.2: 8x8 pixel photos were iput into a Deep Learning network which tried to guesswhat the original face looked like.

In crime scene investigation (CSI), the investigators often zoom into videos beyondthe resolution of the actual video. This remains challenging until the emergence of deeplearning. In 2017, Google Brain researchers trained a Deep Learning network to take verylow-resolution images of faces and predict what each face most likely looks like. Theycall the method Pixel Recursive Super Resolution which enhances the resolution of photossignificantly. Fig. C.2 shows the original 8x8 photos, the ground truth (which was the real

27

Page 33: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

face originally in the photos) and in the middle the guess of the computer. Despite it isnot perfect, it is pretty unbelievable that the computer can guesstimate so well many of thefeatures of the person in the photo.

C.1.2 Fashion Designs

Figure C.3: Draw the outline of a bag or a shoe and the Deep Learning network will colorit for you. The results are pretty creative.

The same idea as in“Let there be color” can be used to for a Deep Learning network tocreate other types of new images. In Pix2Pix, Isola et al taught a Deep Learning networkto perform multiple tasks: create real street scenes from colored blobs, create a map froma real aerial photo, turn day scenes into the night and fill out the colors between edges ofobjects. The example in Fig. C.3 demonstrate that in many cases, the computer gets prettycreative about the designs of the objects.

C.1.3 Machine Translation

Figure C.4: A photo of a packaged food taken by the phone, and Google Translate “reads”the text and replaces it with a text in English in real-time.

28

Page 34: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

Google Translate app can now automatically translate images with text in real-time toa language of your choice. Just hold the camera on top of the object and your phone runs adeep learning network to read the image, OCR it (i.e. convert it to text) and then translateit. Languages will gradually stop being a barrier and we will be able to communicate withother humans universally.

C.2 Computer Games, Robots and Self-driving

C.2.1 AI Game Engines

The Deep Learning community is currently in a race to train computers to beat people atalmost any game you can think of, including: Space Invaders, Doom, Pong, Gathering, anda dozen of other games. In the majority of these games, Deep Learning networks alreadyoutperform experienced players. The computers were not programmed to play the games,instead, they just played the games for a few hours and learned the rules by themselves.

C.2.2 Multi-functional Robotics

Deep Learning is also heavily used in robotics these days. As shown in Fig. C.6, the robotsreact to people pushing them around, they also get up when falling, and can even take careof pretty elaborate tasks that require gentle and care, like unloading a dish washer.

C.2.3 Self-Driving

A self-driving car is the ultimate evolutionary goal of developing ADASes - AdvancedDriver Assistance Systems, to the point when there is nobody to assist anymore. Hence,without additional insight, we can predict the future of self-driving cars tech by looking athow the most advanced ADASes are implemented now.

The importance of deep learning for autonomous driving systems can be illustrated bythe fact that Nvidia maintains long-term relationships with car manufacturers, working onembedded and real-time operating systems designed exactly for these purposes.

C.3 Trip, Traffic and Shipment Routing

AI algorithms can be used to plan the most efficient routes and schedules for travel planners,traffic routers and even shipping companies. The shortest routes for traveling. The timingto avoid traffic tie-ups and rush hours. Most efficient use of transport for shipping, evento including pickup loads and deliveries along the way. The program can be modeling all

29

Page 35: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

Figure C.5: A list of games played by machines. In most games Deep Learning networkstrains well enough to play better than a human player.

30

Page 36: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

Figure C.6: Robotics designed by the Boston Dynamics.

Figure C.7: Self-driving system of Tesla.

31

Page 37: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

this in the background while the human agents do other things, improving productivity aswell. Chances are increasing steadily that when you get that trip plan packet from the travelagency, AI contributed more to it than the agent did.

C.4 Finance and Investment Strategies

In the current unprecedented world economic meltdown, one might legitimately wonder ifsome of those Wall Street gamblers made use of AI-assisted computer modeling of financeand investment strategies to funnel the worlds accumulated wealth into what can best bedescribed as dot-dollar black holes. It is possible that a newer generation of AI-assistedfinancial forecasting would have avoided the black holes and returned something otherthan bad debts the taxpayers get to repay.

C.5 Automotive Design

Figure C.8: AI-assisted car shape design.

Using AI (particular evolutionary computation) algorithms to both design compositematerials and aerodynamic shapes for race cars and regular means of transportation (in-

32

Page 38: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

cluding aviation) can return combinations of best materials and best engineering to providefaster, lighter, more fuel efficient and safer vehicles for all the things we use vehicles for.Rather than spending years in laboratories working with polymers, wind tunnels and balsawood shapes, the processes can be done much quicker and more efficiently by comput-er modeling using AI searches to return a range of options human designers can then puttogether however they please.

C.6 Engineering Design

Figure C.9: AI-assisted truss design.

Getting the most out of a range of materials to optimize the structural and operationaldesign of buildings, factories, machines, etc. is a rapidly expanding application of AI.These are being created for such uses as optimizing the design of heat exchangers, robotgripping arms, satellite booms, building trusses, flywheels, turbines, and just about anyother computer-assisted engineering design application. There is work to combine AI al-gorithms optimizing particular aspects of engineering problems to work together, and someof these can not only solve design problems, but also project them forward to analyze weak-nesses and possible point failures in the future so these can be avoided.

33

Page 39: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

Bibliography

[1] Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. Deep learning. Nature,521(7553):436–444, 2015.

[2] David Silver, Aja Huang, Chris J Maddison, Arthur Guez, Laurent Sifre, GeorgeVan Den Driessche, Julian Schrittwieser, Ioannis Antonoglou, Veda Panneershelvam,Marc Lanctot, et al. Mastering the game of go with deep neural networks and treesearch. Nature, 529(7587):484–489, 2016.

[3] UBER Engineer. Deep Neuroevolution. https://eng.uber.com/deep-neuroevolution/,2017.

[4] Open AI. Nonlinear Computation in Deep Linear Networks.https://blog.openai.com/nonlinear-computation-in-linear-networks/, 2017.

[5] Deep Mind. Population Based Training of Neural Networks.https://deepmind.com/blog/population-based-training-neural-networks/, 2017.

[6] David B Fogel. Evolutionary computation: toward a new philosophy of machine

intelligence, volume 1. John Wiley & Sons, 2006.

[7] Artur Kiulian. Decentralized Artificial Intelligence Is Coming: Here’s What YouNeed To Know, 2017.

[8] Satoshi Nakamoto. Bitcoin: A peer-to-peer electronic cash system. 2008.

[9] Kariappa Bheemaiah. Block chain 2.0: The renaissance of money. wired, 2015.

[10] Ben Dickson. How blockchain can create the worlds biggest supercomputer, 2017.

[11] Vitalik Buterin. https://medium.com/@vitalikbuterin/the-meaning-of-decentralization-a0c92b76a274, 2017.

[12] Gavin Wood. Ethereum: A secure decentralised generalised transaction ledger.Ethereum Project Yellow Paper, 151:1–32, 2014.

34

Page 40: To Develop a Blockchain-based Computing System …ing PCs). As a core of the proposed system, Blockchain serves as the media to connect the computing resources. In this system, AI

[13] Kyle Wang. Cryptoeconomics: Paving the Future of Blockchain Tech-nology. https://hackernoon.com/cryptoeconomics-paving-the-future-of-blockchain-technology-13b04dab971, 2017.

[14] Dr Andreas Freund. Economic incentives and blockchain security. Journal of Secu-

rities Operations & Custody, 10(1):67–76, 2018.

[15] Bitshares. Delegated Proof-of-Stake Consensus.https://bitshares.org/technology/delegated-proof-of-stake-consensus/, 2017.

[16] Stefano De Angelis, Leonardo Aniello, Roberto Baldoni, Federico Lombardi, An-drea Margheri, and Vladimiro Sassone. Pbft vs proof-of-authority: applying the captheorem to permissioned blockchain. 2017.

[17] Jurgen Schmidhuber. Deep learning in neural networks: An overview. Neural net-

works, 61:85–117, 2015.

35