PDF version of this report
You must have Adobe Acrobat reader to view, save, or print PDF files. The
reader is available for free
download.
Edge AI
Copyright 2022, Faulkner Information Services. All
Rights Reserved.
Docid: 00018008
Publication Date: 2202
Report Type: TUTORIAL
Preview
In the view of many experts, the future of artificial intelligence – and,
indeed, edge computing – is “edge AI,” in which machine learning
algorithms process data generated by edge devices locally. In many
respects, edge AI is the technology that will,
ultimately, enable enterprises to exploit the new class of intelligent
resources represented by the Internet of Things.
Report Contents:
- Executive Summary
- Concepts
- Implementation
- Concerns
- Recommendations
- References
- Web Links
- Related Reports
Executive Summary
[return to top of this
report]
In the view of many experts, the future of artificial intelligence – and,
indeed, edge computing – is “edge AI,” in which machine learning
algorithms process data generated by edge devices locally. To borrow a
communications concept, local processing removes the latency
inherent in remote processing, in which data is collected by an edge
device, transmitted to another device or cloud where it is subsequently
processed; the processed data is then returned to the originating device,
where the data is used to facilitate an action.
Related Faulkner Reports |
Edge Computing Tutorial |
Artificial Intelligence Tutorial |
Machine Learning Tutorial |
Enterprise Uses for Artificial Intelligence Market |
In edge AI, this data diversion from and to the edge device is
eliminated by utilizing AI algorithms incorporated within the edge
device or the edge system to process the data directly. This removes the
“middleman” element.
By combining data collection with smart data analysis and smart data
action, edge AI:
- Expedites critical operations, especially where speed is essential
(autonomous vehicle operation is a classic example); and - Eliminates multiple points of failure, since everything occurs
locally, or at the edge.
In many respects, edge AI is the technology (or class of technologies)
that will ultimately enable enterprises to exploit the new class of
intelligent resources represented by the Internet of Things. As Deloitte
explains, “With 41.6 billion Internet of Things (IoT) devices projected
by 20251, there is a growing need for platforms and hardware
to generate real-time analytics, often without access to centralized
data hubs.
“As computing resources become cheaper and more efficient, the [edge
AI] model can create net-new capabilities for an organization. By moving
a higher percentage of the computing power and decision-making
responsibility to the devices located close to the actual events, an
infrastructure is created where the platform improves the devices, and
the devices improve the platform.”2
Edge AI Market
As might be anticipated, the market for Edge AI hardware and software
is robust and rising. According to MarketsandMarkets:
- The global edge AI hardware market should grow from 920 million
units in 2021 to 2,080 million units by 2026; a compound annual growth
rate (CAGR) of 17.7 percent during the forecast period.3 - Correspondingly, the global edge AI software market should grow from
$590 million in 2020 to $1,835 million by 2026, a CAGR of 20.8 percent
during the forecast period.4
Concepts
[return to top of
this report]
“AI is ‘the most common
workload’ in edge computing.”
– Stephanie
Overby5
In traditional edge computing [see Figure 1], edge systems do double
duty: collecting and processing data at the source, and transmitting
that data to the cloud for higher-level processing, like processing by
artificial intelligence algorithms.
Figure 1. Conventional Edge Computing
Source: Wikimedia Commons
In contrast, edge AI brings the artificial intelligence processing to
the data, thus obviating the need for cloud involvement.
The Edge AI Foundation
Edge AI exists at the intersection of three technologies:
- Edge computing, an umbrella term for computing
conducted at the network edge. - Internet of Things (IoT), in which industrial
components like sensors are transformed into “smart machines” capable
of collecting and processing data locally, and transmitting that data
to a central data center for additional processing like trend or
pattern analysis. - Artificial intelligence (AI) or, more specifically,
machine learning (ML), which enables a system to
enhance its awareness and capabilities – that is, to learn – without
being explicitly programmed to do so.
Edge AI accelerates the trend toward decentralized computing, helping
re-purpose “the cloud” from a data warehouse and AI engine to primarily
a data storage and application delivery medium.
How Edge AI Works
Edge AI works something like this. Consider a hypothetical piece of
manufacturing equipment loaded with sensors (IoT elements) that collect
and report performance data to a hardware monitor (an edge AI device).
This monitor analyzes the data for trends and anomalies, decides if any
actions (including remedial actions) are required, and then, when
necessary, initiates those actions.
Integrated within the manufacturing unit itself or existing alongside
it, the edge AI monitor assumes the oversight role previously performed
by a human operator or, more likely, performs a new, value-added
function that contributes to equipment sustainability.
Major Benefits
Edge AI offers a variety of advantages, principal among these are:
Responsiveness – Edge AI removes the cloud middleman.
Reduced cloud traffic means reduced latency and reduced bandwidth.
Cloud Security – Data streamed to the cloud are
subject to interception and manipulation by cyber criminals. Local
processing, however, decreases that potential.
Privacy – Personal data accumulated by edge AI devices
are no longer stored in the cloud.
Control – Edge data processing policies and practices
are regulated on-premise by edge AI owners and operators.
Implementation
[return to top of this
report]
The applications for edge AI are numerous and varied. According to
Samsung, use cases include self-driving vehicles, surveillance and
monitoring, industry IoT, and image and audio analytics.
Self-Driving Vehicles
“For self-driving vehicles [see Figure 2] to be able to respond to
developments on the road in real-time while reacting swiftly and
appropriately to traffic signs, pedestrians and other vehicles, the
ability to process data very rapidly is imperative.”
Figure 2. Edge AI-powered Self-Driving Vehicle
Source: Electric Motor Engineering
Surveillance and Monitoring
“In the past, security cameras would transmit unaltered video signals
directly to the cloud, resulting in heavy bandwidth use and overburdened
servers. Now, machine learning built into cameras [see Figure 3] can
monitor activity, and only transmit events of note to the cloud.”
Figure 3. Montavue AI Smart Motion Detection Surveillance System
Source: Montavue
Industry IoT
“In industries that utilize automation, edge AI will improve safety and
reduce costs. Locally performed AI will monitor machinery for potential
defects and respond to them in real-time, with local deep learning able to
contribute to data collection too.”
Image and Audio Analytics
“The image and audio analytics-related applications of edge AI are vast.
From real-time image and scene recognition to devices responding to audio
triggers, there is a wide range of latency-critical applications [to
which] edge AI can be applied.”6
Healthcare Analytics
In an exciting development, analyst Kishan Bhoopalam predicts that
“[edge] AI will enable autonomous monitoring of hospital rooms, which will
reduce the number of medical errors. By using computer vision and other
sensors, an edge device will be able to monitor rooms and detect potential
issues. Hospital staff could then be notified of the issue.”7
Automated Optical Inspection
A key element of manufacturing quality assurance, Edge AI-enabled visual
analysis allows rapid detection of faulty parts on production assembly
lines, reducing product defects and product replacement costs, and
ushering in the era of the truly-smart factory.8
Amazon Web Services Snowball Edge Compute Optimized
To help facilitate the introduction and use of edge AI technology, Amazon
Web Services (AWS) offers Snowball Edge Compute Optimized, a
family of devices [see Figure 4] providing 52 vCPUs, block and object
storage, and an optional GPU for use cases like advanced machine learning
and full motion video analysis.
Figure 4. AWS Snowball Edge Appliance
Source: Amazon AWS
According to the vendor, customers “can deploy and run machine learning
models, such as document classification and image labeling, directly on
the device” providing the opportunity to “tune processes, improve
efficiency and productivity, and even anticipate model failures.”
Microsoft Azure Percept
Microsoft’s contribution to the edge AI space is Azure Percept, a
family of hardware, software, and services offerings designed to
accelerate business transformation through AI at the edge. According
to the vendor, Percept components include:
- A development kit supporting a wide variety of prototyping scenarios
for device builders, solution builders, and customers. - Services and workflows that expedite edge AI model and solution
development. - “[The] ecosystem of hardware developers with patterns and best
practices for developing edge AI hardware that can be integrated easily
with Azure AI and IoT services.”
Concerns
[return to top of this
report]
Although edge AI offers advantages over conventional edge computing, some
analysts are worried about issues like AI itself, inadequate oversight,
data loss potential, and more.
AI Itself
The first and most obvious objection to edge AI is the machine learning
component itself. ML systems don’t deduce the truth as humans do; rather
they forecast the truth based on the available data.9 Depending
on the application, machine-made mistakes can be serious, even deadly, and
delegating decisions to edge AI systems can induce legal liability.
Realizing their exposure, Gartner predicts that by 2024, 60 percent of AI
providers will include harm/misuse mitigation as a part of their software.10
Lax Oversight
As edge devices achieve greater autonomy through edge AI, they present a
challenge to IT and security governance. Consider, for example, that when
taking tests students are frequently admonished to “show your work,” as a
way of proving that their answers were formulated through logical means.
Unfortunately, one of the troubling aspects of machine learning – and AI
in general – is often the inability to conduct critical human oversight,
particularly as the process of ML decision-making becomes more
sophisticated, and less grounded in traditional data analysis techniques.
The rationale for ML conclusions may, in some cases, be unexplained and
unexplainable, creating a potential crisis in confidence.11 The
coming proliferation in edge AI devices will certainly exacerbate the
problem.
Data Loss
The ability of edge AI devices to gather and process data locally (or
“on-site”) reduces the amount – and, potentially, the quality – of data
transmitted to the cloud or other central repository. This may diminish
the ability of enterprise owners to conduct trend and other analyses aimed
at edge AI optimization.
Edge Security
While edge AI technology tends to increase cloud security by minimizing
edge-cloud interactions, it tends to decrease edge security since that’s
where the data processing action is. MarketsandMarkets observes that
“[the] rise of [cyber attacks, including] distributed denial of service
(DDoS) [attacks on] routers, base stations, and switches, are restricting
the adoption of edge AI solutions.”12 Analyst Kashyap Vyas
shares the concern, adding that “locally pooled data demands security for
more locations. These increased physical data points make an Edge AI
infrastructure vulnerable to various cyber attacks.”13
Size Matters
As edge AI devices become more sophisticated, assume greater
responsibilities, increase in numbers, and begin clustering at select edge
points, the twin problem of device size and power utilization will
materialize, as edge AI devices, especially local collections of edge AI
devices, begin to resemble micro-data centers or mini-clouds.14
Recommendations
[return to top of this
report]
By 2027, machine learning (ML) in the form of deep
learning (DL) will be included in over 65 percent of edge use cases, up
from less than 10 percent in 2021.
– Gartner15
The COVID-19 Imperative
Owing to the coronavirus pandemic, many manufacturing firms are adopting
edge AI software to help automate their operational workflows. Even as
COVID-19 restrictions ease, firms in manufacturing and other sectors
should pursue edge AI solutions as part of their overall workplace
automation program.16
On-Premise Vs. Cloud Redux
The availability of edge AI solutions is renewing the on-premise versus
cloud debate. Enterprise officials should develop an Edge AI Deployment
Plan, dictating where and when edge AI solutions should be deployed.
Importantly, edge AI is not a cloud replacement.
Concerning the pace of edge AI implementation, analyst Stephanie Overby
cautions that “[if] you haven’t already implemented an edge solution, you
can’t leapfrog to edge AI.” First comes an architecture that encompasses
edge and cloud computing. Then comes the edge AI element.17
Edge AI Ethics
Edge AI is, of course, AI, with all the associated dilemmas of machines
making decisions that were previously the province of humans. Enterprise
officials should establish an AI Ethics Board to consider the implications
of possible edge AI actions, and restrict or curtail any actions
potentially harmful to people or property. Involve the Legal and Human
Resources departments in evaluating any damage mitigation options.18
Data Is Everything
Modern artificial intelligence relies on probability and prediction. As
such, it requires more data than human-based intelligence. Prior to
launching an edge (or other) AI project, enterprise officials should
ensure that their:
- Data management practices are sound;
- Data management repositories are sufficiently sized;
- Data security is robust and reliable; and
- AI training data are substantial in volume, varied in nature, and
unbiased in composition.
Until these requirements are met, subject matter experts can substitute
business logic for machine learning.19
Edge AI Cybersecurity Staffing
Edge AI has opened up a new front in the never-ending battle to secure
enterprise information and information systems from hackers and other
cyber criminals. To help prepare for present and future edge AI
deployments, enterprise officials should be actively engaged in recruiting
cybersecurity analysts, hopefully with edge, if not edge AI, experience.
References
[return to top of this
report]
1 IDC.
2 “Considerations for Distributing Artificial Intelligence
Workloads to the Intelligent Edge.” Deloitte Development LLC. 2020:1-2.
3 “Edge AI Hardware Market with COVID-19 Impact Analysis
Device, Processor (CPU, GPU, and ASICs), End User, Function (Training and
Inference), Power (Less Than 1W, 1-3 W, 3-5 W, 5-10W and More Than 10W)
and Region – Global Forecast to 2026.” MarketsandMarkets. August 2021.
4 “Edge AI Software Market by Component (Solutions and
Services), Data Source, Application (Autonomous Vehicles, Access
Management, Video Surveillance, Remote Monitoring and Predictive
Maintenance, and Telemetry), Vertical, and Region – Global Forecast to
2026.” MarketsandMarkets. February 2021.
5 Stephanie Overby. “Edge Computing and AI: 7 Things to Know.”
Enterprisers Project | Red Hat, Inc. May 5, 2020.
6 “On the Edge – How Edge AI Is Reshaping the Future.”
Samsung. April 16, 2020.
7 Kishan Bhoopalam. “5 Ways Edge IoT and Edge AI Will Disrupt
Healthcare in 2020.” ClearObject. 2020.
8 Kashyap Vyas. “Edge AI: The Future of Artificial
Intelligence and Edge Computing.” IT Business Edge | TechnologyAdvice.
August 25, 2021.
9 Nick Heath. “What Is Machine Learning? Everything You Need
to Know.” ZDNet. May 14, 2018.
10 Brandon Vigliarolo. “Gartner: The Future of AI Is Not as
Rosy as Some Might Think.” ZDNet. January 25, 2021.
11 Joe McKendrick. “What’s Inside the ‘Black Box’ of Machine
Learning?” RTInsights. 2016.
12 “Edge AI Software Market by Component (Solutions and
Services), Data Source, Application (Autonomous Vehicles, Access
Management, Video Surveillance, Remote Monitoring and Predictive
Maintenance, and Telemetry), Vertical, and Region – Global Forecast to
2026.” MarketsandMarkets. February 2021.
13 Kashyap Vyas. “Edge AI: The Future of Artificial
Intelligence and Edge Computing.” IT Business Edge | TechnologyAdvice.
August 25, 2021.
14 “Edge AI Hardware Market with COVID-19 Impact Analysis
Device, Processor (CPU, GPU, and ASICs), End User, Function (Training and
Inference), Power (Less Than 1W, 1-3 W, 3-5 W, 5-10W and More Than 10W)
and Region – Global Forecast to 2026.” MarketsandMarkets. August 2021.
15 “The Future of Edge Computing Is Changing.” Channel Futures
| Informa PLC. November 22, 2021.
16 “On the Edge – How Edge AI Is Reshaping the Future.”
Samsung. April 16, 2020.
17 Stephanie Overby. “Edge Computing and AI: 7 Things to
Know.” Enterprisers Project | Red Hat, Inc. May 5, 2020.
18 Brandon Vigliarolo. “Gartner: The Future of AI Is Not as
Rosy as Some Might Think.” ZDNet. January 25, 2021.
19 Stephanie Overby. “Edge Computing and AI: 7 Things to
Know.” Enterprisers Project | Red Hat, Inc. May 5, 2020.
Web Links
[return to top of this
report]
- Amazon Web Services: http://aws.amazon.com/
- Continuity Central: http://www.continuitycentral.com/
- International Organization for Standardization: http://www.iso.org/
- Microsoft: http://www.microsoft.com/
- SANS Institute: http://www.sans.org/
- US National Institute of Standards and Technology: http://www.nist.gov/
About the Author
[return to top of this
report]
James G. Barr is a leading business continuity analyst
and business writer with more than 30 years’ IT experience. A member of
“Who’s Who in Finance and Industry,” Mr. Barr has designed, developed, and
deployed business continuity plans for a number of Fortune 500 firms. He
is the author of several books, including How to Succeed in Business
BY Really Trying, a member of Faulkner’s Advisory Panel, and a
senior editor for Faulkner’s Security Management Practices.
Mr. Barr can be reached via e-mail at jgbarr@faulkner.com.
[return to top of this
report]