SlideShare a Scribd company logo
1 of 23
The professional home for the engineering and technology community worldwide
ieee.org
The professional home for the engineering and technology community worldwide
Approaches in using Generative AI in Automation Solutions
IEEE International Conference on Autonomic Computing and Self-Organizing Systems (ACSOS) 2023
Allen Chan, IBM Distinguished Engineer and CTO, IBM Business Automation
1
The professional home for the engineering and technology community worldwide
2 ieee.org
The professional home for the engineering and technology community worldwide
Allen Chan
2
Co-Author of 2 books on Automation and AI
Recent collaboration with University of Toronto (Prof. Eric Yu) and
York University (Prof. Alex Sendervich) on AI Research
1.Transformer Models for Activity Mining in Knowledge-Intensive Processes. Feb 2023. Book: Business Process
Management Workshops
2.Managing and Simplifying Cognitive Business Operations Using Process Architecture Models. May 2019.
Book: Advanced Information Systems Engineering
3.Solution Patterns for Machine Learning. May 2019. Book: Advanced Information System Engineering.
4.Modeling and Analyzing Process Architecture for Context-Driven Adaptation: Designing Cognitively-Enhanced
Business Processes for Enterprises. Oct 2018. 2018 IEEE 22nd International Enterprise Distributed Object
Computing Conference (EDOC)
5.Designing Process Architectures for User Engagement with Enterprise Cognitive Systems. Oct 2017. IFIP
Working Conference on The Practice of Enterprise Modeling
The professional home for the engineering and technology community worldwide
Agenda
3
▸Generative AI is not Enough
▸Content Use Cases
▸Decisions Use Cases
▸Workflow Use Cases
▸Summary
▸Q&A
The professional home for the engineering and technology community worldwide
1 Generate
▸ Generate content
based on request,
e.g., write an email,
draft a proposal
2 Classify
▸ Read and classify
input based on few
examples. Can be
used to categorize
customer issues or
group inputs.
3 Summarize
▸ Transform dense text
into your
personalized
overview, capturing
key points.
4 Extract
▸ Pull the information
you want from large
documents. Identify
named entities,
parse terms and
conditions, etc.
4
Generative AI Core Capabilities
The professional home for the engineering and technology community worldwide
GenerativeAI alone is not enough
5
Contrary to popular belief, Generative AI does not:
1. Think
2. Reason
3. Check for Facts
4. Coordinate Tasks
5. Perform Real Work
=> To do that, we need to combine Generative AI with Automations
The professional home for the engineering and technology community worldwide
Cost
Enterprise
challenges
with using
Gen AI
▸ Training on LLM is expensive and requires
hundreds or thousands of GPU hours. Simple
approaches can easily lead to token explosion
and exponential cost.
▸ LLM is designed to process “language” data, not
Excel, not PDF, not JSON, not Powerpoint, not
photos… so all information will need to be
“normalized” before it can be stored in LLM.
▸ Many documents in an enterprise content
repository have access control and contain
sensitive and private data. LLMs have no access
control or data governance.
Text Only
6
Access
Control
The professional home for the engineering and technology community worldwide
Sensitive
Data
Enterprise
challenges
with using
Gen AI
▸ Since LLMs do not forget, guardrails must be
provided to ensure sensitive and private data
are not stored intentionally or
unintentionally in the model.
▸ We want the model to answer questions based
on facts and not making it up (“hallucinating”).
In addition, we must be able to trace the
answers back to the original documents.
▸ Different LLMs have different performance and
quality. They use different algorithms, different
training data, different weighting on data, and
different pre- and post-processing. The quality of
the answers will rely on picking the correct LLM for
the problem domain.
Trust &
Traceability
7
Prediction
Quality
The professional home for the engineering and technology community worldwide
Use Cases
Content
▸ Provide answers to questions
with LLMs
Decisions
▸ Add reasoning and decisions to
LLMs
Workflow
▸ Automate work with LLMs
8
The professional home for the engineering and technology community worldwide
1
▸ Feed documents
directly into fine-
tuning
2
▸ Leverage ECM Built-
in Search
3
▸ Refine inputs with
knowledge pre-
processing
4
▸ Adopt a VectorDB
for rich semantic
search
9
Content –
Provide answers to
questions with LLMs
The professional home for the engineering and technology community worldwide
10
Pros
• The approach is straightforward and easy to
implement
• Allows users to get answers directly from LLM
Cons
• Takes a lot of time and money to fine-tune the model with
millions of documents
• Not all documents are textual and can be consumed as-is… so
any answer could be incomplete.
• No role-based access control means all data used in fine-
tuning will be available to anyone that is asking the question
• Mixing enterprise data with public-domain foundational data
can easily lead to hallucination
Content – Feed documents
directly into fine-tuning
1. Take documents in Document Store use them directly in Fine-
Tuning.
2. Build a Q&A system to just the question through the prompt.
The professional home for the engineering and technology community worldwide
11
Pros
• The approach is straightforward and easy to
implement
• Get answers directly from the information source –
chances of hallucination is greatly limited.
• The token count is limited only to information
passed to the model.
• Role-based access control can be applied
Cons
• We can only find documents with matching search
keywords
• Additional processing would be required to limit the
number of matching results as there might be too many
tokens.
• Given the result is the entire document, it would be
difficult to extract the precise relevant sections.
Content – Leverage Document
Store Built-in Search
1. Perform a search first and retrieve the documents
matching the search criteria.
2. Feed the document as part of the prompt and ask
the LLM to limit the answers to only information in
the prompt.
The professional home for the engineering and technology community worldwide
12
Pros
• In addition to having the same pros as the previous
approach, the additional knowledge pre-processor
allows more document types to be passed to the
LLMs
• By removing sensitive data ahead of the prompt, we
reduce the risk of sensitive data being stored in the
LLMs.
Cons
• We can only find documents with matching search
keywords
• Additional processing would be required to limit the
number of matching results as there might be too many
tokens.
• Given the result is the entire document, it would be
difficult to extract the precise relevant sections.
Content – Leverage
Knowledge Pre-processor
1. Leverage existing FileNet Search
2. Inject a Knowledge Pre-processor before we
generate the prompt to perform OCR and other
information filtering to remove sensitive and
personal data.
The professional home for the engineering and technology community worldwide
13
Pros
• Use of VectorDB allows use of document fragments
in search results, thus reducing the token counts
and increases the result relevance.
• Knowledge aggregator/formatter can combine
information from multiple data sources to create a
single answer.
Cons
• Having to pre-process all the documents is a time-
consuming process.
Content – Adopt Retrieval
Augmentation Generation (RAG)
approach for semantic search
1. Leverage a VectorDB to shard, index and perform
semantic search on documents
2. Include a knowledge aggregator/formatter to combine
results from LLMs and original data to created complete
answers.
The professional home for the engineering and technology community worldwide
1
▸ NLU + Reasoning
2
▸ Reasoning followed
by NLG
3
▸ Generate business
rules from business
policies
14
Decisions –
Add reasoning and
decisions to LLMs
The professional home for the engineering and technology community worldwide
15
Pros
• Straightforward implementation requiring minimal
or no fine tuning.
• Allow rules to be used in non-traditional API-based
implementation e.g., in a chatbot.
Cons
• Limited to identified text extraction capabilities.
• In cases where the extraction process does not find the
expected data from the input text, it’s crucial to set
appropriate guardrails to handle such scenarios during
the reasoning phase.
Decisions – NLU + Reasoning
1. Leverage NLU capabilities of LLMs to extract
business entities (e.g., city name, job role) and
pass those as inputs to Decisions.
The professional home for the engineering and technology community worldwide
16
Pros
• Straightforward implementation requiring minimal
or no fine tuning.
• Allow rules to be used in non-traditional API-based
implementation e.g., in a chatbot.
Cons
• Prompt will need to be prepared carefully to ensure
generated outputs reflect the intent of the decision
outcomes.
Decisions – Reasoning
followed by NLG
1. Leverage NLU capabilities of LLMs to extract
business entities (e.g., city name, job role) and
pass those as inputs to Decisions.
The professional home for the engineering and technology community worldwide
17
Pros
• Leveraging the LLM as a knowledge extraction tool
enables the extraction of rules and an underlying
data ontology. The extracted rules, once reviewed
by a human, can automate decisions with
traceability and determinism.
Cons
• Depending on the business domains, further fine-
turning will be required.
• Require a good amount of prompt engineering to
ensure the generated rules model is syntactically and
semantically correct.
• Generate models will require require by SME and still
need to be tested/validated.
Decisions – Generate business
rules from business policies
1. Extract automation assets, including business rules, data
models, and signatures, from plain text business policies.
2. Use the generated rules models as starting point to create
business decisions.
The professional home for the engineering and technology community worldwide
1
▸ Classify inputs and
perform requests
2
▸ Generate responses
as part of a business
processes
3
▸ Automate LLM
finetuning and
exception handling
4
▸ Generate workflow
from description
18
Workflow –
Automate work with LLMs
The professional home for the engineering and technology community worldwide
19
Pros
• Straightforward implementation requiring minimal
or no fine-tuning.
• Allow workflow to work directly with natural
language inputs without pre-processing
Cons
• Limited to identified text extraction capabilities.
• In cases where the extraction process does not find the
expected data from the input text, it’s crucial to set
appropriate guardrails to handle such scenarios during
the reasoning phase.
Workflow – Classify/Extract
inputs and perform work
Leverage LLM to classify and/or extract input data and use
those information to perform work
If LLM fails to classify the information, we can delegate to a
human user to decide and feed the decision back to the
system.
The professional home for the engineering and technology community worldwide
20
Pros
• Support a more intuitive engagement experience
with users.
• Support other non-API based interaction, e.g., with
smart assistants like Alexa, voice-based interaction
with text-to-speech
Cons
• Prompt will need to be prepared carefully to ensure
generated outputs reflect the intent of the work
outcomes.
Workflow – Generate responses
as part of a business processes
Use LLM to generate responses (e.g., email
messages, work summary, task notification) directly
from business processes data.
The professional home for the engineering and technology community worldwide
21
Pros
• This is a must-have for any fine-tuning or indexing
exercise that uses enterprise data.
Cons
• Incur additional review and approval cost.
Workflow – Automate LLM finetuning
and exception handling
Use workflow to provide Human-in-the-loop review
and approval before knowledge data will be fed to
LLM for fine tuning.
The professional home for the engineering and technology community worldwide
22
Pros
▸ Provide a natural language approach to generate
business processes.
▸ The quality of the generated process will rely on the
quality of the LLMs.
▸ It is also possible to ask the LLM to generate the
process from just a simple business goal, but the result
will rely on the foundational data that was used to do
the initial model training.
Cons
▸ Do not expect to get a fully working BPMN model, work such as
defining the data types, decision logic, user interfaces and
validating the process will be required.
Workflow – Generate Workflow
from Description
From a high level description of a business process, generate a
BPMN process that can be used as a starting point for an
automation workflow solution.
Due to different generation quality of the LLM models, it is better
to use LLM to summarize or extract procedures from the business
process document and use a post-processor to take the output
and turn that into a formal BPMN document.
The professional home for the engineering and technology community worldwide
1
▸ Gen AI, when used
appropriately, can
speed up certain
data processing.
2
▸ Combine Gen AI
and Automations
can accelerate your
digital
transformation
journey
3
▸ Careful planning
and consideration
for private and
sensitive data is a
must for enterprise
usage.
4
▸ LLM models are evolving
very fast and each day
there are new models.
Understanding which
models to use for the
purpose at hand is the
key to yield good results.
23
Summary

More Related Content

What's hot

Using Generative AI
Using Generative AIUsing Generative AI
Using Generative AIMark DeLoura
 
Generative-AI-in-enterprise-20230615.pdf
Generative-AI-in-enterprise-20230615.pdfGenerative-AI-in-enterprise-20230615.pdf
Generative-AI-in-enterprise-20230615.pdfLiming Zhu
 
Leveraging Generative AI: Opportunities, Risks and Best Practices
Leveraging Generative AI: Opportunities, Risks and Best Practices Leveraging Generative AI: Opportunities, Risks and Best Practices
Leveraging Generative AI: Opportunities, Risks and Best Practices Social Samosa
 
DevDive_UnleashthFullPotentialofAutomationwithGenAI.pptx
DevDive_UnleashthFullPotentialofAutomationwithGenAI.pptxDevDive_UnleashthFullPotentialofAutomationwithGenAI.pptx
DevDive_UnleashthFullPotentialofAutomationwithGenAI.pptxRohit Radhakrishnan
 
ChatGPT, Foundation Models and Web3.pptx
ChatGPT, Foundation Models and Web3.pptxChatGPT, Foundation Models and Web3.pptx
ChatGPT, Foundation Models and Web3.pptxJesus Rodriguez
 
14 2 2023 - AI & Marketing - Hugues Rey.pdf
14 2 2023 - AI & Marketing - Hugues Rey.pdf14 2 2023 - AI & Marketing - Hugues Rey.pdf
14 2 2023 - AI & Marketing - Hugues Rey.pdfHugues Rey
 
Generative AI Use cases for Enterprise - Second Session
Generative AI Use cases for Enterprise - Second SessionGenerative AI Use cases for Enterprise - Second Session
Generative AI Use cases for Enterprise - Second SessionGene Leybzon
 
Exploring Opportunities in the Generative AI Value Chain.pdf
Exploring Opportunities in the Generative AI Value Chain.pdfExploring Opportunities in the Generative AI Value Chain.pdf
Exploring Opportunities in the Generative AI Value Chain.pdfDung Hoang
 
Generative AI: Past, Present, and Future – A Practitioner's Perspective
Generative AI: Past, Present, and Future – A Practitioner's PerspectiveGenerative AI: Past, Present, and Future – A Practitioner's Perspective
Generative AI: Past, Present, and Future – A Practitioner's PerspectiveHuahai Yang
 
arbar https://www.slideshare.net/Solutionman/charles-caldwell-improve-your-li...
arbar https://www.slideshare.net/Solutionman/charles-caldwell-improve-your-li...arbar https://www.slideshare.net/Solutionman/charles-caldwell-improve-your-li...
arbar https://www.slideshare.net/Solutionman/charles-caldwell-improve-your-li...SOLTUIONSpeople, THINKubators, THINKathons
 
Responsible Generative AI
Responsible Generative AIResponsible Generative AI
Responsible Generative AICMassociates
 
An Introduction to Generative AI - May 18, 2023
An Introduction  to Generative AI - May 18, 2023An Introduction  to Generative AI - May 18, 2023
An Introduction to Generative AI - May 18, 2023CoriFaklaris1
 
Thabo Ndlela- Leveraging AI for enhanced Customer Service and Experience
Thabo Ndlela- Leveraging AI for enhanced Customer Service and ExperienceThabo Ndlela- Leveraging AI for enhanced Customer Service and Experience
Thabo Ndlela- Leveraging AI for enhanced Customer Service and Experienceitnewsafrica
 
AI FOR BUSINESS LEADERS
AI FOR BUSINESS LEADERSAI FOR BUSINESS LEADERS
AI FOR BUSINESS LEADERSAndre Muscat
 
The Rise of the LLMs - How I Learned to Stop Worrying & Love the GPT!
The Rise of the LLMs - How I Learned to Stop Worrying & Love the GPT!The Rise of the LLMs - How I Learned to Stop Worrying & Love the GPT!
The Rise of the LLMs - How I Learned to Stop Worrying & Love the GPT!taozen
 

What's hot (20)

Using Generative AI
Using Generative AIUsing Generative AI
Using Generative AI
 
Generative-AI-in-enterprise-20230615.pdf
Generative-AI-in-enterprise-20230615.pdfGenerative-AI-in-enterprise-20230615.pdf
Generative-AI-in-enterprise-20230615.pdf
 
Leveraging Generative AI: Opportunities, Risks and Best Practices
Leveraging Generative AI: Opportunities, Risks and Best Practices Leveraging Generative AI: Opportunities, Risks and Best Practices
Leveraging Generative AI: Opportunities, Risks and Best Practices
 
Matt Lewis - The Hardest Thing-Final to Host.pdf
Matt Lewis - The Hardest Thing-Final to Host.pdfMatt Lewis - The Hardest Thing-Final to Host.pdf
Matt Lewis - The Hardest Thing-Final to Host.pdf
 
DevDive_UnleashthFullPotentialofAutomationwithGenAI.pptx
DevDive_UnleashthFullPotentialofAutomationwithGenAI.pptxDevDive_UnleashthFullPotentialofAutomationwithGenAI.pptx
DevDive_UnleashthFullPotentialofAutomationwithGenAI.pptx
 
ChatGPT, Foundation Models and Web3.pptx
ChatGPT, Foundation Models and Web3.pptxChatGPT, Foundation Models and Web3.pptx
ChatGPT, Foundation Models and Web3.pptx
 
14 2 2023 - AI & Marketing - Hugues Rey.pdf
14 2 2023 - AI & Marketing - Hugues Rey.pdf14 2 2023 - AI & Marketing - Hugues Rey.pdf
14 2 2023 - AI & Marketing - Hugues Rey.pdf
 
Generative AI Use cases for Enterprise - Second Session
Generative AI Use cases for Enterprise - Second SessionGenerative AI Use cases for Enterprise - Second Session
Generative AI Use cases for Enterprise - Second Session
 
Exploring Opportunities in the Generative AI Value Chain.pdf
Exploring Opportunities in the Generative AI Value Chain.pdfExploring Opportunities in the Generative AI Value Chain.pdf
Exploring Opportunities in the Generative AI Value Chain.pdf
 
Generative AI: Past, Present, and Future – A Practitioner's Perspective
Generative AI: Past, Present, and Future – A Practitioner's PerspectiveGenerative AI: Past, Present, and Future – A Practitioner's Perspective
Generative AI: Past, Present, and Future – A Practitioner's Perspective
 
arbar https://www.slideshare.net/Solutionman/charles-caldwell-improve-your-li...
arbar https://www.slideshare.net/Solutionman/charles-caldwell-improve-your-li...arbar https://www.slideshare.net/Solutionman/charles-caldwell-improve-your-li...
arbar https://www.slideshare.net/Solutionman/charles-caldwell-improve-your-li...
 
Responsible Generative AI
Responsible Generative AIResponsible Generative AI
Responsible Generative AI
 
ChatGPT, Generative AI and Microsoft Copilot: Step Into the Future - Geoff Ab...
ChatGPT, Generative AI and Microsoft Copilot: Step Into the Future - Geoff Ab...ChatGPT, Generative AI and Microsoft Copilot: Step Into the Future - Geoff Ab...
ChatGPT, Generative AI and Microsoft Copilot: Step Into the Future - Geoff Ab...
 
An Introduction to Generative AI - May 18, 2023
An Introduction  to Generative AI - May 18, 2023An Introduction  to Generative AI - May 18, 2023
An Introduction to Generative AI - May 18, 2023
 
Thabo Ndlela- Leveraging AI for enhanced Customer Service and Experience
Thabo Ndlela- Leveraging AI for enhanced Customer Service and ExperienceThabo Ndlela- Leveraging AI for enhanced Customer Service and Experience
Thabo Ndlela- Leveraging AI for enhanced Customer Service and Experience
 
AI FOR BUSINESS LEADERS
AI FOR BUSINESS LEADERSAI FOR BUSINESS LEADERS
AI FOR BUSINESS LEADERS
 
Dr. Nassim Belbaly - Decision Markin pai summit 3v2.pdf
Dr. Nassim Belbaly - Decision Markin pai summit 3v2.pdfDr. Nassim Belbaly - Decision Markin pai summit 3v2.pdf
Dr. Nassim Belbaly - Decision Markin pai summit 3v2.pdf
 
The Rise of the LLMs - How I Learned to Stop Worrying & Love the GPT!
The Rise of the LLMs - How I Learned to Stop Worrying & Love the GPT!The Rise of the LLMs - How I Learned to Stop Worrying & Love the GPT!
The Rise of the LLMs - How I Learned to Stop Worrying & Love the GPT!
 
Charles Caldwell - Improve Your Life with AI.pdf
Charles Caldwell - Improve Your Life with AI.pdfCharles Caldwell - Improve Your Life with AI.pdf
Charles Caldwell - Improve Your Life with AI.pdf
 
Content In The Age of AI
Content In The Age of AIContent In The Age of AI
Content In The Age of AI
 

Similar to GenerativeAI and Automation - IEEE ACSOS 2023.pptx

Building successful data science teams
Building successful data science teamsBuilding successful data science teams
Building successful data science teamsVenkatesh Umaashankar
 
AI hype or reality
AI  hype or realityAI  hype or reality
AI hype or realityAwantik Das
 
Bridging the Gap: from Data Science to Production
Bridging the Gap: from Data Science to ProductionBridging the Gap: from Data Science to Production
Bridging the Gap: from Data Science to ProductionFlorian Wilhelm
 
Open, Secure & Transparent AI Pipelines
Open, Secure & Transparent AI PipelinesOpen, Secure & Transparent AI Pipelines
Open, Secure & Transparent AI PipelinesNick Pentreath
 
Machine Learning Infrastructure
Machine Learning InfrastructureMachine Learning Infrastructure
Machine Learning InfrastructureSigOpt
 
Data Scientist By: Professor Lili Saghafi
Data Scientist By: Professor Lili SaghafiData Scientist By: Professor Lili Saghafi
Data Scientist By: Professor Lili SaghafiProfessor Lili Saghafi
 
EVAIN Artificial intelligence and semantic annotation: are you serious about it?
EVAIN Artificial intelligence and semantic annotation: are you serious about it?EVAIN Artificial intelligence and semantic annotation: are you serious about it?
EVAIN Artificial intelligence and semantic annotation: are you serious about it?FIAT/IFTA
 
Data science tools of the trade
Data science tools of the tradeData science tools of the trade
Data science tools of the tradeFangda Wang
 
Online talent sourcing - a future essentia
Online talent sourcing - a future essentiaOnline talent sourcing - a future essentia
Online talent sourcing - a future essentiaHSE Guru
 
Choosing the right IDP Solution
Choosing the right IDP SolutionChoosing the right IDP Solution
Choosing the right IDP SolutionProvectus
 
The Path to Digital Engineering
The Path to Digital EngineeringThe Path to Digital Engineering
The Path to Digital EngineeringElizabeth Steiner
 
Digital Transformation: How to Run Best-in-Class IT Operations in a World of ...
Digital Transformation: How to Run Best-in-Class IT Operations in a World of ...Digital Transformation: How to Run Best-in-Class IT Operations in a World of ...
Digital Transformation: How to Run Best-in-Class IT Operations in a World of ...Precisely
 
Algorithm Marketplace and the new "Algorithm Economy"
Algorithm Marketplace and the new "Algorithm Economy"Algorithm Marketplace and the new "Algorithm Economy"
Algorithm Marketplace and the new "Algorithm Economy"Diego Oppenheimer
 
Introducción al Machine Learning Automático
Introducción al Machine Learning AutomáticoIntroducción al Machine Learning Automático
Introducción al Machine Learning AutomáticoSri Ambati
 
Addressing learning gaps and career oppurtunities after B.Sc computer science
Addressing learning gaps and career oppurtunities after B.Sc computer scienceAddressing learning gaps and career oppurtunities after B.Sc computer science
Addressing learning gaps and career oppurtunities after B.Sc computer sciencesandhya12bansal
 
What Does Artificial Intelligence Have to Do with IT Operations?
What Does Artificial Intelligence Have to Do with IT Operations?What Does Artificial Intelligence Have to Do with IT Operations?
What Does Artificial Intelligence Have to Do with IT Operations?Precisely
 
CIO 101 for Entrepreneurs (2016)
CIO 101 for Entrepreneurs (2016)CIO 101 for Entrepreneurs (2016)
CIO 101 for Entrepreneurs (2016)Michael King
 

Similar to GenerativeAI and Automation - IEEE ACSOS 2023.pptx (20)

Building successful data science teams
Building successful data science teamsBuilding successful data science teams
Building successful data science teams
 
AI hype or reality
AI  hype or realityAI  hype or reality
AI hype or reality
 
Bridging the Gap: from Data Science to Production
Bridging the Gap: from Data Science to ProductionBridging the Gap: from Data Science to Production
Bridging the Gap: from Data Science to Production
 
Open, Secure & Transparent AI Pipelines
Open, Secure & Transparent AI PipelinesOpen, Secure & Transparent AI Pipelines
Open, Secure & Transparent AI Pipelines
 
Machine Learning
Machine LearningMachine Learning
Machine Learning
 
MIS.pptx
MIS.pptxMIS.pptx
MIS.pptx
 
Chatbots: Automated Conversational Model using Machine Learning
Chatbots: Automated Conversational Model using Machine LearningChatbots: Automated Conversational Model using Machine Learning
Chatbots: Automated Conversational Model using Machine Learning
 
Machine Learning Infrastructure
Machine Learning InfrastructureMachine Learning Infrastructure
Machine Learning Infrastructure
 
Data Scientist By: Professor Lili Saghafi
Data Scientist By: Professor Lili SaghafiData Scientist By: Professor Lili Saghafi
Data Scientist By: Professor Lili Saghafi
 
EVAIN Artificial intelligence and semantic annotation: are you serious about it?
EVAIN Artificial intelligence and semantic annotation: are you serious about it?EVAIN Artificial intelligence and semantic annotation: are you serious about it?
EVAIN Artificial intelligence and semantic annotation: are you serious about it?
 
Data science tools of the trade
Data science tools of the tradeData science tools of the trade
Data science tools of the trade
 
Online talent sourcing - a future essentia
Online talent sourcing - a future essentiaOnline talent sourcing - a future essentia
Online talent sourcing - a future essentia
 
Choosing the right IDP Solution
Choosing the right IDP SolutionChoosing the right IDP Solution
Choosing the right IDP Solution
 
The Path to Digital Engineering
The Path to Digital EngineeringThe Path to Digital Engineering
The Path to Digital Engineering
 
Digital Transformation: How to Run Best-in-Class IT Operations in a World of ...
Digital Transformation: How to Run Best-in-Class IT Operations in a World of ...Digital Transformation: How to Run Best-in-Class IT Operations in a World of ...
Digital Transformation: How to Run Best-in-Class IT Operations in a World of ...
 
Algorithm Marketplace and the new "Algorithm Economy"
Algorithm Marketplace and the new "Algorithm Economy"Algorithm Marketplace and the new "Algorithm Economy"
Algorithm Marketplace and the new "Algorithm Economy"
 
Introducción al Machine Learning Automático
Introducción al Machine Learning AutomáticoIntroducción al Machine Learning Automático
Introducción al Machine Learning Automático
 
Addressing learning gaps and career oppurtunities after B.Sc computer science
Addressing learning gaps and career oppurtunities after B.Sc computer scienceAddressing learning gaps and career oppurtunities after B.Sc computer science
Addressing learning gaps and career oppurtunities after B.Sc computer science
 
What Does Artificial Intelligence Have to Do with IT Operations?
What Does Artificial Intelligence Have to Do with IT Operations?What Does Artificial Intelligence Have to Do with IT Operations?
What Does Artificial Intelligence Have to Do with IT Operations?
 
CIO 101 for Entrepreneurs (2016)
CIO 101 for Entrepreneurs (2016)CIO 101 for Entrepreneurs (2016)
CIO 101 for Entrepreneurs (2016)
 

Recently uploaded

"Federated learning: out of reach no matter how close",Oleksandr Lapshyn
"Federated learning: out of reach no matter how close",Oleksandr Lapshyn"Federated learning: out of reach no matter how close",Oleksandr Lapshyn
"Federated learning: out of reach no matter how close",Oleksandr LapshynFwdays
 
New from BookNet Canada for 2024: BNC BiblioShare - Tech Forum 2024
New from BookNet Canada for 2024: BNC BiblioShare - Tech Forum 2024New from BookNet Canada for 2024: BNC BiblioShare - Tech Forum 2024
New from BookNet Canada for 2024: BNC BiblioShare - Tech Forum 2024BookNet Canada
 
Transcript: New from BookNet Canada for 2024: BNC BiblioShare - Tech Forum 2024
Transcript: New from BookNet Canada for 2024: BNC BiblioShare - Tech Forum 2024Transcript: New from BookNet Canada for 2024: BNC BiblioShare - Tech Forum 2024
Transcript: New from BookNet Canada for 2024: BNC BiblioShare - Tech Forum 2024BookNet Canada
 
Science&tech:THE INFORMATION AGE STS.pdf
Science&tech:THE INFORMATION AGE STS.pdfScience&tech:THE INFORMATION AGE STS.pdf
Science&tech:THE INFORMATION AGE STS.pdfjimielynbastida
 
"LLMs for Python Engineers: Advanced Data Analysis and Semantic Kernel",Oleks...
"LLMs for Python Engineers: Advanced Data Analysis and Semantic Kernel",Oleks..."LLMs for Python Engineers: Advanced Data Analysis and Semantic Kernel",Oleks...
"LLMs for Python Engineers: Advanced Data Analysis and Semantic Kernel",Oleks...Fwdays
 
AI as an Interface for Commercial Buildings
AI as an Interface for Commercial BuildingsAI as an Interface for Commercial Buildings
AI as an Interface for Commercial BuildingsMemoori
 
Human Factors of XR: Using Human Factors to Design XR Systems
Human Factors of XR: Using Human Factors to Design XR SystemsHuman Factors of XR: Using Human Factors to Design XR Systems
Human Factors of XR: Using Human Factors to Design XR SystemsMark Billinghurst
 
Advanced Test Driven-Development @ php[tek] 2024
Advanced Test Driven-Development @ php[tek] 2024Advanced Test Driven-Development @ php[tek] 2024
Advanced Test Driven-Development @ php[tek] 2024Scott Keck-Warren
 
SQL Database Design For Developers at php[tek] 2024
SQL Database Design For Developers at php[tek] 2024SQL Database Design For Developers at php[tek] 2024
SQL Database Design For Developers at php[tek] 2024Scott Keck-Warren
 
Are Multi-Cloud and Serverless Good or Bad?
Are Multi-Cloud and Serverless Good or Bad?Are Multi-Cloud and Serverless Good or Bad?
Are Multi-Cloud and Serverless Good or Bad?Mattias Andersson
 
Snow Chain-Integrated Tire for a Safe Drive on Winter Roads
Snow Chain-Integrated Tire for a Safe Drive on Winter RoadsSnow Chain-Integrated Tire for a Safe Drive on Winter Roads
Snow Chain-Integrated Tire for a Safe Drive on Winter RoadsHyundai Motor Group
 
Artificial intelligence in the post-deep learning era
Artificial intelligence in the post-deep learning eraArtificial intelligence in the post-deep learning era
Artificial intelligence in the post-deep learning eraDeakin University
 
CloudStudio User manual (basic edition):
CloudStudio User manual (basic edition):CloudStudio User manual (basic edition):
CloudStudio User manual (basic edition):comworks
 
APIForce Zurich 5 April Automation LPDG
APIForce Zurich 5 April  Automation LPDGAPIForce Zurich 5 April  Automation LPDG
APIForce Zurich 5 April Automation LPDGMarianaLemus7
 
costume and set research powerpoint presentation
costume and set research powerpoint presentationcostume and set research powerpoint presentation
costume and set research powerpoint presentationphoebematthew05
 
Making_way_through_DLL_hollowing_inspite_of_CFG_by_Debjeet Banerjee.pptx
Making_way_through_DLL_hollowing_inspite_of_CFG_by_Debjeet Banerjee.pptxMaking_way_through_DLL_hollowing_inspite_of_CFG_by_Debjeet Banerjee.pptx
Making_way_through_DLL_hollowing_inspite_of_CFG_by_Debjeet Banerjee.pptxnull - The Open Security Community
 
SIEMENS: RAPUNZEL – A Tale About Knowledge Graph
SIEMENS: RAPUNZEL – A Tale About Knowledge GraphSIEMENS: RAPUNZEL – A Tale About Knowledge Graph
SIEMENS: RAPUNZEL – A Tale About Knowledge GraphNeo4j
 
My INSURER PTE LTD - Insurtech Innovation Award 2024
My INSURER PTE LTD - Insurtech Innovation Award 2024My INSURER PTE LTD - Insurtech Innovation Award 2024
My INSURER PTE LTD - Insurtech Innovation Award 2024The Digital Insurer
 

Recently uploaded (20)

"Federated learning: out of reach no matter how close",Oleksandr Lapshyn
"Federated learning: out of reach no matter how close",Oleksandr Lapshyn"Federated learning: out of reach no matter how close",Oleksandr Lapshyn
"Federated learning: out of reach no matter how close",Oleksandr Lapshyn
 
New from BookNet Canada for 2024: BNC BiblioShare - Tech Forum 2024
New from BookNet Canada for 2024: BNC BiblioShare - Tech Forum 2024New from BookNet Canada for 2024: BNC BiblioShare - Tech Forum 2024
New from BookNet Canada for 2024: BNC BiblioShare - Tech Forum 2024
 
Transcript: New from BookNet Canada for 2024: BNC BiblioShare - Tech Forum 2024
Transcript: New from BookNet Canada for 2024: BNC BiblioShare - Tech Forum 2024Transcript: New from BookNet Canada for 2024: BNC BiblioShare - Tech Forum 2024
Transcript: New from BookNet Canada for 2024: BNC BiblioShare - Tech Forum 2024
 
Science&tech:THE INFORMATION AGE STS.pdf
Science&tech:THE INFORMATION AGE STS.pdfScience&tech:THE INFORMATION AGE STS.pdf
Science&tech:THE INFORMATION AGE STS.pdf
 
"LLMs for Python Engineers: Advanced Data Analysis and Semantic Kernel",Oleks...
"LLMs for Python Engineers: Advanced Data Analysis and Semantic Kernel",Oleks..."LLMs for Python Engineers: Advanced Data Analysis and Semantic Kernel",Oleks...
"LLMs for Python Engineers: Advanced Data Analysis and Semantic Kernel",Oleks...
 
AI as an Interface for Commercial Buildings
AI as an Interface for Commercial BuildingsAI as an Interface for Commercial Buildings
AI as an Interface for Commercial Buildings
 
Human Factors of XR: Using Human Factors to Design XR Systems
Human Factors of XR: Using Human Factors to Design XR SystemsHuman Factors of XR: Using Human Factors to Design XR Systems
Human Factors of XR: Using Human Factors to Design XR Systems
 
Advanced Test Driven-Development @ php[tek] 2024
Advanced Test Driven-Development @ php[tek] 2024Advanced Test Driven-Development @ php[tek] 2024
Advanced Test Driven-Development @ php[tek] 2024
 
SQL Database Design For Developers at php[tek] 2024
SQL Database Design For Developers at php[tek] 2024SQL Database Design For Developers at php[tek] 2024
SQL Database Design For Developers at php[tek] 2024
 
Are Multi-Cloud and Serverless Good or Bad?
Are Multi-Cloud and Serverless Good or Bad?Are Multi-Cloud and Serverless Good or Bad?
Are Multi-Cloud and Serverless Good or Bad?
 
Snow Chain-Integrated Tire for a Safe Drive on Winter Roads
Snow Chain-Integrated Tire for a Safe Drive on Winter RoadsSnow Chain-Integrated Tire for a Safe Drive on Winter Roads
Snow Chain-Integrated Tire for a Safe Drive on Winter Roads
 
Artificial intelligence in the post-deep learning era
Artificial intelligence in the post-deep learning eraArtificial intelligence in the post-deep learning era
Artificial intelligence in the post-deep learning era
 
CloudStudio User manual (basic edition):
CloudStudio User manual (basic edition):CloudStudio User manual (basic edition):
CloudStudio User manual (basic edition):
 
Vulnerability_Management_GRC_by Sohang Sengupta.pptx
Vulnerability_Management_GRC_by Sohang Sengupta.pptxVulnerability_Management_GRC_by Sohang Sengupta.pptx
Vulnerability_Management_GRC_by Sohang Sengupta.pptx
 
APIForce Zurich 5 April Automation LPDG
APIForce Zurich 5 April  Automation LPDGAPIForce Zurich 5 April  Automation LPDG
APIForce Zurich 5 April Automation LPDG
 
costume and set research powerpoint presentation
costume and set research powerpoint presentationcostume and set research powerpoint presentation
costume and set research powerpoint presentation
 
Making_way_through_DLL_hollowing_inspite_of_CFG_by_Debjeet Banerjee.pptx
Making_way_through_DLL_hollowing_inspite_of_CFG_by_Debjeet Banerjee.pptxMaking_way_through_DLL_hollowing_inspite_of_CFG_by_Debjeet Banerjee.pptx
Making_way_through_DLL_hollowing_inspite_of_CFG_by_Debjeet Banerjee.pptx
 
SIEMENS: RAPUNZEL – A Tale About Knowledge Graph
SIEMENS: RAPUNZEL – A Tale About Knowledge GraphSIEMENS: RAPUNZEL – A Tale About Knowledge Graph
SIEMENS: RAPUNZEL – A Tale About Knowledge Graph
 
DMCC Future of Trade Web3 - Special Edition
DMCC Future of Trade Web3 - Special EditionDMCC Future of Trade Web3 - Special Edition
DMCC Future of Trade Web3 - Special Edition
 
My INSURER PTE LTD - Insurtech Innovation Award 2024
My INSURER PTE LTD - Insurtech Innovation Award 2024My INSURER PTE LTD - Insurtech Innovation Award 2024
My INSURER PTE LTD - Insurtech Innovation Award 2024
 

GenerativeAI and Automation - IEEE ACSOS 2023.pptx

  • 1. The professional home for the engineering and technology community worldwide ieee.org The professional home for the engineering and technology community worldwide Approaches in using Generative AI in Automation Solutions IEEE International Conference on Autonomic Computing and Self-Organizing Systems (ACSOS) 2023 Allen Chan, IBM Distinguished Engineer and CTO, IBM Business Automation 1
  • 2. The professional home for the engineering and technology community worldwide 2 ieee.org The professional home for the engineering and technology community worldwide Allen Chan 2 Co-Author of 2 books on Automation and AI Recent collaboration with University of Toronto (Prof. Eric Yu) and York University (Prof. Alex Sendervich) on AI Research 1.Transformer Models for Activity Mining in Knowledge-Intensive Processes. Feb 2023. Book: Business Process Management Workshops 2.Managing and Simplifying Cognitive Business Operations Using Process Architecture Models. May 2019. Book: Advanced Information Systems Engineering 3.Solution Patterns for Machine Learning. May 2019. Book: Advanced Information System Engineering. 4.Modeling and Analyzing Process Architecture for Context-Driven Adaptation: Designing Cognitively-Enhanced Business Processes for Enterprises. Oct 2018. 2018 IEEE 22nd International Enterprise Distributed Object Computing Conference (EDOC) 5.Designing Process Architectures for User Engagement with Enterprise Cognitive Systems. Oct 2017. IFIP Working Conference on The Practice of Enterprise Modeling
  • 3. The professional home for the engineering and technology community worldwide Agenda 3 ▸Generative AI is not Enough ▸Content Use Cases ▸Decisions Use Cases ▸Workflow Use Cases ▸Summary ▸Q&A
  • 4. The professional home for the engineering and technology community worldwide 1 Generate ▸ Generate content based on request, e.g., write an email, draft a proposal 2 Classify ▸ Read and classify input based on few examples. Can be used to categorize customer issues or group inputs. 3 Summarize ▸ Transform dense text into your personalized overview, capturing key points. 4 Extract ▸ Pull the information you want from large documents. Identify named entities, parse terms and conditions, etc. 4 Generative AI Core Capabilities
  • 5. The professional home for the engineering and technology community worldwide GenerativeAI alone is not enough 5 Contrary to popular belief, Generative AI does not: 1. Think 2. Reason 3. Check for Facts 4. Coordinate Tasks 5. Perform Real Work => To do that, we need to combine Generative AI with Automations
  • 6. The professional home for the engineering and technology community worldwide Cost Enterprise challenges with using Gen AI ▸ Training on LLM is expensive and requires hundreds or thousands of GPU hours. Simple approaches can easily lead to token explosion and exponential cost. ▸ LLM is designed to process “language” data, not Excel, not PDF, not JSON, not Powerpoint, not photos… so all information will need to be “normalized” before it can be stored in LLM. ▸ Many documents in an enterprise content repository have access control and contain sensitive and private data. LLMs have no access control or data governance. Text Only 6 Access Control
  • 7. The professional home for the engineering and technology community worldwide Sensitive Data Enterprise challenges with using Gen AI ▸ Since LLMs do not forget, guardrails must be provided to ensure sensitive and private data are not stored intentionally or unintentionally in the model. ▸ We want the model to answer questions based on facts and not making it up (“hallucinating”). In addition, we must be able to trace the answers back to the original documents. ▸ Different LLMs have different performance and quality. They use different algorithms, different training data, different weighting on data, and different pre- and post-processing. The quality of the answers will rely on picking the correct LLM for the problem domain. Trust & Traceability 7 Prediction Quality
  • 8. The professional home for the engineering and technology community worldwide Use Cases Content ▸ Provide answers to questions with LLMs Decisions ▸ Add reasoning and decisions to LLMs Workflow ▸ Automate work with LLMs 8
  • 9. The professional home for the engineering and technology community worldwide 1 ▸ Feed documents directly into fine- tuning 2 ▸ Leverage ECM Built- in Search 3 ▸ Refine inputs with knowledge pre- processing 4 ▸ Adopt a VectorDB for rich semantic search 9 Content – Provide answers to questions with LLMs
  • 10. The professional home for the engineering and technology community worldwide 10 Pros • The approach is straightforward and easy to implement • Allows users to get answers directly from LLM Cons • Takes a lot of time and money to fine-tune the model with millions of documents • Not all documents are textual and can be consumed as-is… so any answer could be incomplete. • No role-based access control means all data used in fine- tuning will be available to anyone that is asking the question • Mixing enterprise data with public-domain foundational data can easily lead to hallucination Content – Feed documents directly into fine-tuning 1. Take documents in Document Store use them directly in Fine- Tuning. 2. Build a Q&A system to just the question through the prompt.
  • 11. The professional home for the engineering and technology community worldwide 11 Pros • The approach is straightforward and easy to implement • Get answers directly from the information source – chances of hallucination is greatly limited. • The token count is limited only to information passed to the model. • Role-based access control can be applied Cons • We can only find documents with matching search keywords • Additional processing would be required to limit the number of matching results as there might be too many tokens. • Given the result is the entire document, it would be difficult to extract the precise relevant sections. Content – Leverage Document Store Built-in Search 1. Perform a search first and retrieve the documents matching the search criteria. 2. Feed the document as part of the prompt and ask the LLM to limit the answers to only information in the prompt.
  • 12. The professional home for the engineering and technology community worldwide 12 Pros • In addition to having the same pros as the previous approach, the additional knowledge pre-processor allows more document types to be passed to the LLMs • By removing sensitive data ahead of the prompt, we reduce the risk of sensitive data being stored in the LLMs. Cons • We can only find documents with matching search keywords • Additional processing would be required to limit the number of matching results as there might be too many tokens. • Given the result is the entire document, it would be difficult to extract the precise relevant sections. Content – Leverage Knowledge Pre-processor 1. Leverage existing FileNet Search 2. Inject a Knowledge Pre-processor before we generate the prompt to perform OCR and other information filtering to remove sensitive and personal data.
  • 13. The professional home for the engineering and technology community worldwide 13 Pros • Use of VectorDB allows use of document fragments in search results, thus reducing the token counts and increases the result relevance. • Knowledge aggregator/formatter can combine information from multiple data sources to create a single answer. Cons • Having to pre-process all the documents is a time- consuming process. Content – Adopt Retrieval Augmentation Generation (RAG) approach for semantic search 1. Leverage a VectorDB to shard, index and perform semantic search on documents 2. Include a knowledge aggregator/formatter to combine results from LLMs and original data to created complete answers.
  • 14. The professional home for the engineering and technology community worldwide 1 ▸ NLU + Reasoning 2 ▸ Reasoning followed by NLG 3 ▸ Generate business rules from business policies 14 Decisions – Add reasoning and decisions to LLMs
  • 15. The professional home for the engineering and technology community worldwide 15 Pros • Straightforward implementation requiring minimal or no fine tuning. • Allow rules to be used in non-traditional API-based implementation e.g., in a chatbot. Cons • Limited to identified text extraction capabilities. • In cases where the extraction process does not find the expected data from the input text, it’s crucial to set appropriate guardrails to handle such scenarios during the reasoning phase. Decisions – NLU + Reasoning 1. Leverage NLU capabilities of LLMs to extract business entities (e.g., city name, job role) and pass those as inputs to Decisions.
  • 16. The professional home for the engineering and technology community worldwide 16 Pros • Straightforward implementation requiring minimal or no fine tuning. • Allow rules to be used in non-traditional API-based implementation e.g., in a chatbot. Cons • Prompt will need to be prepared carefully to ensure generated outputs reflect the intent of the decision outcomes. Decisions – Reasoning followed by NLG 1. Leverage NLU capabilities of LLMs to extract business entities (e.g., city name, job role) and pass those as inputs to Decisions.
  • 17. The professional home for the engineering and technology community worldwide 17 Pros • Leveraging the LLM as a knowledge extraction tool enables the extraction of rules and an underlying data ontology. The extracted rules, once reviewed by a human, can automate decisions with traceability and determinism. Cons • Depending on the business domains, further fine- turning will be required. • Require a good amount of prompt engineering to ensure the generated rules model is syntactically and semantically correct. • Generate models will require require by SME and still need to be tested/validated. Decisions – Generate business rules from business policies 1. Extract automation assets, including business rules, data models, and signatures, from plain text business policies. 2. Use the generated rules models as starting point to create business decisions.
  • 18. The professional home for the engineering and technology community worldwide 1 ▸ Classify inputs and perform requests 2 ▸ Generate responses as part of a business processes 3 ▸ Automate LLM finetuning and exception handling 4 ▸ Generate workflow from description 18 Workflow – Automate work with LLMs
  • 19. The professional home for the engineering and technology community worldwide 19 Pros • Straightforward implementation requiring minimal or no fine-tuning. • Allow workflow to work directly with natural language inputs without pre-processing Cons • Limited to identified text extraction capabilities. • In cases where the extraction process does not find the expected data from the input text, it’s crucial to set appropriate guardrails to handle such scenarios during the reasoning phase. Workflow – Classify/Extract inputs and perform work Leverage LLM to classify and/or extract input data and use those information to perform work If LLM fails to classify the information, we can delegate to a human user to decide and feed the decision back to the system.
  • 20. The professional home for the engineering and technology community worldwide 20 Pros • Support a more intuitive engagement experience with users. • Support other non-API based interaction, e.g., with smart assistants like Alexa, voice-based interaction with text-to-speech Cons • Prompt will need to be prepared carefully to ensure generated outputs reflect the intent of the work outcomes. Workflow – Generate responses as part of a business processes Use LLM to generate responses (e.g., email messages, work summary, task notification) directly from business processes data.
  • 21. The professional home for the engineering and technology community worldwide 21 Pros • This is a must-have for any fine-tuning or indexing exercise that uses enterprise data. Cons • Incur additional review and approval cost. Workflow – Automate LLM finetuning and exception handling Use workflow to provide Human-in-the-loop review and approval before knowledge data will be fed to LLM for fine tuning.
  • 22. The professional home for the engineering and technology community worldwide 22 Pros ▸ Provide a natural language approach to generate business processes. ▸ The quality of the generated process will rely on the quality of the LLMs. ▸ It is also possible to ask the LLM to generate the process from just a simple business goal, but the result will rely on the foundational data that was used to do the initial model training. Cons ▸ Do not expect to get a fully working BPMN model, work such as defining the data types, decision logic, user interfaces and validating the process will be required. Workflow – Generate Workflow from Description From a high level description of a business process, generate a BPMN process that can be used as a starting point for an automation workflow solution. Due to different generation quality of the LLM models, it is better to use LLM to summarize or extract procedures from the business process document and use a post-processor to take the output and turn that into a formal BPMN document.
  • 23. The professional home for the engineering and technology community worldwide 1 ▸ Gen AI, when used appropriately, can speed up certain data processing. 2 ▸ Combine Gen AI and Automations can accelerate your digital transformation journey 3 ▸ Careful planning and consideration for private and sensitive data is a must for enterprise usage. 4 ▸ LLM models are evolving very fast and each day there are new models. Understanding which models to use for the purpose at hand is the key to yield good results. 23 Summary