Artificial Intelligence Overview

Objectives

 Upon completion of this course, you will understand:

 Basic concepts of AI
 AI technologies and development history
 Application technologies and fields of AI
 Huawei's AI development strategy
 AI development trends

1 Huawei Confidential
 Contents

1. AI Overview

2. Application Fields of AI

3. Huawei's AI Development Strategy

4. Controversies Over AI and Its Future

2 Huawei Confidential
Insights from Prominent Computer Scientists
"I propose to consider the question, 'Can machines think?'"
— Alan Turing, 1950

"...making a machine behave in ways that would be called intelligent if a human were so
behaving."
— John McCarthy et al., 1956

"Artificial intelligence is the science of making machines do things that would require intelligence
if done by men."
— Marvin Minsky

3 Huawei Confidential
What is intelligence?
 According to the theory of multiple intelligences proposed by Professor Howard Gardner,
multiple intelligences are manifested by eight capabilities:
 Linguistic-verbal intelligence
 Logical-mathematical intelligence
 Visual-spatial intelligence
 Bodily-kinesthetic intelligence
 Musical-rhythmic and harmonic intelligence
 Interpersonal intelligence
 Intrapersonal intelligence
 Naturalistic intelligence
4 Huawei Confidential
5 Huawei Confidential
What Is Artificial Intelligence?
 "Artificial" in artificial intelligence (AI) means that it is designed by and is created for humans.
 AI is a scientific discipline that studies and develops theories, techniques, and application systems used to simulate
and extend human intelligence. The term was first coined by John McCarthy in 1956, who defined it as the "science
and engineering of making intelligent machines, especially intelligent computer programs". The very premise of AI
technology is to enable machines to learn from collected data, and make human-like decisions.
 Today, AI has become an interdisciplinary course that involves various fields.

6 Huawei Confidential
Relationship Between AI, Machine Learning, and Deep Learning

7 Huawei Confidential
Relationship Between AI, Machine Learning, and Deep Learning
 AI is a scientific discipline that studies and develops theories, techniques, and
application systems used to simulate and extend human intelligence.
 Machine learning (ML) refers to the ability of computers to learn, simulate, or implement
human behavior to acquire new knowledge or skills, and continuously update existing
knowledge structures to improve performance.
 Deep learning (DL) is a research field in ML and originates from artificial neural network (NN)
studies. Multilayer perceptron (MLP) is a deep learning structure. Deep learning uses higher
level features derived from the lower level features to form a hierarchical representation, in
which it simulates the mechanisms of the human brain to interpret data, such as images, voice,
and text.
8 Huawei Confidential
Major Schools of AI - Symbolism
 Symbolicism, also called logicism, psychologism, or computerism, refers to the symbolic AI that
is derived from mathematical logic.
 It suggests the basic cognition units of humans are symbols, and human cognition is a
reasoning process based on various symbols. As humans and computers are both physical
symbol systems, computers can be used to simulate intelligent human behaviors.
 This theory, proposed by McCarthy et al in 1956, first used the term "artificial intelligence" and
made dominant contributions to the field's development, especially the success of expert
systems.

9 Huawei Confidential
Major Schools of AI - Connectionism
 Connectionism, also known as bionicsism or physiologism, argues AI originates from bionics,
that is, it is the study of human brain.
 Researchers believe the neuron, and not symbol processes, is the basic thinking unit.
Connectionism starts with neurons and studies NNs and brain models, to create a new
development path for AI.
 The McCulloch-Pitts (MP) neuron model was proposed in 1943. However, the study of brain
models was limited due to biological prototypes and technical conditions in the 1970s and 80s.
It was not until the proposal of hardware-simulated neural networks that the trend of
connectionism reemerged again.
 Nowadays, the artificial neural network (ANN) is a common technique but its complexity and
scale has brought many interpretability problems.

10 Huawei Confidential
Major Schools of AI - Actionism
 Actionism is also known as evolutionism or cyberneticsism and states that AI originates from
cybernetics.
 It suggests intelligence depends on perception and actions, and that intelligence does not
require knowledge, representation, or reasoning. AI can evolve like human intelligence, and
intelligent behavior can only be manifested in the real world by interacting with the
surrounding environment.
 Early research on actionism focused on simulating intelligent behavior of people in the control
process. It led to the birth of intelligent control and robotics in the 1980s.

11 Huawei Confidential
AI Development History (1)

12 Huawei Confidential
AI Development History (2)

13 Huawei Confidential
Types of AI
 Strong AI:
 This hypothesis aims to create intelligent machines that replicate human functions, such as reasoning
and problem-solving, and are perceptive and self-conscious. Strong AI will be able to think
independently and teach itself to solve new problems, and have its own values and worldviews, and
will even have the same instincts as creatures, such as survival and safety needs. In a sense, strong AI
can be seen as a new species.

 Weak AI:
 Weak AI aims to build intelligent machines that can perform specific tasks but rely heavily on human
interference. These machines may seem intelligent but are not self-conscious.

14 Huawei Confidential
AI Industry Ecosystem
 Data, algorithms, computing power, and application scenarios are the basic elements of AI
applications. We must combine AI with premium cloud computing, big data, and IoT to facilitate
our intelligent society.

15 Huawei Confidential
 Contents

1. AI Overview

2. Application Fields of AI

3. Huawei's AI Development Strategy

4. Controversies Over AI and Its Future

16 Huawei Confidential
Popular AI Subfields

17 Huawei Confidential
AI Technology and Application Fields

Source: Deloitte Global Artificial Intelligence Industry

Whitepaper

18 Huawei Confidential
Distribution of AI Application Technologies in China
 Computer vision, robotics, natural language processing (NLP), machine learning, and biometric
recognition are the most popular technology fields for China's AI enterprises.
 Currently, the directions of general AI
technologies are:
 Computer vision
 The study to make computers see things
fast and accurately.
 NLP
 The study making computers understand
and use natural languages.
Source: EqualOcean 2022 China AI chips Industry Research

19 Huawei Confidential
Smart Mining
Scenario 1: Mining chamber that houses important devices such as Solution for scenario 1: Automatic
power and communications devices inspection robot
Goal: Automatic inspection and unattended operation. Video surveillance: Machine vision-based video
surveillance for 24-hour monitoring of sensitive
areas
Challenge 1: The chamber Challenge 2: Manual inspection Sound detection: AI sound detection techniques
environment needs to be monitored must be performed periodically. to intelligently monitor devices
24/7.

Scenario 2
Belt tear monitoring Belt deviation monitoring

Scenario 2: Underground video surveillance system

Challenges
Coal conveyor The length of the coal conveying belt exceeds 20 km. Manual
belt inspection is time-consuming and insecure. Belt cleanliness monitoring
Multiple belts need to work together to prevent coal accumulation.
Violation Monitoring of personnel violations, such as not wearing safety
activities helmets, smoking, passing under equipment, walking in roadways,
and sitting on belts.
Violation Gas drainage, water exploration and drainage.
operations

20 Huawei Confidential
AI Safeguards Nature – A Protective Umbrella for Biodiversity in Chile
 The AI-powered "Nature Guardian" project has landed in the Nahuelbuta mountain range to
study and protect the Darwin's fox. The "Nature Guardian", an acoustic monitoring system
developed by the rainforest conservation organization Rainforest Connection (RFCx), has been
deployed in several projects and is working effectively.
 It consists of solar devices equipped with microphones and antennas. These devices collect
sound data from the surrounding environment and transmit it to the cloud through wireless
networks for AI to analyze. Each device covers 3 square kilometers and runs around the clock.
 The "Nature Guardian" monitors gunshots from poachers, trucks and saws from illegal
loggers, and cries from different animals to provide data for research.
 Trained AI analysis can identify the sounds of different animals, enabling experts to study the
distribution and behavior of specific species and better implement environmental protection
through adaptive management.
 If a threat is identified, the system sends a real-time alert to the ranger's mobile app, enabling
the ranger to respond rapidly.

Darwin's fox
Source: https://www.huawei.com/en/tech4all/stories/nature-guardians-for-
biodiversity-in-chile

21 Huawei Confidential
AI Safeguards Nature – Protecting Wildlife in Greece with a Shield of Sound

22 Huawei Confidential
NLP Application Scenarios
 The research topics of natural language processing include machine translation, text mining, and sentiment analysis.
Current NLP is technically difficult and not mature enough, and due to the highly-complex semantics, machines are
unable to achieve the human understanding through deep learning based on big data and parallel computing.
 Development: From shallow semantics understanding to automatic feature extraction and deep semantics
understanding; from single intelligence (ML) to hybrid intelligence (ML, DL, etc.).
 Application scenarios:

Sentiment analysis

Knowledge graph
23 Huawei Confidential
 Contents

1. AI Overview

2. Application Fields of AI

3. Huawei's AI Development Strategy

4. Controversies Over AI and Its Future

24 Huawei Confidential
Huawei Full-Stack All-Scenario AI Solution
Industry
Smart city, manufacturing, energy, transportation, finance, carrier, education, and more
applications

Third-party
ModelArts HiAi Service
platform

Ascend Community
Full-pipeline toolchain

O&M tools
Application MindX: Ascend application enablement
enablement
MindX DL MindX Edge ModelZoo MindX SDK
Deep learning Intelligent edge Preferred Industry SDKs
enablement enablement model library

AI framework

SmartKit
Fusion Director/
All-scenario AI framework that best manifests the computing power of Ascend AI processors

hiascend.com
MindStudio
Heterogeneous
compute Unified heterogeneous compute architecture to unleash hardware power
architecture

Device Edge Cloud

Hardware
products

Atlas Hardware

25 Huawei Confidential
Ascend AI Processors Empower Smarter, Superior Computing

Ascend 310 series Ascend 910 series

Energy-efficient AI SoC Powerful AI Processor

26 Huawei Confidential
CANN Heterogeneous Computing Architecture
Software and Hardware Synergy to Double Computing Power

CANN 1.0 CANN 2.0

CANN 3.0 >>Support for multiple AI frameworks
Doubled performance
Inference Training Inference + training, with software-hardware
device-edge-cloud synergy synergy TensorFlow Caffe

2018.09 2019.09 2020.08 2021.12 PyTorch PaddlePaddle ONNX

>> Continuous performance improvement

Industry-leading ResNet-50 training speed
59.8s Test benchmark: ResNet-50 V1.5
Model training Inference efficiency Ecosystem
model, ImageNet-1k dataset
acceleration improvement completeness
Graph compilation
High-performance
technology matches 28s
inference engine
Ascend processors
ResNet-50 ATC Suite 16s
2x 1300+ high-performance
CANN 3.0 50%
operators
better inference
BERT-Large
efficiency 800+ ModelZoo models CANN 2.0 CANN 3.0 CANN 5.0
2.5x
CANN 3.0 Comprehensive coverage

27 Huawei Confidential
MindSpore AI Computing Framework
AI computing center, manufacturing, finance, Internet, transportation, government, energy, university
scientific research, public safety, and carrier Key Features

Ease of use
Dynamic and static graph
Model/suite library Python native expression
unification Ease of use All-scenario AI
Models for popular scenarios, Diversified computing power,
Ultra-large-scale AI AI converged computing AI security and trustworthiness Python native expression, and collaborative systems, and smart
dynamic and static graph devices over edge and cloud.
Multi-scale hybrid Robustness evaluation and unification.
Automatic parallelism
computing enhancement
Visualized cluster Higher-order
Interpretability
optimization differentiation

High-performance execution High-performance execution Ultra-large-scale AI

Heterogeneous Memory Full-stack collaborative Distributed parallel training of
Graph kernel fusion On-device execution acceleration and Ascend-affinitive ultra-large trillion-parameter
acceleration overcommitment
execution engine to maximize models, with a series of pre-trained
MindSpore IR (unified intermediate representation) performance. influential models available.

Collaborative AI across device-edge-cloud

Adaptive lightweight High-performance Distributed parallel
Lite/Micro
deployment inference scheduling AI converged computing AI security and trustworthiness
X-fold acceleration of AI+scientific Secure and trustworthy training,
The hardware abstraction layer (HAL) supports Huawei-developed and third-party processors.
computing converged applications evaluation, and deployment of
for electromagnetic, models.
biopharmaceutical, and other
Ascend/Kunpeng/Kirin/GPU/CPU/... usages.

28 Huawei Confidential
ModelArts: Ascend-based Full-Stack Independence and Controllability
Independent and controllable AI technology stack
ModelArts is a one-stop AI development platform that offers tools to help developers
and data scientists of all skill sets build, train, and deploy models (cloud to edge), and
manage full-lifecycle AI workflows. The platform makes it easier to develop AI by
providing key capabilities, including data preprocessing and auto labeling, distributed
ModelArts: full-process AI enablement platform training, automated model building, and one-click workflow execution.
Algorithm
Automatic
development Pangu models (D) Industry datasets
learning
environment ModelArts Pro: Enterprise-Grade Development Platform for Specific
Distributed AI scientific Ascend built-in Industry Needs
Data labeling
debugging computing algorithms
Software
platform

Data AI asset
AI training AI inference
management marketplace
Full-stack AI

Independently-developed, controllable chip enablement toolchain

MindSpore: deep learning MindStudio: Ascend operator

computing framework development tool
CANN: Ascend chip heterogeneous computing driver

China-developed chips with independent intellectual

Hardware

property rights

Kunpeng (general computing) Ascend NPU (AI computing)

Planning and Cooling,

Civil works Electricity
Inf

Design ventilation, etc

29 Huawei Confidential
MindX: Application Enablement with Ascend
MindX: Ascend application enablement

MindX DL MindX Edge ModelZoo

MindX SDK
Deep learning Intelligent edge 800+ high-performance pre-trained models
Industry application development kit
enablement enablement 110+ reference designs

MindX SDK: Accumulated industry knowledge to simplify application development

Smart Search clustering

Video analysis SDK Scenario -specific SDKs
manufacturing SDK SDK Shorten
development time
Traditional SDK-based
application
development development
mxVision mxManufacture mxIndex OCR, customer service, voice...

Available Coming soon

30 Huawei Confidential
From AI+ to +AI: The Never-Ending Evolution of Best Practices

AI+ +AI
Exploring AI capabilities Powering core production systems with AI

Image EfficientNet model accuracy:

classification 98.7%, higher than human
Smart city Smart campus
accuracy (96%)

...
RNN-T model accuracy: 96.8%,
Speech
higher than human accuracy Smart
recognition Industrial
healthcare
(94.17%) Internet

BERT model accuracy: 90.9%,

Reading
higher than human accuracy Autonomous Smart finance
comprehension
(82%) driving

31 Huawei Confidential
 Contents

1. AI Overview

2. Application Fields of AI

3. Huawei's AI Development Strategy

4. Controversies Over AI and Its Future

▫ Controversies Over AI

▫ The Future of AI

32 Huawei Confidential
Controversies Over AI
 Every year we are gainin a clearer understanding of underlying AI concepts. However, many are
concerned about the future of AI:
 Do we still see and hear reality?
 What are the ethical issues of AI?
 Will AI take over our jobs?
 Will our privacy be violated? How to ensure privacy security?
 Is future AI controllable?
 ......

33 Huawei Confidential
Seeing Is Not Believing
 Breakthroughs in computer vision, such as photoshop and generative adversarial network
(GAN), means now we can generate fake images that indistinguishable from real photos,
impacting the credibility of image and video data.
 For example:
 Weight-loss shops use Photoshop to produce fake images that show fast weight loss.
 Lyrebird is a tool that can clone human voices based on a few minutes of recording samples. Such a
tool could be exploited for malicious purposes.
 Room pictures on rental or hotel booking platforms could be fake and generated by a GAN.

34 Huawei Confidential
Ethical Concerns of AI
 Discrimination and stigmatization: AI can be used to
impact on social justice and freedom.
 The autonomous decision-making mechanism of AI: This
mechanism endangers human autonomy and the results
cannot be traced or reproduced to provide effective
remedies for incorrect decisions.
 Defects of AI: The objectives, methods, and decisions of
AI systems cannot be interpreted.
 Damage to human dignity, minds, and bodies: Robots
are vulnerable to malicious use and can be altered to
make environments unsafe, while AI can expose people
to deception and manipulation.

35 Huawei Confidential
Will AI take over our jobs?
 Throughout the course of history, humans have sought ways to improve efficiency, that is, to obtain more
from fewer resources. Just as sharp stones improve hunting, and steam engines reduce the demand for
horses, every step towards automation has changed our society. In the AI era, what tasks will be taken
over by AI?
 AI may take over:
 Repeatable tasks
 Non-creative tasks
 Tasks that require weak social skills
 Dangerous tasks
 ......

36 Huawei Confidential
Privacy and Data Security
 In the era of big data and AI, we can obtain information with convenience and efficiency. However, while we are
obtaining convenience, our behavior is recorded, learned, and used for different purposes. How can we protect
private data?
• 1. End users need to read privacy terms before giving
consent to a mobile app to collect user information,
and limit the information available to the app.
• 2. Privacy protection laws and regulations should be
developed, to implement industry code of conduct
and strengthen supervision.
• 3. Confidential computing, model privacy protection,
federated learning, or adversarial learning, in addition
to protection frameworks, such as MindArmour, can
prevent reverse attacks and protect user privacy.

Typical application scenarios of MindArmour

37 Huawei Confidential
Issues to Be Resolved
 Are AI-generated content protected by copyright laws?
 Who will grant rights to robots?
 What rights can be granted to robots?
 ......

38 Huawei Confidential
 Contents

1. AI Overview

2. Application Fields of AI

3. Huawei's AI Development Strategy

4. Controversies Over AI and Its Future

▫ Controversies Over AI

▫ The Future of AI

39 Huawei Confidential
Policy Trends - United States (1)
 Executive Order - American Artificial Intelligence Initiative
 In February 2019, the US released the American Artificial Intelligence Initiative, which states that the US
will take various ways to improve American leadership in AI. It covers federal investment, access to
federal data and computing resources, development of AI-related regulations, employment crisis
response, and participation in formulation of international standards. It specifies the AI development
direction of the US in the future, proposes key propositions in fields such as smart healthcare and smart
city, and clearly expresses the exclusion of cross-border acquisitions of key AI technologies from
competitive countries.

Access to Participation in
Development
Federal federal data Employment international
of AI-related
investment and computing crisis response standard
regulations
resources formulation

40 Huawei Confidential
Policy Trends - United States (2)
 Main Content of American Artificial Intelligence
Initiative: Year One Annual Report:
1. Invest in AI Research and Development
2. Unleash AI Resources
3. Remove Barriers to AI Innovation
4. Train an AI-ready Workforce
5. Promote an International Environment
Supportive of American AI Innovation
6. Embrace Trustworthy AI for Government Services
and Missions

41 Huawei Confidential
Policy Trends - United States (3)
 To promote technological innovation and scientific research development in the US, the US
government passed the United States Innovation and Competition Act of 2021 in June 2021. The act
covers multiple topics including chips, 5G, aerospace, cyber security and AI, medical research, and
American manufacturing. The act proposes to reform the National Science Foundation (NSF) and
authorizes $81 billion for it to conduct research in 10 key fields including AI and computer
technology. In addition, $10 billion is authorized for technology centers and innovation research
institutes of universities for research in related fields.

42 Huawei Confidential
Policy Trends - China (1)
 In July 2017, AI was first written into the government's work
report. The New Generation of Artificial Intelligence
Development Plan issued by the State Council proposed a
three-step strategic goal:
 Step 1: By 2020, the overall technology and application of AI
will be in step with globally advanced levels.
 Step 2: By 2025, achieve major breakthroughs in basic
theories for AI, and be a world-leading player for
technologies and applications.
 Step 3: By 2030, realize world-leading status of nationally-
developed AI, technologies, and applications, ensure China is
the world's primary AI innovation center, achieve tangible
results in intelligent economy and intelligent society, and lay
an core to become a leading innovation-style nation and an
economic power.

43 Huawei Confidential
Policy Trends - China (2)
 The Outline of the 14th Five-Year Plan for Economic and Social Development and Long-Range Objectives
Through the Year 2035 of the People's Republic of China was released in March 2021, which emphasized
that China "will explore the establishment of regulatory frameworks for autonomous driving, online
medical care, financial technology, and intelligent delivery, and improve the relevant laws, regulations, and
rules for ethical reviews". The statements reflect China's priority for and determination to overhaul AI
ethics and its governance.
 The infringement on personal information is becoming more rampant, and there is an urgent need to
protect personal information. To protect the rights and interests on personal information, regulate its
processing, and guide its reasonable use, Personal Information Projection Low of People's Republic of
China was passed on August 20th, 2021 and came into force on November 1st, 2021.

44 Huawei Confidential
AI in More Industry Practices
 With further breakthroughs in AI in various vertical industries, AI technologies will be used in an
increasing number of scenarios.
 Smart transportation - highways, airports, railways, logistics, ports...
 Energy - electric power, oil and gas...
 Finance - bank, insurance, securities...
 ......

45 Huawei Confidential
 Quiz

 1. What are the AI application scenarios in these vertical fields? What technologies
(not limited to AI) are used in these scenarios?
 2. Based on the policy trends,summarize the core concerns for future AI
development of each country.

46 Huawei Confidential
 Summary

 This course introduces the definition and development history of AI, describes the
popular fields and application scenarios of AI, briefly introduces Huawei's AI
development strategy, and discusses its most pressing topics and trends.

47 Huawei Confidential
 Quiz

1. (Multiple-answer question) Which of the following are Huawei's full-stack all-scenario AI

solutions?
A. ModelArts

B. MindSpore

C. Atlas 800 Servers

D. ModelZoo

48 Huawei Confidential
 Recommendations

 Huawei Talent
 https://e.huawei.com/en/talent/#/home

 Ascend official website

 https://www.hiascend.com/

 MindSpore
 https://mindspore.cn/

 Huawei Cloud
 https://www.huaweicloud.com/
49 Huawei Confidential
Thank you. 把数字世界带入每个人、每个家庭、
每个组织，构建万物互联的智能世界。
Bring digital to every person, home, and
organization for a fully connected,
intelligent world.

Copyright©2023 Huawei Technologies Co., Ltd.

All Rights Reserved.

The information in this document may contain predictive statements

including, without limitation, statements regarding the future financial
and operating results, future product portfolio, new technology, etc.
There are a number of factors that could cause actual results and
developments to differ materially from those expressed or implied in the
predictive statements. Therefore, such information is provided for
reference purpose only and constitutes neither an offer nor an
acceptance. Huawei may change the information at any time without
notice.