Content uploaded by Ishan Gupta
Author content
All content in this area was uploaded by Ishan Gupta on May 12, 2020
Content may be subject to copyright.
Decentralization of Artificial Intelligence:
Analyzing Developments in Decentralized Learning
and Distributed AI Networks
Ishan Gupta
Student, Seth Anandram Jaipuria School, Kanpur
Abstract— While Artificial Intelligence models are traditionally
centralized at all stages (development, training, testing,
deployment etc.) throughout their lifecycle, there are various
disadvantages and problems, as discussed in this paper,
associated with such a centralized model. Convergence of various
new technologies are facilitating the development of
unconventional decentralized or distributed frameworks in
which AI models learn from decentralized data; the problem-
solving process can also be decentralized by breaking down the
problem into subproblems which are handled separately by
different AI programs specializing in them. This decentralization
has been catalyzed by developments in Blockchain technology
and Cryptography. The growth of Decentralized Artificial
Intelligence (DAI) has been fueled by various factors and has
often been termed as the “democratization” of AI. Emerging
decentralized Machine Learning (ML) frameworks, Federated
Learning and Distributed AI marketplaces are indicating the
growth of DAI. This paper closely analyzes the functioning and
growth of DAI systems and recent developments in the field. It
also discusses various challenges that arise in a decentralized
model along with numerous potential solutions.
Keywords— Decentralized Artificial Intelligence, Federated
Learning, Data Privacy, Blockchain.
I. INTRODUCTION
Distributed Artificial Intelligence or DAI can be defined as
a subfield of Artificial Intelligence which, unlike a traditional
AI model, follows a decentralized and collaborative approach
towards solving problems by breaking a complex problem into
smaller subproblems and handling each one individually.
The models that handle these “subproblems”, in a DAI
system, are called the ”agents” or “nodes” which are generally
autonomous or semi-autonomous and are located
geographically far away from each other. Furthermore, these
agents are able to communicate with each other; enabling this
collaborative approach and communication requires a specific
protocol which is developed by various Decentralized AI
networks. The characteristic feature of a DAI agent is its
autonomy and ability to communicate and function
independently.
In a decentralized model, there are various data privacy
concerns that must be resolved; this is where cryptography
(homomorphic encryption, SMPC, etc.) plays an important
role. Additionally, Blockchain and Smart Contracts are being
used to reward the developers and data scientists and enable
the monetization of individual AI solutions (without the need
to trust a central authority) available on the DAI network.
Such a decentralized model has various features that solve
the challenges posed by a traditional centralized AI system;
these features and challenges have been closely analyzed in
this paper.
The field of Decentralized AI has recently witnessed
numerous developments and can be viewed as a rapidly
evolving industry. There are various organizations, projects,
developers and researchers that are contributing to this
development; however, all of them do not follow a common
approach towards Decentralized AI leading to inconsistency
in definitions and ambiguity surrounding the implementation
and structure of these systems.
The core principle remains the same- decentralizing a
traditionally centralized method of AI development. However,
in recent years, two different approaches towards
decentralization have emerged with both focusing on
decentralized two different stages in the lifecycle of an AI
model.
1) Decentralized data acquisition, learning and training-
This approach focuses on decentralizing the process
of Deep Learning and training the AI model through
Federated or Distributed Learning. Unlike traditional
AI, instead of a large dataset compiled in a single
location, multiple and relatively smaller datasets are
used to train the AI model.
2) Decentralized model development and solution-
This approach focuses on decentralizing the process
of developing solutions and creating the AI model. In
this approach, the problem-solving process is
decentralized by collaboration of various
independent AI solutions specializing in different
tasks. It can be compared to a “divide-and-conquer”
algorithm which breaks problems into subproblems
and handles them individually. Individual results are
combined to reach a final solution.
Although both of the aforementioned approaches use
decentralization to solve different problems, the central idea
of democratizing the Artificial Intelligence ecosystem (which
presently seems to be moving towards monopolization) is
common. [1]
II. CONVENTIONAL CENTRALIZED AI MODEL AND LIFE
CYCLE
Traditionally, the life cycle of an AI Model is mostly
centralized at every stage from data acquisition and learning
to deployment.
Fig. 1 Life cycle of an AI Model
Most Data Science projects developed in the last few decades
follow this standard development pattern and life cycle.
Multiple iterations of this end-to-end cycle might be required
for creating an efficient model. In most processes involved in
this lifecycle, some degree of decentralization is possible.
III. DISADVANTAGES OF CONVENTIONAL CENTRALIZED
AI
As demonstrated in Figure 1, the development of a
decentralized AI model involves various stages. All these
stages are traditionally centralized. This centralized approach
of creating and deploying AI has resulted in numerous issues
which have been discussed below.
A. Oligopoly in the AI ecosystem
The Artificial Intelligence industry has been witnessing
monopolization in the past few years; a few “technology
giants" (primarily Google, Apple, Facebook, Microsoft and
Amazon) govern major portions of the industry and are
largely responsible for spearheading the AI movement.
Numerous upcoming ventures in the field of AI have also
been acquired by these corporations. The main reason for this
monopolization is the interdependence of various stages of AI
development. [2]
Fig. 2 Cyclic AI Development Process
As demonstrated in Fig. 2, access to huge datasets have
enabled industry leaders to develop more accurate and
successful AI models and further acquire data for training
future models; this has often been termed as the "rich get
richer" problem. The ownership of the acquired datasets is
centralized; therefore, the large tech companies are the sole
owners of these datasets. While these companies are able to
successfully train and optimize AI models using these
datasets, most new ventures lack the access to such datasets
necessary for building a successful model. In a decentralized
approach towards AI, multiple agents can contribute to the
training of AI models through Federated Learning and
Collaborative AI. [3]
It has been argued that this monopolization has been
restricting the capabilities of AI in the area of social
innovation. Large technology corporations have access to
huge datasets, resource capabilities and exceptional
professionals in data science and machine learning.
Corporations are primarily developing AI for general
commercial usage; a decentralized AI network can encourage
innovation and efficient usage of AI in resolving major social
issues and global challenges since it allows access to large
decentrally-owned datasets and most importantly, a medium
to collaborate for machine learning and model training.
B. Data privacy issue (centralized learning)
This is one of the most serious problems caused due to
traditional centralized AI models and the undemocratic AI
ecosystem. As discussed above, major tech corporations have
access to large datasets acquired through various sources.
Firstly, the ownership of data is centralized i.e. the companies
are the owners of the data acquired from users. Secondly, the
usage of data is generally not transparent i.e. a user is
generally not informed about how and where their data is
being used by the corporation. Today, data privacy is a major
concern due to various cases of data leaks witnessed in the
past few years.
Data privacy issues are particularly harmful when dealing
with sensitive data. For example, in order to develop an AI
model for hospitals which predicts diseases or deals with
particular diagnostics, patient data and medical history would
be required. However, medical history is very sensitive
information and is vulnerable to misuse. Following a
centralized AI approach, the organization creating the model
would either fail to secure enough datasets to develop accurate
models or would get access to sensitive medical records, both
of which are undesirable situations. Increasing cases of data
harvesting and data scandals are a matter of grave concern;
centralized ownership of data (in traditional AI development)
has often been termed as “unethical” and “harmful” as it puts
users’ personal data at risk. [4]
Federated Learning or Collaborative Learning aims to
tackle this problem through decentralized deep learning. The
“incentive mechanism” and a few other components
incorporated in Microsoft’s Decentralized and Collaborative
Data Acquisition
and Processing
Model Training
Using Acquired
Datasets
Model
Regularization
and
Optimization
Model
Deployment
Problem
Analysis and
Overview
Collect large
amount of data
Create models
and train using
acquired data
Deploy model,
optimize and
acquire more
data
AI on Blockchain aims to prevent “bad data” in decentralized
training.
C. Unnecessary efforts and reduced efficiency for
developers
Due to the absence of frameworks for a decentralized
collaborative AI development, traditional practices do not
facilitate the sharing of AI models or tools. Due to this,
unnecessary reduplication of AI models is required. Most AI
tools that predict results and aim to optimize solutions require
a collection of numerous models performing various specific
functions. Again, large technology corporations have access to
various such tools due to huge resources and competent teams
of professionals.
Today, almost all industries have some requirement or
application for AI services. From healthcare and finance to
education and manufacturing, all industries can benefit from
the use of AI in understanding consumer behaviours,
personalizing features, predicting market responses etc.
However, a significant percentage of businesses are unable to
efficiently use AI to benefit their operations because of two
reasons. Firstly, pre-existing models that can be purchased are
usually not optimized enough for every business to produce
desirable results. Secondly, most businesses, excluding large
technology corporations, are unable to hire competent teams
of AI professionals who are able to create customized AI
models to suit their purpose. Through emerging decentralized
AI networks, companies could actually combine various semi-
autonomous AI services to create a custom model, without a
lot of effort, which is able to fulfill the requirements.
If the AI ecosystem is democratized, such specific tools
could be easily monetized and made available through
different frameworks and networks. Furthermore, in a
decentralized AI model, the different “agents” or, in this case,
AI tools performing specific subtasks are autonomous or
semi-autonomous. These agents can communicate with each
other and get tasks performed as per the requirement. In such
networks, the use of models to get results is monetized i.e. it
requires certain “bounty” to be paid to the creator of the
model.
This problem has been specifically targeted by
decentralized AI marketplaces such as SingularityNET. In the
past few years, various research teams and developers have
been working on creating protocols, frameworks and networks
for enabling the interoperability of various AI models to work
together and for the developers of these models to be
rewarded for the usage of their services.
IV. MACHINE LEARNING USING DECENTRALIZED DATA
Traditionally, AI Models are trained using centralized data
i.e. data stored in a central location. This, of course, raises
various privacy concerns since entire datasets are accessible.
Recently, various approaches to train AI Models without the
need to compile all data in a central location have emerged.
A. Federated Learning
Federated Learning is one of the most promising techniques
aimed at decentralizing the process of AI development.
Collaborative or Federated Learning is a Machine Learning
(ML) in which AI models are locally trained on various
decentralized devices or clients holding their own local data
samples. In this approach, data samples are not exchanged or
compiled in a central server.
With the emergence of IoT (Internet of Things) and the
growth of portable technology, there are billions of devices,
which are connected to the internet, around the world. With
Federated Learning, the huge data collected by these devices
can help train very accurate and efficient AI models without
the risk of data leaks and privacy concerns.
Each round of the process of Federated Learning can be
summarized in the following five steps-
● The central server selects a statistical model to be trained
and a subset of nodes or clients for the training process.
● The central server transmits the same model to all nodes
selected (based on specific criterion) in the current round.
● The clients retrieve the model and locally train it using
the local user data on the device.
● The nodes send the updated results back to the central
server. Results from all selected clients are aggregated to
get a final updated model.
● The updates are made to the original model and the
process is repeated again with the updated model and a
new set of clients. [5]
Fig. 3 Federated Learning Process
Notable points related to the process-
● Nodes are usually selected on the basis of numerous
factors. For example, in the case of mobile devices-
battery life, connectivity, network bandwidth, available
computation power etc.
● As opposed to various ML training methods, in Federated
Learning, datasets and mostly heterogeneous i.e. they
differ in magnitude, distribution, etc.
● In a node, the resulting model might be iterated over the
local data numerous times before averaging all nodes’
results in the central server. The number of times a client
must train the model in each round before sending back
the updates depends upon various circumstantial factors.
A February 2017 study related to Federated Learning
analyzes two algorithmic approaches towards Federated
Learning using decentralized data- FedSGD and FedAvg.
SGD stands for Stochastic Gradient Descent, a common
and effective approach often used in Machine Learning
algorithms. FedSGD or FederatedSGD is a federated
variant of the SGD algorithm with a few upgrades and
additional parameters for batch size, clients etc. FedAvg
stands for Federated Averaging algorithm; it works
similar to the FedSGD but the nodes perform multiple
updates using local data, with weights instead of
gradients, before transmitting the results to the central
server for weighted averaging. [6]
● Numerous iterations of the above process with different
subsets of clients are usually performed in order to
achieve desired accuracy. Generally, after a given number
of rounds, the model stops showing improvement in
accuracy.
In this approach of Machine Learning, a user’s private data
never actually leaves their personal device. The developers are
able to train their models on the user's own device using their
data. Since no transmission of data takes place, there is no
possibility of data leaks. Additionally, the user data cannot be
reconstructed from the results since it is encrypted. Therefore,
Federated Learning tackles various problems such as data
privacy, security, handling sensitive data etc. FL models
might still have a risk of unintended memorization; this risk
can be handled using differential privacy which might come at
the cost of increased computations.
Federated Learning is also quite different from Distributed
Learning. In Distributed Learning, it is a common assumption
that the datasets are homogenous or evenly distributed;
however, this is not true for Federated Learning. Since
datasets are stored locally and the magnitude of data is
dependent upon various unique factors related to the usage,
datasets are mostly heterogeneous.
It must be noted that Federated Learning decentralizes three of
the steps show in Figure 1-
● Data acquisition and processing
● Model training using acquired datasets
● Model regularization and optimization
However, Federated Learning also has a few limitations as
listed below.
● Datasets are heterogeneous and non-uniformly
distributed. Datasets vary a lot in magnitude and range
and might have unintended effects on the model.
● Since datasets are not accessible, it is difficult to
recognize parameters decreasing the accuracy or “bad
data”.
● Models trained through Federated Learning are mostly
static i.e. changes to the model can only be provided with
regular updates.
● Specific communication mechanisms are required to
facilitate the training of models.
● Limited time is available for the models to be trained
without affecting the node device’s performance.
● Node devices must have required computational power
and specifications to execute the training process locally.
Libraries and Applications:
1) TensorFlow Federated (TFF)-
TensorFlow is a renowned open source platform for
developing Machine Learning applications. It
provides various libraries, tools and resources for
developers to develop ML-powered applications.
TensorFlow Federated (TFF) is an open-source
framework developed by TensorFlow for machine
learning on decentralized data. Besides Machine
Learning, it also allows users to perform other
decentralized computations such as aggregated
analytics. The TFF platform has two layers-
Federated Learning (FL) and Federated Core (FC).
2) Google’s Gboard-
Gboard is Google’s virtual keyboard for mobile
devices. Next-word prediction is a crucial feature of
the Gboard and a February 2019 study shows the
results of training Gboard to give more accurate next-
word predictions using Federated Learning. The
study compares results of server-based training using
SGD (Stochastic Gradient Descent) and training
using the FederatedAveraging (FedAvg) algorithm.
This is one of the first few real-world applications of
Federated Learning. [7]
CIFG (Coupled Input-Forget Gates) is a variant of
RNN (Recurrent Neural Network), a class of
Artificial Neural Networks. CIFG is specifically
optimized for mobile devices since it significantly
reduces the number of computations and parameter
set size without affecting the models performance.
The aforementioned study finally compares the
accuracy and results from Federated CIFG and
Server-based CIFG. The metric used to compare the
two is “Recall”, the ratio of the number of correct
predictions to the total number of tokens.
Furthermore, results from top-1 recall (one word
suggestion) and top-3 recall (three word suggestions)
were both considered separately. For both, the
Federated CIFG model performed better than the
server-based CIFG model. Most importantly,
federated learning models protect the user’s privacy.
B. Microsoft’s Decentralized and Collaborative AI on
Blockchain
In a July 2019 study, researchers at Microsoft proposed a
framework called “Decentralized & Collaborative AI on
Blockchain” for decentrally training and improving AI
models. Presently, this framework is operational and publicly
available as it has been open-sourced on GitHub. Using this
framework, developers can host their model using smart
contracts on a Blockchain. [8]
This framework consists of three primary components that
facilitate its functioning-
● Incentive Mechanism-
It coordinates smart contract rewards and validates
the transactions made (Ethereum Blockchain-based).
● Data Handler-
Stores data and meta-data on Blockchain and ensures
it is publicly available.
● Machine Learning Model-
It is concerned with predictions and training of
models uploaded by developers.
Unlike the Federated Learning approach, in this framework,
the process of training AI models and contributing datasets
has been incentivized for the users who wish to share data and
train models. Contributors can easily contribute to improve
the model; in order to encourage contributions, several
incentives (financial and non-financial) have been provided
for the user within the framework. One of the unique features
of this framework is the incentive mechanism to promote
submission of “good data” which works simultaneously with a
penalty mechanism to discourage submission of “bad data” by
the contributors.
Initially, this framework has been implemented using
Ethereum, an open source, Blockchain-based distributed
computing platform offering smart contract functionality, as it
provides the most optimum solution. However, it is suggested
that the framework can also use another Blockchain if a better
suited alternative is available.
The following 4 steps summarize the basic process of
model training through this framework-
1. Developers upload a lightly pretrained (very limited
accuracy) AI model on the Blockchain through a
smart contract.
2. Contributors test the model by getting predictions for
input data without any additional costs.
3. Contributors “stake” a required deposit in order to
submit a data contribution for training purposes.
4. One of the following two scenarios takes place-
4.1. If the data submitted is “good”, the deposit
is refunded with some additional incentive.
4.2. If the data submitted is “bad”, the deposit is
lost as penalization.
Using a Blockchain instead of traditional source code
hosting practices provides various advantages to this
framework. Firstly, Blockchain is a Distributed Ledger
Technology (DLT) and most importantly, it provides a
“trustless system” i.e. there is no need to trust individual
entities or a third party for any transactions, modifications,
additions, etc. Blockchain technology has been discussed in a
more detailed manner later in this paper. In the context of this
framework, Blockchain provides a transparent and reliable
method of sharing models. Customers and contributors can
therefore view the model’s smart contract due to its public
availability. Since the usefulness or quality of a data provided
might be a subjective metric in various cases, such
transparency and fairness provided by smart contracts assures
the data contributors that they would be fairly compensated
for their contributions.
The aforementioned compensation is provided through any
of the three components of the incentive mechanism-
● Gamification-
In this mechanism, a non-monetary compensation is
provided to the data contributor. It can be compared
to the badges or points awarded on various platforms.
● Reward Mechanism-
The contributor is provided some financial rewards
based on the improvement in the accuracy of the
model due to the data contributed. These financial
rewards are supplied from a pool of reward funds
provided by the development company or any entity
that wishes to encourage the training of the model.
● Deposit, Refund, Penalty-
Since a smart contract cannot legally compel a user
to pay a penalty, the deposit submitted by the
contributors before making the contribution is
actually “staked” by them. In case the data provided
is good, the deposit is returned and financial
incentive is provided.
Microsoft claims that the vision of this project is to
democratize the AI field.
Although this framework is quite progressive and it will
certainly encourage developments in the field of
Decentralized Artificial Intelligence, there are still quite a few
limitations related to this mechanism-
● It cannot be used with unsupervised models; it is
only available for supervised models i.e. models
trained on labeled datasets containing both input and
output parameters.
● It cannot be used with complicated models. It is
presently viable for only simpler models with simple
input formats.
● The incentive mechanisms do not offer a proper
source of compensation for contributors who might
not be encouraged to participate in the training
process.
V. INTEROPERABILITY BETWEEN AI MODELS IN A
DECENTRALIZED NETWORK
In Decentralized AI, the handling of problems is different
from traditional centralized AI systems because of the absence
of a central server which coordinates all tasks. In recent years,
various decentralized AI networks and frameworks have
emerged. However, in such networks, there are various
problems that arise.
The following sections discuss the architecture and
problems associated with decentralized systems.
A. Structure and Functioning
Fig. 4 below demonstrates the functioning of a centralized
multi-AI model.
Fig. 4 Centralized Multi-AI Model
In such a model, the working of all AI services is
coordinated by a central authority “X”. For example, let us
assume that Model A is a tool used for clarifying images
containing handwritten text to increase legibility; Model B is
an OCR (Optical Character Recognition) tool which reads
handwritten text and changes it to machine-encoded text;
Model C translates machine-encoded texts in French to
English and vice-versa; Model D is a text-to-speech tool
which processes machine-encoded text in English and reads it
aloud.
To better understand how a centralized model would try
and solve a multi-level problem, we can refer to Fig. 4. As
demonstrated in Fig 4, if an unclear image “i” of a
handwritten text in French would be fed as input to the central
point X and the desired output is an audio file “a” containing
the input text read aloud in English, then X would sequentially
process the image through nodes A, B, C, and D to generate
the desired output.
Fig. 5 below demonstrates the functioning of a
decentralized multi-AI model.
Fig. 5 Decentralized Multi-AI Model
In such a model, all nodes can independently interact with
each other under given protocols or communication network
guidelines. If the aforementioned text-to-speech problem is
processed in a decentralized architecture, a different approach
is observed. The input (an unclear image of a handwritten text
in French (i)) is fed to Model D. Since the input is an image
which D cannot process, it requests B to process the image
and send back a machine-encoded text. B finds that the image
is not clear and the text is not adequately legible thus B sends
the image i to A; A sends back a clarified image i’. B
processes the clarified image i’ and extracts a machine-
encoded text t from it and sends this text to D. D finds that the
text t is in French; therefore, D sends this text t to C and C
sends back the translated text t’ to D. D now receives a
machine-encoded text t’ in English; it generates the desired
output i.e. the audio file “a” containing the text read aloud in
English.
Architecture of decentralized AI systems resembles Multi-
agent systems and Decentralized Autonomous Organizations
(DAOs). The nodes or agents (Model A, Model B, etc. in Fig.
5) in a decentralized AI system generally have the following
properties-
● The nodes are independent i.e. all nodes are distinct
entities.
● The nodes are autonomous or semi-autonomous i.e.
capable of independent decisioning.
● The nodes are loosely coupled.
● The nodes might be located geographically far away
from each other.
B. Challenges in decentralization
With such a system, there is a problem that arises- finding a
suitable mechanism to incentivize developers for the use of
their models. Since models operate with each other
decentrally and problems are solved collaboratively, all
models in a system are not owned by a central entity. [9]
Therefore, in a decentralized network, there is a
requirement for the following two things-
● A fair way to compensate for the usage of models,
without the need to trust a third party to manage or
oversee transactions, through a democratic process.
● A method to make executions and transactions
related to third-party AI models transparently
available without having the need for a centralized
authority.
The most optimum solution for both these problems is
Distributed Ledger Technology (DLT) or specifically,
Blockchain and smart contracts. The specifications and details
of Blockchain technology is beyond the scope of this paper;
however, in simple terms, a Blockchain is an immutable
ledger which stores transactions in a verifiable way using
decentralization and cryptographic hashing. One of the
Blockchain-based platforms, most commonly used for
facilitating transactions in decentralized AI networks, is
Ethereum. Ethereum has its own crypto currency (Ether) and
tokens for smart contracts.
Smart contracts are self-executing contracts or computer
programs that facilitate or initiate a pre decided transaction
when given conditions are met. These smart contracts are
stored on the Blockchain for permanency and transparency.
Generally, Blockchain and smart contracts are used to host
AI models in a decentralized network; for using a model, a
user must pay a given amount or “bounty”, usually in the form
of crypto currency or tokens, to the creator of the model.
Since it is based on a Blockchain, all transactions are
immutably stored and it provides a reliable, verifiable,
decentralized and efficient way of incentivizing AI models in
a decentralized network.
Smart contracts are also used in decentralized learning
frameworks in order to provide incentives to users who
choose to train models using their own data. This method has
been implemented in projects such as OpenMined.
However, in such a case, the model must be encrypted.
Since the model is sent to the user’s personal device (in
Federated Learning), there is a risk of the model being stolen.
To prevent this, AI models can be homomorphically
encrypted before sending it to the users for training.
Homomorphic encryption (HE) allows operations on cipher
text without decryption; therefore, the model is completely
protected during the local training. Although homomorphic
encryption might seem ideal, it is still mostly theoretical, less
efficient and not optimized for application. Alternatively,
GAN cryptography or Adversarial Neural Cryptography,
introduced in 2016, and SMP (Secure Multi-Party)
computations provide more efficient approaches. SMPC is
much more practical and efficient to work with Blockchain; in
2014, a study demonstrated implementation of SMPC on
Bitcoin. [10] [11]
C. Frameworks and Organizations
In the past few years, various organizations have been
venturing into the field of Decentralized Artificial Intelligence
and Machine Learning. Most initiatives follow a common
theme or a similar vision; however, they greatly differ in
approach and features.
1) SingularityNET-
SingularityNET is a decentralized open market and
network for AIs; it is a leading initiative in the field
of Decentralized AI. It serves as both- a “commercial
launchpad” for developers and a mechanism for “AIs
to interoperate”. Transactions in SingularityNET are
based on the AGI token. Users can use this token to
purchase AI services from the marketplace.
According to SingularityNET’s Whitepaper 2.0, in
the future, AGI tokens might also be used as a basis
to provide access to voting rights for the network’s
democratic governance. [12]
SingularityNET presently uses Ethereum as the
underlying Blockchain. It provides the
SingularityNET SDK for clients who wish to use the
network’s services in their applications.
SingularityNET is certainly one of the most
ambitious and progressive projects which aim to
democratize the AI ecosystem.
2) OpenMined-
OpenMined is an open-source community which
focuses on privacy preserving and accessible AI and
provides a decentralized AI platform. OpenMined
leverages cryptography techniques such as SMPC
and HE and Blockchain technology to decentralize
machine learning. Similar to Federated Learning,
OpenMined’s decentralized ML focuses on methods
to train on sensitive and private user data without
raising privacy concerns; additionally, like
Microsoft’s Decentralized and Collaborative AI, it
aims to incentivize users to encourage them to train
AI models uploaded by developers. OpenMined
contains two primary components- PySyft library and
PyGrid platform.
3) Decentralized Artificial Intelligence Alliance
(DAIA)-
DAIA, the Decentralized AI Alliance is an alliance
of various organizations that work with AI and
Blockchain to develop decentralized applications,
networks, protocols, etc. This alliance aims to
formalize the association between the members who
share a common vision for the future of AI. It aims to
encourage collaborations amongst members,
encourage new ventures in the field of DAI and
accelerate the decentralization of AI.
VI. CONCLUSION
Decentralized AI is presently an active area of research and
application; however, almost all projects in this field are still
in their initial stage. The recent emergence and growth of
DApps (Decentralized Applications) and Distributed Ledger
Technologies (DLT) demonstrate the progress towards
distributed computing.
In the current AI paradigm, the growth of AI is largely
propelled by huge technology corporations; in view of
frequent data scandals, data leaks and data security issues,
Decentralized AI seems to provide a reliable solution to
prevent such privacy issues without hindering the progress of
AI. Contrary to common belief, besides new ventures and
emerging startups, even large technology corporations such as
Google and Microsoft are actively researching potential
methods to decentralize processes related to AI.
This paper has vividly discussed the working and
specifications of various decentralized AI systems and has
closely analyzed numerous approaches, technological
enablers, applications, frameworks, and practices related to
Decentralized AI. Relatively modern Cryptography techniques
and developments in Blockchain-based platforms have
certainly catalyzed the growth of Decentralized AI. With so
many initiatives, projects and organizations aiming to
decentralize AI and making it more accessible, Decentralized
AI seems to be a promising area of research and development
with huge potential for growth in the coming years.
REFERENCES
[1] Schneegans, Susan. (2018). Towards a monopolization of research in
artificial intelligence?.
[2] Khan, A. (2018, September 21). The Tech Oligopoly — Part 1.
Retrieved April 8, 2020, from https://blog.singularitynet.io/the-tech-
oligopoly-part-1-5d76df9d09aa
[3] Rodriguez, J. (2018, April 5). Why Decentralized AI Matters Part I:
Economics and Enablers. Retrieved April 8, 2020, from
https://medium.com/datadriveninvestor/why-decentralized-ai-matters-
part-i-economics-and-enablers-5576aeeb43d1
[4] Timan, T. &Z. Á. Mann (eds) (2019) “Data protection in the era of
artificial intelligence. Trends, existing solutions and recommendations
for privacy-preserving technologies”, October 2019. BDVA.
[5] McMahan, B., & Ramage, D. (2017, April 6). Federated Learning:
Collaborative Machine Learning without Centralized Training Data.
Retrieved April 8, 2020, from
https://ai.googleblog.com/2017/04/federated-learning-
collaborative.html
[6] McMahan, H. B., Moore, E., Ramage, D., Hampson, S., & Arcas
Blaise Aguera y. (2017). Communication-Efficient Learning of Deep
Networks from Decentralized Data. arXiv:1602.05629v3 [cs.LG].
[7] Hard, A., Rao, K., Mathews, R., Ramaswamy, S., Beaufays, F.,
Augenstein, S., Eichner, H., Kiddon, C., Ramage, D. (2019). Federated
Learning for Mobile Keyboard Prediction. arXiv:1811.03604v2
[cs.CL].
[8] Harris, J. D., & Waggoner, B. (2019). Decentralized and Collaborative
AI on Blockchain. 2019 IEEE International Conference on Blockchain
(Blockchain). doi: 10.1109/blockchain.2019.00057
[9] Rodriguez, J. (2018, April 6). Why Decentralized AI Matters Part II:
Technological Enablers. Retrieved April 8, 2020, from
https://medium.com/datadriveninvestor/why-decentralized-ai-matters-
part-ii-technological-enablers-a67e3115312e
[10] Abadi, M., & Andersen, D. G. (2016). Learning to Protect
Communications with Adversarial Neural Cryptography.
arXiv:1610.06918v1 [cs.CR].
[11] Andrychowicz, Marcin & Dziembowski, Stefan & Malinowski, Daniel
& Mazurek, Lukasz. (2014). Secure Multiparty Computations on
Bitcoin. Communications of the ACM. 59. 443-458.
10.1109/SP.2014.35.
[12] SingularityNET (2019, February). Whitepaper 2.0. Retrieved April 8,
2020, from https://public.singularitynet.io/whitepaper.pdf