I

Artificial Intelligence and

OMG Standards
Document omg/19-08-01
26 August 2019
© 2019 Object Management Group 1
Claude R. Baudoin, cébé IT & Knowledge Management
Larry L. Johnson, Technical Director, OMG
Pete Rivett, CTO, Adaptive Inc.
Copyright Notice
© 2019 Object Management Group. All rights reserved. You may download, store, display
on your computer, view, print, and link to the Artificial Intelligence and OMG Standards white
paper subject to the following: (a) the white paper may be used solely for your personal,
informational, non-commercial use; (b) the white paper may not be modified or altered in
any way; (c) the white paper may not be redistributed; and (d) the trademark, copyright or
other notices may not be removed. You may quote portions of the white paper as
permitted by the Fair Use provisions of the United States Copyright Act, provided that you
attribute the portions to the Object Management Group White Paper on Artificial
Intelligence and OMG Standards (2019).

Acknowledgements
The authors would like to thank the following OMG members and supporters for their
expertise on the various standardization efforts discussed in this paper. In particular, we
extend our sincerest gratitude to Sridhar Iyengar, IBM; Uwe Kaufmann, ModelAlchemy;
Elisa F. Kendall, Thematix Partners LLC; Dr. Said Tabet, DELL; Bobbin Teegarden, OntoAge;
and Ron Zahavi, Microsoft.

About OMG
The Object Management Group® (OMG®) is an international, open membership, not-for-
profit technology standards consortium with representation from government, industry
and academia. OMG Task Forces develop enterprise integration standards for a wide range
of technologies and an even wider range of industries. OMG's modeling standards enable
powerful visual design, execution and maintenance of software and other processes. Visit
www.omg.org for more information.

© 2019 Object Management Group 2

 Table of Contents

1 Introduction and Executive Summary ...................................................................................... 4

2 The New Scope of AI.................................................................................................................... 5

3 Why Standards in AI?................................................................................................................... 6

3.1 Phases of Standards Adoption........................................................................................... 6

3.2 The Increasing Maturity of AI ............................................................................................. 6

3.3 Impact of the Lack of Standards ........................................................................................ 8

3.4 Preserving Competitiveness – AI Platforms vs. AI Applications .................................... 9

4 Why OMG ...................................................................................................................................... 9

4.1 OMG’s Proven Capabilities ................................................................................................. 9

4.2 Existing OMG Standards in Support of AI...................................................................... 10

4.3 OMG’s Policies and Processes ......................................................................................... 10

5 A Roadmap for AI Standards................................................................................................... 11

6 Next Steps and Call to Action .................................................................................................. 14

6.1 Formation of an AI-Specific Subgroup at OMG ............................................................ 14

6.2 External Liaisons and Collaborations............................................................................. 14

6.3 Call for Participation ......................................................................................................... 15

6.4 AI Forums and Special Events ......................................................................................... 15

6.5 Toward an AI Standards Council? ................................................................................... 16

6.6 Conclusion.......................................................................................................................... 16

Appendix A: AAAI Taxonomy of Artificial Intelligence 17

Appendix B: AI-Related OMG Standards 19

Appendix C: References 21

© 2019 Object Management Group 3

1 Introduction and Executive Summary
The Board of Directors of the Object Management Group believes that the time has come
for defining standards in Artificial Intelligence (AI). Such standards will greatly accelerate
and improve the creation of useful AI applications by reducing the amount of low-level
interchange, integration, or interfacing work currently required.

The OMG is therefore establishing an initiative in AI standardization.

OMG is the best place for industry, academia, government and non-profit organizations to
come together to define such standards, because of the best practices it has developed
over the last thirty years and can easily extend to this new scope of work.

We initiated this paper to convey this message to both current and prospective members
of OMG. The paper explains the following.

• An expanded scope of AI has emerged from the renaissance of the discipline since
the beginning of the millennium. AI now reaches into many domains and integrates
new technologies such as ontologies, vision recognition, analytics and machine
learning for the industrial IoT, and more.
• When a technology area reaches a certain degree of maturity, standards can enable
innovation—rather than impede it—by freeing organizations from having to
constantly worry about the “plumbing” of systems or re-inventing platform
techniques and tools.
• OMG is uniquely placed to lead this effort because it has a proven process, it has
developed a series of foundational capabilities that can bootstrap the effort, and it
covers many industry domains that can benefit from AI standards.

Based on these observations, we propose actions to facilitate the development by our

members, in liaison with other organizations, of a roadmap for AI-enabling standards, and
we invite all interested parties to join us in this effort.

© 2019 Object Management Group 4

2 The New Scope of AI
While there are many competing definitions of AI, it is clearly not just rule-based expert
systems (the domain of AI that seemed the most promising during the early years, and
disappointed enough to cause the long period known as the “AI winter”), nor neural
networks or other forms of machine learning.

Table 1 provides a simplified list of key AI capabilities. This is a simplification of several

authors’ taxonomies of AI, such as [1] and [2]. The AAAI taxonomy of AI [1] is shown in
Appendix A as an example. Among this abundance of topics, we focused on those related
to existing OMG domains of interest.

Of course, in certain domains multiple AI capabilities are jointly used. Not all capabilities
are at the same level either – for example, machine learning can be used in vision,
advanced robotics or natural language processing.

Table 1 – Key AI Capabilities

AI Capability Definition Sample Applications

Machine Creation of an optimal response to Predictive maintenance.

Learning and a set of inputs, usually obtained Pattern recognition.
Deep Learning through the training of a neural Abnormality detection (safety,
network. security).

Vision Image processing to recognize Obstacle avoidance.

shapes. Facial recognition (security
applications).
Vehicle navigation systems.

Smart Robots Sensing, route planning, adaptive Mobile industrial robots.

prehension, ability to react to Human-assistive robots.
changes in the surroundings, etc. Telesurgery.

Natural Understanding of the semantics of Real-time translation.

Language natural language in spite of the Spam blocking.
Processing ambiguity of the vocabulary and Smart assistants (Alexa,
grammar. Cortana, chatbots, etc.)

Rules-Based These are the successors to the Loan application processing.

Systems expert systems of the 1960s-70s. Medical triage by the UK
They apply in an algorithmic National Health Service hotline.
manner rules captured from Robotic process automation
human experts. (RPA).

© 2019 Object Management Group 5

3 Why Standards in AI?
3.1 Phases of Standards Adoption
Every technology goes through a difficult relationship with standards in its early years:

• During the emerging phase, where technology development is led by research

groups, or by startups, standards are not seen as important or may even be
considered harmful.

• During early commercialization, each supplier is keen to create and preserve an

advantage, and to attract customers and lock them in to their proprietary
technology.

• When the technology matures, customers discover the need for integration,
interoperability, and migration from one system to another. At the same time,
suppliers need to attract a broader clientele. The need for standards thus emerges.
At the end of this phase, ad hoc standards are defined but are not yet well adopted.

• After a while, it becomes clear that more formality and governance are needed. The
industry then comes together within standards organizations and initiatives.
Compliance with standards becomes a selling point and a procurement criterion.

Premature standardization could stifle innovation; however, delayed standardization

creates unnecessary interchange, interoperability and integration difficulties. Today, many
organizations are still in the first or second stage listed above—they have not recognized
the need for standards in AI, or they have recognized it but are not convinced that adopting
standards is yet in their best interest. [3]

3.2 The Increasing Maturity of AI

By the early 2000s, it became clear that some of the early visions of AI could actually be
realized [4]. In particular, new practical applications of neural networks (under the name of
“machine learning” or “deep learning”) emerged. AI also proved critical in deciphering and
making decisions from the flood of data collected by Internet of Things (IoT) systems.

AI now has a significant impact and applications in practically all industries. Sectors that
lead this adoption include legal, insurance, crime and fraud investigation and prevention,
meteorology, media management, marketing, and more.

AI is now supported by comprehensive IT offerings, evidence by:

● the availability of platforms from big players such as IBM, Google, Amazon, or
Microsoft;
● the pervasiveness of cloud-based capabilities (with a consequent focus on APIs);

© 2019 Object Management Group 6

 ● the availability of massive amounts of data, both structured and unstructured,
including real-time IoT data, which can be exploited by AI and machine learning
algorithms.

The following activities and emerging standards (not an exhaustive list) indicate that the
time is ripe for standardization of certain aspects of AI:

• The Neural Network Exchange Format (NNEF)1 “reduces machine learning

deployment fragmentation by enabling a rich mix of neural network training tools
and inference engines to be used by applications across a diverse range of devices
and platforms.”
• The Open Neural Network eXchange (onnx)2 is an open-source, community-
driven effort to allow developers to more easily move between machine learning
frameworks.
• The Hierarchical Data Format 5 (HDF5 or .h5)3 is a standard representation of
scientific data sets, together with metadata, and is used in particular for the
interchange of machine learning training data sets.
• The ISO Subcommittee on Artificial Intelligence4 has been working on three
standards related to big data (ISO/IEC 20546, 20547-2 and 20547-5).
• The Consumer Technology Association (CTA)5 launched an initiative on AI aimed
at “improving efficiencies in AI and Health Care.”
• The National Institute for Science and Technology (NIST) is conducting several
related activities:
o It undertook a proof-of-concept project to develop an Industrial Ontology
Foundry (IOF)6.
o the Multimodal Information Group7 has conducted Language Recognition
Evaluation studies.
o In May 2019, it issued a Request for Information about the need for AI
standards [5], which received 98 responses [6], including an extensive one
from OMG. [7]

1
https://www.khronos.org/nnef
2
https://onnx.ai/
3
https://www.hdfgroup.org/
4
https://www.iso.org/standard/72826.html
5
https://www.cta.tech/News/Press-Releases/2019/April/CTA-Brings-Together-Tech-Giants,-Trade-
Association.aspx
6
https://www.nist.gov/publications/industrial-ontologies-foundry-proof-concept-project
7
https://www.nist.gov/itl/iad/mig
© 2019 Object Management Group 7
 • There are also AI initiatives in organizations such as OpenAI8, the Artificial
Intelligence Open Network (AI-ON)9, the Machine Intelligence Research
Institute (MIRI)10, the Allen Institute for Artificial Intelligence (AI2)11, the
Partnership on AI to Benefit People and Society12, the Cognitive Computing
Consortium13, the Consortium for Safer Artificial Intelligence14, and more.
• Open-source AI frameworks such as TensorFlow, Keras, Caffe, Scikit-learn,
Theano, and Torch are starting to be widely adopted.
• We now see a growing interest in the field of AI Ethics, including the work of Dr.
Andreas Vogel15 and position papers from various companies.16

3.3 Impact of the Lack of Standards

There is clear evidence from multiple sectors (finance, space, robotics, manufacturing,
healthcare, energy, and more) that conflicting models, languages and data formats may
impede the progress of applying AI. If AI models cannot be used together or do not have
consistent semantics, one may get the wrong results. Here are some more specific
examples of this impact.

• Deep or unsupervised learning algorithms are incredibly opaque and difficult to

understand, which impacts their reliability, maintainability, reuse, transparency,
respect for privacy, and more. Some research work has started in academia and in
the finance community to address these issues by attempting to combine
declarative ontologies and rules with the systems they specify. [8] [9]
• Without standard interfaces and well-defined ontologies, the robot industry cannot
evolve toward flexible, upgradable systems assembled from interchangeable
modules.

• Google’s new dataset search capability (https://toolbox.google.com/datasetsearch)

allows scientists to find datasets, and the World Wide Web Consortium (W3C) has
developed a data catalog vocabulary (DCAT); but the use of those datasets is
hindered by the lack of associated AI-specific metadata.

8
https://openai.com
9
https://ai-on.org
10
https://intelligence.org
11
https://allenai.org
12
https://partnershiponai.org
13
https://cognitivecomputingconsortium.com
14
https://makingaisafer,org
15
http://www.aisociety.life/
16
Such as for example https://www.ibm.com/watson/ai-ethics/
© 2019 Object Management Group 8
3.4 Preserving Competitiveness – AI Platforms vs. AI Applications
To properly focus AI-related standardization efforts, we must distinguish between
platforms and applications.

● Common foundations for AI, such as the representation of neural network building
blocks, APIs for knowledge bases and ontologies, or libraries of natural language
processing primitives, will accelerate development. Instead of users and tool
vendors wasting time supporting multiple APIs for these common capabilities, they
will be able to use standards-based tools of commercial or open source origin.

● Value-adding applications of AI should—and will—remain an open field for

worldwide innovation and competitiveness, with more time and resources devoted
to this level once the foundations and associated tools have become more easily
available.

4 Why OMG
4.1 OMG’s Proven Capabilities
Readers who are not familiar with the Object Management Group® (OMG®) should refer
to our website, www.omg.org, for more information, including OMG’s history since its
foundation [10], its standards development process, lists of our 245+ members and 225+
adopted specifications, and the organizations with which we maintain liaison relationships,
including ISO (see https://www.omg.org/about/liaison.htm).
In its thirty-year history, OMG has shown its ability to expand to new areas of concern.
From its beginning in object-oriented middleware—with CORBA® and related object
services—OMG transitioned to Model Driven Architecture (MDA), to cloud computing, to
software modernization and quality, to knowledge representations and reasoning, and to
standards supporting the Industrial Internet of Things.

OMG recognized early that horizontal or industry-generic standards were not sufficient to
help users, but that specific industries required their own standards. The organization
adapted to serve this need, pursuing both “platform” and “domain” standards through
separate task forces and technical committees. Over the years, the list of domains
addressed by OMG has evolved to stay aligned with industry needs. It now includes
finance, healthcare, manufacturing, C4I (Command, Control, Communications,
Computers, and Intelligence), robotics, space, and retail. All those areas are being
transformed by AI and will benefit from the development of AI-related standards.

© 2019 Object Management Group 9

4.2 Existing OMG Standards in Support of AI
OMG has developed, and is still developing, several AI-related cross-domain specifications,
in particular in the areas of knowledge representation and reasoning (KR&R). The complete
list of these specifications appears in Appendix B.
We expect more AI-related specifications to appear not only in current areas of work (such
as retail and robotics), but in other domains such as finance, space, C4I, and more.

4.3 OMG’s Policies and Processes

Many aspects of OMG’s policies, processes and procedures give the organization a
significant advantage when addressing a new area. Below is a list of those strengths.

Open Process. OMG’s standards development process has matured over the years, and
has been applied to 225+ specifications. It is recognized by ISO as being sufficiently
rigorous and disciplined as to qualify OMG for participation in the ISO Publicly Available
Specification (PAS) and FastTrack programs. As a result, many OMG standards have
become ISO standards.

The open process is implemented through a set of subgroups (Task Forces and Special
Interest Groups) open to all members, and through facilities such as wikis, a Jira® issues
database, etc. The practices employed to execute this process include the issuance of
discussion papers, Requests for Information (RFI), Requests for Proposals (RFPs), or the
adoption of Requests for Comments (RFCs).

The OMG’s policies and procedures are publicly available, even to non-members, at
https://www.omg.org/cgi-bin/doc.cgi?pp.

One Vote per Member. OMG is open to organizations of all sizes. Each organization has
one vote in the subgroups to which it contributes. Government and academia also
participate in the process under the same terms – this is important since researchers in
universities or national laboratories often do the leading-edge work in areas such as AI.

Moreover, an OMG member can contribute to a standard without being accredited by a

national standards organization, as is the case for example in ISO.

Free Specifications. OMG’s approved specifications are available to the public free of
charge.

Simple Intellectual Property (IP) policies. During the development of a specification, an

IP mode is selected. Almost all OMG standards are available under a royalty-free or “non-
assert” mode, where the holders of any precursor IP agree to give other companies the
right to use the IP they contributed.

© 2019 Object Management Group 10

No shelfware. A specification is only deemed final after evidence is provided that there are
existing implementations of the proposed standard—whether they are commercial
offerings, open-source versions, or internal implementations.

Living Standards. There is a formal change process to ensure that specifications do not
become stale. As soon as a specification is adopted, a task force is formed to address any
issues of interpretation or implementation raised by the public, or address new
requirements within its scope. When a revised version is published, a new such task force
is formed.

Architectural compatibility. All requirements documents and specifications must be

approved by the OMG Architecture Board, which is elected by the members. This ensures
compatibility and coherence across OMG’s set of specifications.

International applicability. OMG is an international organization and all its deliverables

are equally available worldwide.

5 A Roadmap for AI Standards

This section outlines actions that OMG envisions to develop standards for AI. It should be
noted that this is a roadmap, not a detailed plan, and that OMG does not dictate actions to
its members. It is OMG members, through their participation, who will define or refine the
actual roadmap and transform it into specific actions according to our process.

First, an AI Reference Model—similar to the seminal work that NIST performed to create
its widely recognized Cloud Computing Reference Architecture [11]—would be useful to
categorize cross-domain vs. domain-specific capabilities, platforms and tools.
Once this Reference Model is agreed upon, each type of organization (AI suppliers, AI users,
government entities, etc.) can determine which part of the model their initiatives will
address. The Reference Model could distinguish:

• Technology building blocks that address AI-specific as well as other needs.

• Knowledge representation technologies (semantic web, ontologies, rules…).
• AI capabilities (neural networks and other forms of machine learning, pattern
recognition, planning, etc.).
• Cross-sector AI applications (such as facial recognition).
• Sector-specific AI applications (e.g., medical diagnostic)

Table 2 below lists some of these potential areas of standardization.

© 2019 Object Management Group 11

 Table 2 – AI Standard Areas

AI Domains Future Standard Areas Goals and Benefits of Standards

AI Architecture or reference Achieve a common understanding and make it

Architecture model to categorize cross- easier for participants to contribute to parts of the
and Logical domain vs. domain- architecture.
Components specific capabilities,
platforms and tools.

Machine Training data set Allow the sharing of data that can be used to train
Learning representation and models. While the models may be proprietary, the
metadata data sets (e.g., anonymized equipment or patient
data) could, if associated with appropriate
metadata and shared in a standard format,
accelerate the improvement of the models.

IoT interoperability Allow vendors and users, especially in critical

language industries with a potential impact on the public and
the environment, to combine datasets or share
operational rules to improve safety or lower costs.

Machine learning decision Address a growing demand for the ability to “audit”
explanation model. how a neural network arrived at a certain
conclusion. There are technical, legal, regulatory
and ethical reasons why the ability to explain the
decision may be required.

Cognitive Standard APIs for access Allow users to substitute components from multiple
Services to Ai algorithms in vision, suppliers providing those services, without
speech recognition, impeding the competition between those suppliers.
language understanding,
intelligent search and
more.

Facial and Protocols, APIs, Provide a secure and traceable way for justified
biometrics encryption, access rules. access while protecting personal data against
recognition unintended use through encryption or obfuscation
techniques.

Speech Evaluation metrics, test A published standard would make the test
Recognition sets, evaluation methodology, metrics, and test sets available to all
methodology, APIs. developers of speech/language recognition
systems.

© 2019 Object Management Group 12

 AI Domains Future Standard Areas Goals and Benefits of Standards

Smart Standardized planning Enhance the ability to replace a component of a

Robotics language. robotic system with another one by making the
output of planning software transferable from one
brand of robot to another.

Natural Information classification Help solve the information overflow problem (the
Language and rule representation challenge of processing the mass of data received
Processing for automatic message by humans on a daily basis) by providing a common
processing by intelligent representation of the non-confidential content of
agents. messages, which will allow machine learning-based
intelligent assistants and spam filters.

Agents Agent modeling languages Increase the rigor and consistency of agent-related
and techniques. Alert and specifications, and ensure interoperability of agent-
notification interfaces. based systems.

Augmented Content markup and Ensure that AR reaches its full potential as an
Reality management, object enhancement to human life and information use.
identification, navigation.

Sector- Rules and decision models Enable various levels of reasoning and automation,
Specific that leverage, but go as appropriate for each sector, through the ability
Information beyond, the work already to interchange rules and decision models that are
Models and done on sector-specific used by AI applications.
Decision ontologies.
Models

Security of AI Authentication, Practices that relate to the security of the AI

Components authorization and access components of IoT and other systems, for example
control for AI components. to prevent the injection of illegitimate data into the
training of a machine learning algorithm.

AI Ethics Reference architecture for Diminish the risk of accidents or social rejection by
confidentiality, privacy and provide guidance to developers and users so that AI
ethical decision-making in can be applied responsibly, ethically and legally.
AI.

© 2019 Object Management Group 13

6 Next Steps and Call to Action
6.1 Formation of an AI-Specific Subgroup at OMG
Having addressed a “slice” of AI through its Agent Platform Special Interest Group, OMG
recently broadened its involvement by chartering an AI Platform Special Interest Group in
June 2019. The creation of an AI Platform Task Force is currently being undertaken in order
to enable the authorship of specifications, thus accelerating the solicitation and adoption of
AI standards.

Subgroups (SIGs or Task Forces) do not work in isolation within OMG. They routinely confer
and collaborate on common interests. The AI Task Force will influence work done in other
Task Forces such as Healthcare, Finance, Manufacturing, Retail, Robotics, etc.

6.2 External Liaisons and Collaborations

OMG will continue to leverage its existing liaison agreements with other standards
organizations and associations17, and develop new ones as appropriate, including the IEEE
Society on Social Implications of Information Technology’s Standards Committee (IEEE-SSIT
SC)18 or the Augmented Reality for Enterprise Alliance (AREA)19.

In May 2019, OMG responded to NIST’s RFI mentioned earlier [5], aimed at developing a
U.S. “AI standards engagement plan.” While NIST’s effort is U.S.-specific and OMG is
international in scope, our response [7] can lead to a fruitful collaboration. In particular,
OMG suggested that NIST:

• Develop the reference model mentioned in Section 5 above.

• Sponsor some of the incipient work to improve the reliability, maintainability,
reusability, transparency, respect for privacy, etc., of deep learning algorithms
through standards for combining declarative ontologies and rules with the systems
they specify.
• Provide additional funding for NIST’s Industrial Ontology Foundry and similar
projects to accelerate their progress.

OMG is ready to discuss collaborating with any of these organizations to advance our
common interests.

17
https://www.omg.org/about/liaison.htm)
18
http://sites.ieee.org/sagroups-ssit/
19
https://thearea.org/
© 2019 Object Management Group 14
6.3 Call for Participation
The vision expressed in this paper remains a potential roadmap until realized as OMG
members “roll up their sleeves” and participate in the activity. We call on participants in the
AI community to provide input from as many sources as possible, discuss and improve the
roadmap, and help select the top priority areas. While we will use and analyze the
responses to the NIST RFI, a logical vehicle for additional information gathering is for OMG
to issue its own RFI, in particular to involve international organizations. This will be the first
step in applying OMG’s open process to start developing appropriate standards.

We invite any whose concerns and interests have been touched upon in this paper to join
us as we move forward. Concrete steps include:

● Write to [email protected] to express your interest and provide feedback about this
paper as well as the specifications mentioned in it, additional ideas that will
influence OMG’s AI roadmap, and to inform OMG of other relevant efforts.
● Visit the AI subgroup’s wiki at https://www.omgwiki.org/AI.
● Attend meetings organized by OMG’s AI subgroup (to be posted at
https://www.omg.org/events).
● Inform OMG of other relevant efforts.
● Consider an OMG membership in order to have a real impact (including voting
rights) on OMG’s work.

6.4 AI Forums and Special Events

OMG has a track record of holding special events (which may be called forums, summits,
symposia, or workshops) to address the needs for standards in specific sectors. Since one
of our quarterly meetings takes place each year in March in the Washington, D.C. area, we
plan to hold an annual full-day event on AI standards, jointly with other interested
organizations, starting in March 2020. This first AI Workshop would have the following
goals:

● Present analyses by various organizations of the needs for AI standards.

● Expose participants to the relevance of existing OMG standards to the building of AI
capabilities.
● Provide an opportunity for industry, government and academia participants—not
only OMG members—to exchange ideas.
● Attract those players who want to participate in the standards development process
to become members of OMG in order to pursue this work.

© 2019 Object Management Group 15

“Challenge” events (defined in a way similar to DARPA, see https://www.darpa.mil/work-
with-us/public/prizes) are a good way to invite the providers of technology to demonstrate
their capabilities, including standards-based interoperability. To promote several of our
standards such as UML and BPMN, OMG routinely hosts interoperability demonstrations
during our meetings. We propose holding, jointly with other interested organizations, “AI
interoperability challenges.” These may leverage OMG standards once developed, or could
equally be used to expose areas where standards are needed.

6.5 Toward an AI Standards Council?

OMG has the capacity and motivation to create an AI Standards Council, bringing together
representatives from industry, academia and other government agencies, with a mission
similar to those of other OMG managed programs such as the Industrial Internet
Consortium (IIC)20 or the Consortium for Information and Software Quality (CISQ)21. This
would represent the next level of activity and visibility.

A separate council requires significant levels of resources, provided or funded by a slate of

sponsoring organizations. On the other hand, it provides a locus for a number of activities
that are complementary to the development of standards, such as:

● Advocacy—promoting the application of AI to various domains.

● Education and Marketing—helping improve the understanding of AI capabilities,
best practices among end users, practical guides to getting started, success stories,
etc.
● Demonstrations— “laboratory” work or “testbeds” developed as collaboration
between members.
● Liaison—connecting with industry bodies that are not involved in standards as well
as with standards organization to which requirements for new capabilities,
expressed by members, can be sent for action.
● Certification—providing evidence of the qualification of member companies’
employees who demonstrate higher levels of proficiency.

6.6 Conclusion
AI has matured and its successful application can be enhanced by the development and
adoption of standards. OMG has the capability and motivation to successfully expand its
activities in this domain. We encourage the AI community—across all domains and
regions—to get involved in this effort by contacting us, participating in our AI-related
activities and events, and joining OMG to take an active role.

20
https://www.iiconsortium.org
21
https://www.it-cisq.org/
© 2019 Object Management Group 16
Appendix A: AAAI Taxonomy of Artificial Intelligence
This de facto taxonomy of AI material (articles, papers, books…) is extracted from the search
menu of AITopics, an official publication of the Association for the Advancement of Artificial
Intelligence (AAAI), at www.aitopics.org/search. It is provided here solely as an example of
such taxonomies. Some of the subcategories include a third level of detail; we only
included the first two levels here. For the complete taxonomy, see the source.

Artificial Intelligence
• Assistive Technologies o Fuzzy Control
o Inductive Learning
• Challenges
o Kernel Methods
• Cognitive Science o Learning Graphical Models
o Childhood Development o Learning in High Dimensional
o Cognitive Architectures Spaces
o Creativity and Intelligence o Memory-Based Learning
o Emotion o Neural Networks
o Neuroscience o Pattern Recognition
o Problem Solving o Performance Analysis
o Simulation of Human Behavior o Reinforcement Learning
• Games o Statistical Learning
o Supervised Learning
o [list of games omitted]
o Transfer Learning
• History o Unsupervised or Indirectly
• Human-Centered Computing Supervised Learning

• Issues • Natural Language

o Arguments Against AI o Chatbots
o Philosophy o Discourse and Dialogue
o Social and Ethical Issues o Explanation and Argumentation
o Turing’s Test o Generation
o Grammars and Parsing
• Machine Learning
o Information Extraction
o Association Learning
o Information Retrieval
o Bayesian Networks
o Machine Storytelling
o Computational Learning Theory
o Machine Translation
o Control Theory
o Question Answering
o Decision Tree Learning
o Text Classification
o Ensemble Learning
o Text Processing
o Evolutionary Systems
o Understanding
o Forecasting
© 2019 Object Management Group 17
 • Representation and Reasoning • Robots
o Abductive Reasoning o Autonomous Vehicles
o Agents o Humanoid Robots
o Analogical Reasoning o Locomotion
o Automatic Programming o Manipulation
o Belief Revision o Robot Planning and Action
o Blackboard Systems o Robots in the Home
o Case-Based Reasoning o Robots in the Workplace
o Commonsense Reasoning o Soccer Robots
o Constraint-Based Reasoning
• Science Fiction
o Description Logic
o Diagnosis • Speech
o Diagrams and Models o Acoustic Processing
o Expert Systems o Speech Recognition
o Frame-Oriented Architecture o Speech Synthesis
o Information Fusion • Systems and Languages
o Logic and Formal Reasoning o Distributed Architectures
o Mathematical and Statistical o Problem-Independent
Methods Architectures
o Metareasoning o Problem-Specific Architectures
o Model-Based Reasoning o Programming Languages
o Nonmonotonic Logic
• The Future
o Object-Oriented Architecture
o Ontologies • Vision
o Optimization o Face Recognition
o Personal Assistant Systems o Gesture Recognition
o Planning and Scheduling o Handwriting Recognition
o Qualitative Reasoning o Image Understanding
o Rule-Based Reasoning o Optical Character Recognition
o Scientific Discovery o Sketch Understanding
o Scripts and Frames o Video Understanding
o Search
o Semantic Networks
o Spatial Reasoning
o Temporal Reasoning
o Uncertainty

© 2019 Object Management Group 18

Appendix B: AI-Related OMG Standards
The following list is an Appendix to section 4.2 of this paper. It lists OMG specifications that
are either published or under development, and that establish standard foundations
required by AI platforms and applications.

● General knowledge representation and reasoning (KR&R) standards already

published:

o Ontology Definition Metamodel (ODM) enables ontology management and

development using OMG’s Model Driven Architecture (MDA) stack for the
Resource Description Framework (RDF) and RDF Schema, the Web Ontology
Language (OWL), ISO Common Logic (CL), and Topic Maps. The specifications,
and hence tools, in the stack provide support for metamodel and model
storage, versioning, querying and transformation. ODM also provides a
profile for use of UML-compliant tooling for graphical modeling of
ontologies. [12]
o Distributed Ontology, Modeling, and Specification Language (DOL)
provides a language and transformations at the semantic level aimed at
achieving integration and interoperability of ontologies, specifications and
models developed independently and in differing ontology languages and
logic frameworks. [13]
o Semantics of Business Vocabularies and Rules (SBVR) enables a
structured English representation and the interchange of business
statements. [14]
o Decision Modeling and Notation (DMN) enables the executable
representation of business decisions linked to their data sources; and
management of the rules in the business context. [15]
• Non-interface-oriented robotics standards (i.e., those that enable knowledge
interchange or other interactions rather than strictly providing interfaces):
o Robotic Technology Component (RTC) defines a component model and
certain infrastructure services supporting robotics software development.
o Finite State Machine Component for RTC (FSM4RTC) extends the RTC
specification for interchange of state and state machine related content
o Robotic Interaction Service Framework (RoIS) defines a framework for
services supporting interactions between humans and robots, including but
not limited to facial detection and identification, sound detection, language
recognition and understanding, speech generation, interpretation of
gestures, and the like.

© 2019 Object Management Group 19

 • OMG has also published sector-specific standards in areas including knowledge
representation and reasoning for Finance. Our Financial Industry Business
Ontology (FIBO) standard, jointly developed and evolving through our liaison with
the Enterprise Data Management Council, provides an ontology for legal entities,
financial instruments and related concepts as well as reference data for the
representation of currencies, various banking identifiers, legal entity identifiers,
market identifiers, and so forth. [16]
• Many other OMG standards provide the supporting infrastructure that enable the
development of software, systems, and interfaces that include AI as a component.
The above-mentioned OMG specifications (in bold characters) can all be found on the
OMG website at https://www.omg.org/spec/.

• General KR&R standards under development:

o Application Programming Interfaces for Knowledge Platforms (API4KP)

defines a set of ontologies and interfaces needed to incorporate knowledge
representation and reasoning tools, as well as other AI capabilities, in a
broader enterprise environment, including but not limited to interfaces
between inference engines, rule engines, knowledge graphs, and various
sources of information required to build out a comprehensive environment.
[17]
Initial reference implementations have been deployed at the Mayo Clinic and
are under development at the Veterans’ Administration.
• General robotics standards under development:

o Robotics Service Ontology (RoSO) defines a set of ontologies for robot-to-

human interactions and the services needed to perform such interactions.
This work is being done in conjunction with the IEEE Robotics & Automation
Society’s Autonomous Robotics Group, which is responsible for IEEE 1872 –
an ontology that focuses on core (generic) terminology and capabilities of
robot systems at a relatively high level.
• OMG’s sector-specific AI-related specifications under development:

o A retail specification for digital receipts that embodies an ontology defining

not only the receipts themselves, but also content related to jurisdiction-
specific taxation.
o A joint effort between OMG’s Retail and Robotics Task Forces to create a
standard for point-of-sale/point-of-service (POS) robotic interfaces for the
2020 Olympics specifically, but which will be broadly applicable to POS
robotic services.

© 2019 Object Management Group 20

Appendix C: References
[1] Association for the Advancement of Artificial Intelligence (AAAI): a taxonomy of AI.
www.aitopics.org/search (see the “Technology” drop-down list in the left column).

[2] Golstein, B. (2018): A Brief Taxonomy of AI. White paper by Sharper AI Pte Ltd.
http://www.sharper.ai/taxonomy-ai/

[3] Johnson, L. (2004): A False Sense of Proprietary. The Standards Edge: Dynamic Tension,
Chapter 17. Bolin (ed.), Sheridan Books.
https://www.thebolingroup.com/standards_series.html

[4] Smith, C., McGuire, B., Huang, T., & Yang, J. (2006): The History of Artificial Intelligence.
University of Washington materials for course CSED 590A.
https://courses.cs.washington.edu/courses/csep590/06au/projects/history-ai.pdf

[5] National Institute of Standards and Technology (2019): Request for Information on
Artificial Intelligence Technical Standards and Tools.
https://www.federalregister.gov/documents/2019/05/01/2019-08818/artificial-
intelligence-standards

[6] National Institute of Standards and Technology (2019): Comments Received for RFI
about Federal Engagement in Artificial Intelligence Standards.
https://www.nist.gov/topics/artificial-intelligence/comments-received-rfi-about-
federal-engagement-artificial

[7] Object Management Group (2019): Response to the NIST RFI on Artificial Intelligence
Technical Standards and Tools. https://www.nist.gov/document/nist-ai-rfi-omg-001pdf
[8] Antoniou, Grigoris, et al. (2005): Combining Rules and Ontologies. A Survey. REWERSE.
http://rewerse.net/deliverables/m12/i3-d3.pdf

[9] Xu, Li, et al.: Combining Declarative and Procedural Knowledge to Automate and Represent
Ontology Mapping. https://www.deg.byu.edu/papers/SWAT06-131.pdf

[10] OMG (1989): Object Management Group Established. Press release.

https://www.omg.org/marketing/25th/OMG_Forms_PR_1989.pdf

[11] National Institute of Standards and Technology (2011): NIST Cloud Computing Reference
Architecture. Special Publication 500-292. https://doi.org/10.6028/NIST.SP.500-292

[12] OMG: Ontology Definition Metamodel™ (ODM™) https://www.omg.org/odm/

[13] OMG: Distributed Ontology, Model, and Specification Language™ (DOL™).

https://bei.omg.org/dol/
[14] OMG: About the Semantics of Business Vocabulary and Rules Specification Version 1.4.
https://www.omg.org/spec/SBVR/

© 2019 Object Management Group 21

[15] OMG: Decision Model and Notation™ (DMN™). https://www.omg.org/dmn/

[16] OMG: Financial Services Standards. https://www.omg.org/hot-topics/finance.htm

[17] Athan, T., Bell, R., Kendall, E., Paschke, A., & Sottara, D. (2015): API4KP Metamodel: A
Meta-API for Heterogeneous Knowledge Platforms. Conference paper, International
Symposium on Rules and Rule Markup Languages for the Semantic Web. Springer.
https://link.springer.com/chapter/10.1007/978-3-319-21542-6_10

© 2019 Object Management Group 22