GDG Cloud Southlake #28: Brad Taylor and Shawn Augenstein: Old Problems in the New Frontiers of AI
• Brad discusses how decades-old laws and expanding regulation have new implications in the ML and Large Model age, and will touch on:
• Legal and Regulatory: Data usage rights, cautionary tale of stability.ai and Getty Images, EU's planned expansion of GDPR re models
• How Neural Networks, zero and one-shot learning, and LLMs have increased the need for better data governance, lineage management
• Shawn speaks on the coming "Data Renaissance"
• The New IP: Prompts and Internatl Interaction Data
• Where GenAI can be used right now and where it maybe shouldn't be used yet
• The Power of the Diversity of Insight
• What is making the future look bright!
Brad has been an intrapreneur and entrepreneur in data, AI, and IoT and has led teams in the creation of NLP, data products and predictive analytics for retention, churn, driver safety, traffic, CX and fleet risk. He has built solutions on global hyperscalers GCP, AWS, Azure, and IBM. Brad is a former founding partner at Tech Wildcatters, and worked with dozens of mobile, SaaS and AI start-ups, many of which became both job creators and profitable exits for TW investors. He is currently a Senior Manager in Pepsico's global Strategy and Transformation group, where he focuses on delivering AI/ML driven solutions.
Shawn Augenstein is a dynamic and highly experienced professional, who is driven by educating, providing equal access to technology and equitable access to information. Currently, Shawn serves as Principal Data & AI Consultant at CDW, where he develops the curriculum and architectures for understanding and furthering the use of AI, as well as developing solutions for both partners and clients. In his spare time, he enjoys exploring new frontiers of Diffusers, capturing moments through photography, and listening to music as a passionate melophile.
Kubernetes & AI - Beauty and the Beast !?! @KCD Istanbul 2024
GDG Cloud Southlake 28 Brad Taylor and Shawn Augenstein Old Problems in the New Frontiers of AI.pdf
1. AUSUM (Latin): a venture,or bold undertaking Confidential
Old Problems in the New Frontier of AI
AI is turning data ownership to gold
LLM and generative’scomplexity, data and
compute needs differentiatethe AI companies
that have the resources to compete
Cloud services, processes and databases will
adapt to support AI and LLM-type training
Generative and LLM’s drive statutory,
contractualand regulatory initiatives
and a need for guard rails (and lawyers)
LLM and AI in general will create
unexpected big winners and big losers
Agile, DevOps, Data Engineering, Data
Science and ML Ops will need to mature to
manage the AI and LLM lifecycle
2. AUSUM (Latin): a venture,or bold undertaking Confidential
Data rights: a cautionary tale…
3. AUSUM (Latin): a venture,or bold undertaking Confidential
And a new opportunity for web companies
4. AUSUM (Latin): a venture,or bold undertaking Confidential
New and distinct LLM solutions
5. AUSUM (Latin): a venture,or bold undertaking Confidential
General & Generative AI – a cloud discipline
Ingestion
Landing
Zone
Data
Sources
Data Acquisition & Preparation Data Lineage | Vector Storage | Modeling
Data
Governance
ADO - Data Engineering ADO - Data Science and Machine Learning
ML Ops
- YAML
- Config
- Repo
- YAML
- Config
- Repo
6. AUSUM (Latin): a venture,or bold undertaking Confidential
But AI will still be remarkably human…
Image from EvilAICartoons.com
7. AUSUM (Latin): a venture,or bold undertaking Confidential
...Bias can emerge in AI systems due to the data they are
trained on… AI systems might inadvertently perpetuate
existing inequalities if not carefully monitored and
addressed.
Moreover, the complexity of AI systems can introduce
risks. If the underlying algorithms are not thoroughly tested
or properly validated, they might produce erroneous
results or unexpected behaviors, which could have serious
consequences in critical domains like healthcare or
autonomous vehicles…By implementing robust validation
processes, conducting bias audits, and adopting
continuous monitoring practices, you can help ensure…
the responsible and beneficial use of AI in our daily lives.
Final thoughts from…
8. DATA, AI & EMPATHY
Data Driven Approaches
Human-centric Ideals
9. 9
HOW FAR AWAY ARE
WE?
Fujitsu has built the K computer, which is one
of the fastest supercomputers in the world. It is
one of the significant attempts at achieving
strong AI. It took nearly 40 minutes to simulate
a single second of neural activity.
Tianhe-2 is a supercomputer that was
developed by the China's National University of
Defense Technology. It holds the record for cps
(calculations per second) at 33.86 petaflops
(quadrillions of cps). Although that sounds
exciting, the human brain is estimated to be
capable of one exaflop, i.e., a billion cps.
11. Actually Happened
In 2018, Amazon scrapped its AI recruiting tool after realizing it was
discriminating against female candidates.
In 2016, a self-driving Tesla car crashed into a truck, killing the
passenger. The accident was caused by the car's failure to recognize
a white truck against a bright sky.
In 2020, an algorithm used to grade high school students’ essays in
the United Kingdom gave higher scores to longer, nonsensical essays,
as long as they contained certain keywords. This led to concerns
about the reliability of automated grading systems.
In 2019, the facial recognition system used by London's Metropolitan
Police was found to be wrong 81% of the time and disproportionately
false-positive for people of color.
These incidents highlight the potential dangers of AI and the
importance of developing ethical frameworks and safeguards to
mitigate these risks.
12. Human-centric design
places human beings at
the center of work and
doesn’t treat them as
secondary components
of the work environment.
The human-centric work
model includes three
dimensions; flexible work
experiences, intentional
collaboration and
empathy-based solutions.
HUMAN-
CENTRIC
APPROACH
13. HUMAN-CENTRIC DELIVERY
Design work models around
human needs, rather than
expecting humans to conform
to legacy practices or locations
that constrain them.
Drive work outcomes by
focusing on three dimensions of
human-centric work: flexible
experiences, intentional
collaboration and empathy-
based management.
Deliver human-centric work
models by soliciting employee
input into the design and
engaging them in the rollout,
especially through co-creation.
Adapt the work model over time
— particularly the next 12 to 18
months — by experimenting,
measuring and continually
optimizing to achieve key
outcomes.
14. • Human-centric work drives superior
outcomes. It is 3.8 times more likelyto have
high employee performance, 3.2 times
more likely to have high intent to stay and
3.1times more likely to have low fatigue
than when human-centricwork attributes
are not present.
• Implementing work models with
employee input and adapting over time
out-perform stop-down static approaches.
• Employees able to provide input to
design and implementation of the future
post pandemic work model are 2.5 times
more likely to have high performance.
HUMAN-CENTRIC
VALUE
17. GEN Z COMING
ON
17
• About half of working Millennials and
Gen Zs are in non-frontline office/desk
roles (52%) and about half (48%) are in
frontline/people-facing roles. More
specifically, working Millennials and Gen
Zs are employed in:
• Office/Desk Positions (52%)
• Student-, Customer-, and Patient-
Facing Positions (26%)
• Manufacturing/Warehouse Positions
(11%)
• Construction Positions (7%)
• Delivery Positions (4%)
18. WHILE STEM, HEALTHCARE, BUILDERS, AND
PROFESSIONAL ELDS CONTINUE TO ADD JOBS,
GENERATIVE AI COULD CHANGE WORK
ACTIVITIES SIGNIFICANTLY FOR MANY
OCCUPATIONS.
18
20. DIVERSITY OF INSIGHT
20
Bias in AI Systems 🔬🌍:
AI systems can
unintentionally amplify
societal biases, leading
to unfair outcomes in
areas such as
sentencing, recruitment,
or voice recognition
systems. Therefore,
diverse insight is crucial in
AI development to
identify and mitigate
these biases.
Tech Industry's Lack of
Diversity🔬🌍: The AI
sector has an
underrepresentation of
women and minorities,
which can limit the
perspectives and
approaches used in AI
development. Greater
diversity can contribute
to more inclusive,
innovative, and effective
AI systems.
Diversityin Leadership 🌍:
Diverseleadership in the
tech industry can foster a
more inclusive
environment,
encouraging diverse
talent and perspectives.
This can lead to more
innovativeproblem-
solving and more
inclusive AI systems.
21. DIVERSITY OF
INSIGHT
21
Impact of Diverse Dat a Sets 🌍: AI
models are only as good as t he dat a
t hey're t rained on. Diverse
development t eams can help ensure
t hat algorit hms are t rained on diverse
dat a set s, resulting in more inclusive
and equit able AI syst ems.
Import ance of Diverse Input in AI
Development 🌍: AI systems
need diverse human input
t hroughout t he development
process t o ensure t hey are
represent at ive and inclusive. This
includes not only diverse dat a
set s but also diverse perspect ives
in t he design and implement ation
of AI syst ems.
Built -inAccount ability 🔬: Implementing
measures such as ensemble methods
and loss functions can help ensure t hat
AI syst ems are diverse by design and
account able for t heir performance. This
can help avoid inherent biases in AI
syst ems.
Diversit y Fuels Innovation 🌍:
Diversit y of insight in AI can lead
t o new and improved ideas,
avoiding an echo chamber of
ideas. Ignoring diversit y can limit
innovat ion and opport unit ies in
t he t ech industry.
M ult icultural Aw areness in AI Use 🌍:
M arket ers, t hough not direct ly
involved in building AI t ools, can bring
mult icultural aw areness to their use of
AI t ools. This can help ensure t hat
t hese t ools are used responsibly and
inclusively.
22. DIVERSITY OF
INSIGHT
22
Stifling Innovation 🔬: Lack of diversity can limit the range
of ideas and approaches in AI development, stifling
innovation in the field.
Limited Consideration of Needs 🔬: AI systems developed
withoutdiverse input may fail to consider the needs of
diverse groups, leading to services that are not inclusive or
equitable.
Biased AI Outcomes 🔬: If AI systems are biased, they can
negatively impact marginalized individuals, influencing
crucial decisions like loan approvals or public services.
23. DIVERSITY OF INSIGHT
23
Perpetuation of Stereotypes
🔬: Without diverse insight,
AI systems can inadvertently
perpetuate harmful
stereotypes and biases.
Missed Opportunities 🔬:
Ignoring diversity in AI is a
missed opportunity to make
AI systems more robust,
innovative, and inclusive.
Lack of Cultural Sensitivity
🔬: Without diverse insight,
AI tools might lack the
cultural sensitivity needed
for global markets,
potentially leading to
cultural missteps or offenses.
24. 24
EQUITY OF
INFORMATION
Equity of Information is an
approach that aims to provide
individuals with convenient access
to technical or business expertise
without the need for extensive
training or high costs. This
approach emphasizes four
important areas which include
application development, data
and analytics, design, and
knowledge. It is also sometimes
known as "citizen access," which
has led to the emergence of
professionals such as citizen data
scientists and citizen programmers.
25. Generative AI can make complex data more accessible to
laypersons by transforming it into simpler, more understandable
formats. Here are some ways it achieves this:
USAGE: EQUITY OF INFORMATION
Data visualization
Generative AI can create easy-to-
understand visual representations of
complex data, such as graphs, charts,
and infographics. These visuals help
convey the main insights and trends in
the data without overwhelming the
reader with technical details.
Natural language generation
AI can convert complex data into human-
readable text by summarizing it or
highlighting key points. This process,
known as natural language generation,
makes it easier for laypersons to
comprehend the information without
needing expertise in the subject matter.
Interactive interfaces
Generative AI can be used to develop
interactive tools that enable users to explore
and manipulate complex data sets. By
allowing users to filter, sort, and customize
their view of the data, these tools make it
easier for them to find relevant insights and
understand the underlying patterns.
1
Personalized content
AI can analyze user preferences and
behavior to tailor content based on
individual needs and interests. This
personalization helps laypersons access
information that is most relevant to them,
making it easier to digest complex data.
Simplification and explanation
Generative AI can break down complex
concepts and relationships within data sets
into simpler explanations. By providing
context and examples, AI can help
laypersons better understand the significance
of the data and its implications.
2 3
4 5
26. 26
ETHICS IN AI
• Empowerment and Supervision of Humans: AI
technologies ought to uphold the autonomy of
individuals and their decision-making abilities. They
should serve to promote democratic values, enhance
societal prosperity and equality, amplify the power of
users, uphold fundamental human rights, and allow for
human supervision.
• Reliability and Security: AI technologies should be
built to withstand various conditions and maintain
security. They should have safety mechanisms,
including contingency plans for unexpected
circumstances, while also guaranteeing accuracy,
dependability, and repeatability.
• Data Confidentiality and Governance: Ensuring full
compliance with privacy and data protection standards
is paramount. It's also crucial to have sufficient data
governance structures in place, considering the data's
quality and integrity, as well as providing legitimate
data access.
• Openness: AI technologies should adhere to the principle of
explain-ability, embodying transparency and facilitating
communication about the components involved: the data, the
system, and business models.
• Inclusivity, Non-Bias, and Justice: This involves eliminating
unjust biases, incorporating accessibility, universal design, and
stakeholder involvement throughout the AI system lifecycle, in
addition to promoting diversity and inclusivity.
• Wellbeing of Society and Environment: AI technologies
should be designed to serve all of humanity, including future
generations, ensuring they are sustainable and do not harm the
environment.
• Responsibility: The need for accountability complements other
requirements and is closely associated with the principle of
fairness. It's necessary to establish mechanisms to ensure
responsibility and accountability for AI technologies and their
outcomes, both pre and post their development,
implementation, and use.
28. THE DATA RENAISSANCE
🧠 Let's break down these terms:
Data: In the context of technology and business, data refers to
distinct pieces of information, usually formatted in a special way
and can exist in a variety of forms, such as numbers, text, bits and
bytes, images, etc. Data is typically the basis for algorithms,
computations, and statistical analysis.
"Data RenAIssance”: a period of significant change and renewed
interest in the field of data. This could imply dramatic
advancements in how data is collected, analyzed, and utilized,
leading to transformative impacts in various fields, such as
business, technology, science, and culture. It denotes a period
where data becomes central to decision making and innovation,
driving a new era of development and opportunities.
RenAIssance: This term literally means "rebirth" in
French. Historically, it refers to the period in European
history marking the transition from the Middle Ages to
modernity, characterized by dramatic changes in art,
science, culture, and thought. In a more general sense, a
renaissance signifies a period of significant change,
rebirth, and renewed interest in a particular area.
29. THE DATA RENAISSANCE
Increased Volume and Diversity of Data: Just as the
Renaissance was characterized by an explosion of
knowledge and ideas, our modern Data Renaissance is
characterized by an exponential increase in the volume,
variety, and velocity of data being generated. This includes
structured and unstructured data from diverse sources like
social media, IoT devices, business transactions, and more.
Democratization of Data: The Renaissance was a period of
increased accessibility of knowledge and learning. In the
Data Renaissance, we're seeing the democratization of
data and AI, where more people have access to data and
the tools to analyze and interpret it. This is leading to a more
data-literate society where data-driven insights are not
confined to a select few data scientists or analysts
Ethical Considerations and Data Governance: The use of
Generative AI and the resulting increase in data collection
and use bring about new ethical and governance
considerations. Questions around data privacy, consent,
and the potential misuse of generatively created content
become significant. Just as the Renaissance led to new
societal norms and legal frameworks, the Data Renaissance
will likely require new norms and regulations for data
governance.
Role of Prompts and Interactions: With Generative AI,
prompts and interactions become a significant part of the
data landscape. They not only serve as inputs for generative
models but also become valuable data points themselves.
These prompts and interactions can provide insights into user
behavior, preferences, and needs, feeding into the data
streams that are crucial for improving AI systems.
Data as a Creative Medium: In the Data Renaissance, driven
by Generative AI, data is not just a source of insight but also
a creative medium. Generative AI models can create new
content, from articles and poems to designs and music,
transforming data into a tool for creativity.
Innovation and Creativity: The Renaissance was a golden
age of innovation and creativity. In the Data Renaissance,
AI and Big Data are driving innovation across sectors, from
healthcare and finance to entertainment and
transportation. All bootstrapped by the advent of
Generative AI!
30. Generative AI, prompts and interaction data hold immense value and can indeed
be treated as a form of Intellectual Property (IP). Protecting this IP requires a
combination of legal, technical, and operational measures.
THE NEW IP
Understanding theNew IP: This includes prompts
created by companies, external interaction data from
clients,and internal interactiondata from employees.
Legal Protection of New IP: This includes prompts
created by companies,external interaction data from
clients,and internal interactiondata from employees.
31. 31
Challenges in Legally Protecting New IP:
1. Unclear Legal Frameworks: Current intellectual property laws
were not designed with AI-generated content in mind, making it
unclear how these laws apply to prompts and interaction data.
2. Ownership and Authorship: It can be complex to determine
ownership and authorship for AI-generated content. For instance,
who owns the IP for a piece of content generated by an AI: the
developer who created the AI, the user who provided the prompt,
or the AI itself?
3. Jurisdictional Issues: As AI and data are often global in nature,
jurisdictional issues can arise. Different countries have different
laws regarding data protection and IP, complicating legal
protection efforts.
THE NEW IP Strategies for Legally Safeguarding Promptsand Interaction
Data:
1. Patents and Copyrights: While the application of
patents and copyrights to AI-generatedcontent is still a
gray area, businesses can potentiallypatent unique
methods of generatingprompts or interaction data.
2. Trade Secrets: If prompts and interaction data are
valuable to a business and give it a competitive
advantage, the business may protect them as trade
secrets. This requires the business to take reasonable
steps to keep the information secret.
3. Contracts and Licenses: Businesses can use contracts
and licenses to control how their prompts and
interaction data are used by third parties or how IP is
handledwhen generated internally. For example, a
business could require employees to agree to terms that
specify how the data can be used and how residuals may
work.
4. Data Protection Laws: Businesses must comply with
data protection laws, like GDPR in the EU, which provide
certain protections for personal data. This is particularly
relevantfor interaction data that includes personal
information.
32. THANK YOU
THE END (IS JUST THE BEGINNING)
Thursday, November 30, 2023