Artificial Intelligence

Zenith Special Interest Group

Executive Primer

TRL

Artificial Intelligence

3 Types of

Artificial Intelligence:

Weak AI

▪Limited by programming –

won’t develop new skills

▪Analyses preferences and

improves over time

Strong AI

▪Learns new skills through

contextualisation

▪Applies knowledge to plan

ahead

▪Can adapt as changes occur

Superintelligence

▪Self-aware

▪Surpasses human

intelligence

▪Only exists in science

fiction

Artificial Intelligence is revolutionizing the financial

technology (fintech) industry, offering unprecedented

opportunities and challenges alike. As a rapidly advancing

branch of computer science, AI enables the simulation of

human thought and language through code, with recent

breakthroughs in machine learning and deep learning

propelling the field's growth.

For fintech companies, embracing AI presents a pathway to

stay competitive amid the increasing investments in AI

innovation by big tech firms. However, alongside these

opportunities, several critical challenges arise, demanding

careful attention.

In this executive briefing, we will explore the key challenges facing the fintech industry as it harnesses the

potential of AI. Specifically, we will examine the issues of bias in AI systems, the ethical implications of AI

adoption, and the importance of addressing security and maintenance concerns for AI systems.

By addressing these challenges, fintech companies can pave the way for widespread and responsible

adoption of AI tools, ensuring a safe, secure, and sustainable transformation of the financial services

industry.

In the following sections, we will delve deeper into each challenge, highlighting the Zenith program's

opportunities to assist fintech and financial services firms in overcoming these hurdles and unlocking the full

potential of AI technology. In our appendices, we will brief you on the landscape of each underlying

technology.

Biases in AI Systems

Key Highlights

▪AI algorithms can

inherit biases from the

data they are trained

on, leading to unfair

and discriminatory

outcomes.

▪Identifying and

mitigating bias in AI

systems is crucial to

building trust and

ensuring ethical AI

adoption.

▪Biases in AI can have

significant

consequences in

sensitive domains such

as finance, impacting

customer satisfaction

and regulatory

compliance.

Artificial Intelligence (AI) systems are prone to inheriting biases present in the data used for their training.

Technologies such as Data Annotation and Data Quality & Observability play vital roles in identifying potential

biases and ensuring that training data is diverse and representative. By incorporating these technologies, fintech

companies can reduce the risk of biased AI outcomes and promote fairness in their algorithms.

By adopting Synthetic Data generating practices, fintech companies

can broaden the training data sets to include more diverse

demographics in their customer data to improve the quality of the data

in.

To ensure ethical AI adoption, fintech companies should also invest in

technologies that relate to Model Validation & Monitoring. These

technologies help in understanding how AI models arrive at decisions,

making the decision-making process more transparent and

interpretable for customers and regulators.

To tackle biases effectively, fintech firms should adopt strategies such as diverse and representative data

collection, regular audits, and fairness-aware machine learning algorithms. Transparency in AI decision-making

can also aid in identifying potential biases and gaining customer trust.

Ethical Implications of AI Adoption

Key Highlights

▪Ethical considerations

in AI adoption involve

privacy, transparency,

fairness, and

accountability.

▪Fintech companies

must balance AI-driven

innovation with

societal and ethical

responsibilities.

▪Responsible AI

practices enhance

brand reputation and

foster long-term

customer loyalty.

As fintech companies embrace AI-driven innovation, they must navigate the ethical implications associated

with its adoption. Ethical considerations involve ensuring customer privacy, maintaining transparency in AI

decision-making, promoting fairness in algorithms, and being accountable for AI-driven outcomes.

To address ethical concerns related to data privacy, fintech companies should implement

Data De-Identification techniques. These technologies protect sensitive customer information while still

allowing AI models to derive valuable insights.

Transparency and explain-ability are vital to address ethical concerns in AI adoption. Consumers and

regulators are increasingly demanding explanations for AI-driven decisions. To enhance transparency, fintech

firms should adopt Interpretable AI models and establish clear communication channels to explain the logic

behind AI recommendations.

Fintech companies should acknowledge their responsibility for AI-driven

outcomes and establish mechanisms to rectify errors and address

customer complaints. Being accountable for AI systems' behaviour

fosters trust and loyalty among customers and stakeholders.

Responsible AI practices not only align with societal values but also

enhance the reputation of fintech companies. Customers are more likely

to engage with companies that demonstrate a commitment to ethical AI

practices, leading to long-term customer loyalty and sustainable growth.

Interpretable AI Explainable AI

Small models that

are inherently

interpretable

e.g. Small decision

trees or linear

models with a small

number of input

variables

The process of

applying a method

that models the

output of a more

complex model

This is performed

after model training

has been

completed

Security & Maintenance Concerns

of Artificial Intelligence

Key Highlights

▪The integration of AI in

fintech increases

cybersecurity risks and

potential

vulnerabilities.

▪Continuous monitoring

and updates are

essential to ensure the

robustness and

reliability of AI

systems.

▪Investing in AI security

and maintenance is a

proactive approach to

safeguarding financial

data and operations.

While AI offers transformative opportunities for fintech, it also introduces new cybersecurity risks and

vulnerabilities. The integration of AI in financial services creates potential entry points for cyber threats,

necessitating a proactive approach to security and maintenance.

Technologies like Resource Optimisation and Machine Learning Deployment are essential in enhancing AI

system robustness. These technologies ensure that AI models are efficiently deployed and managed,

minimizing potential vulnerabilities and optimizing real-time resource usage.

AI can also play a pivotal role in fraud

prevention and risk mitigation within fintech.

Implementing AI-driven Fraud Detection

systems enhances the security of financial

transactions and improves overall risk

management.

Considering the rapid evolution of cyber

threats, fintech companies must stay informed

about emerging security challenges and invest

in state-of-the-art AI security technologies.

Collaborating with industry experts and

sharing insights on security best practices can

further fortify fintech systems against

potential threats.

Zenith Opportunities

Benefits of Exploration

Through Zenith

1. Common frameworks for

adoption can be

crowdsourced

2. Reduce internal expertise

limitations through

community engagement

3. Time-to-market of

exploration in innovation

labs reduced

4. Reserve resources for

high-value development

5. Interoperability improves

adoption across the

industry

6. Keep in step with

security, standards, and

new feature development

7. Benefit from community

contributions across key

themes

8. Empower, retain, and

attract innovators and

developers

Identified

AI Themes

In the Appendix, we have outlined 13 key themes that serve as innovation verticals

within the Zenith program.

These themes encompass cutting-edge technologies and trends in Artificial

Intelligence.

Embracing these themes will drive transformative advancements in fintech, fostering

collaboration, and collective innovation within our foundation.

Our aim is to collaborate on an industry-led, financial services

focused, open-source, learning language model.

Together, we can leverage the power of collective expertise to shape the future of AI-powered fintech

solutions. Join us in our mission to create secure, transparent, and ethical AI applications that redefine the

financial services landscape.

Appendices

Data Sheets

▪AI Chipsets

▪AI-Driven Fraud

Detection

▪Computer Vision

▪Data Annotation

▪Data De-Identification

▪Data Quality &

Observability

▪Generative AI

▪Machine Learning

Platforms & Deployment

▪Model Validation &

Monitoring

▪Natural Language

Processing

▪Resource Optimisation

▪Synthetic Data

▪Version Control &

Experiment Tracking

This report uses Technology Readiness Levels. For more information as to this

classification system, please refer to https://zenith.finos.org/docs/roadmap/TRL 1

456

AI Chipsets

Specialized processors

designed to accelerate AI

computations, enabling faster

and more efficient AI model

training and inference.

In fintech, AI chipsets drive

groundbreaking

advancements, powering

complex algorithms for fraud

detection, risk assessment,

and personalized financial

recommendations.

The high-performance

computing capabilities of AI

chipsets empower fintech

companies to deliver real-

time, data-intensive services,

transforming the way

financial institutions operate

and serve their customers.

You can find out more about

this subject in our AI Chipset

Primer on the Zenith GitHub.

Surging Demand

for applications

Deep learning

advancements

Need for

fast/efficient data

processing

Regulatory pressure to

improve fraud detection

Cost savings through AI-

driven automation

Integration across multiple

processes

AI-powered data analytics

Desire to enhance financial

inclusivity & accessibility

Legacy hardware

manufacturers

Traditional financial services

requiring adaptation

Skill shifts in job market as AI

impacts workforce

Faster & more accurate decision making

Risk mgmt. & fraud detection capabilities

Real-time data processing for insights

Optimised AI model training on chipsets

AI-powers Fintech expands market reach

Increased investment for innovative startups

Data storage and integration chipsets

Collaboration with data/chipset manufacturers

Disruption to trad. hardware supply chains

Success of

accelerated AI

model training

Scalability of AI

chipsets for high

volume data

Need for continuous upskilling to ensure

workforce proficiency

Growing ecosystem

of service providers

Availability of dev

tools & frameworks

Need for continuous upskilling to ensure

chipsets are implemented correctly

Data privacy

concerns

Integration

challenges with

legacy hardware

Regulatory compliance complexities

Short term

Initial adoption and

integration of chipsets

Medium term

Widespread deployment

in financial services

Long term

Evolution of chipsets for

autonomous solutions

Potential delays in

advancements and

innovations

Market resistance and

scepticism in AI-decision

making

Ethical considerations

surrounding responsible

use

AI chipsets offer a compelling opportunity for fintech, revolutionizing financial services through accelerated AI computations. Advancements in specialized hardware enable groundbreaking

applications in fraud detection, risk assessment, and personalized financial recommendations. The high-performance computing capabilities of AI chipsets empower real-time, data-intensive

financial services, transforming traditional banking and ushering in a new era of customer-centric financial experiences. Despite technical feasibility and potential financial benefits, challenges

like data privacy, regulatory compliance, and integration frictions require careful navigation. With a promising future outlook, early adoption and collaboration between fintech players and AI

chipset manufacturers will play a crucial role in shaping the trajectory of AI-driven fintech innovation.

AI-Driven

Fraud Detection

This leverages advanced

machine learning algorithms

to detect and prevent

fraudulent activities in real-

time.

In fintech, this technology

acts as a vigilant security

layer, continuously analysing

vast volumes of financial data

to identify suspicious

patterns and transactions.

By swiftly detecting and

mitigating fraud, AI-driven

systems protect financial

assets, preserve customer

trust, and enhance overall

cybersecurity in the rapidly

evolving digital financial

landscape.

You can find out more about

this subject in our AI Chipset

Primer on the Zenith GitHub.

Escalating cyber

threats

AI advancements

increase accuracy

& efficiency

Growing adoption

of digital financial

services

Regulatory mandates &

compliance reqs.

Significant cost

saving through

loss prevention

Brand reputation

& Consumer trust

Widespread adoption leads

to collective intelligence

Collaborative Industry-wide

defence strategy

Increased trust in FS

promotes further adoption

Cybercriminals facing more

evolved systems

Legacy fraud detection

losing competitiveness

Shift in the dynamics of

financial crime investigation

toward automation

Superior fraud detection & reduced losses

Enhanced customer trust & loyalty

Real-time response capabilities to incidents

Cost savings & less post-fraud investigations

Customer retention through reputation

Investments & partnerships for potential

applications following successful tests

Integration of AI-driven solutions to core infra

Collaboration between AI solution providers,

data aggregators & financial entities

Potential changes in risk assessment and

underwriting of Financial Services

Demonstrable

success in real world

scenarios

Scalability of AI

models to process

vast volumes of data

Continuous improvement and refinement of ML

models to adapt to evolving threat landscape

Readily available

dev. frameworks

Growing ecosystem

of talent working

with AI tools

Need for continuous upskilling to keep up with

emerging fraud techniques

Data privacy

concerns

Over-reliance on AI

models leading to

blind spots

Regulatory compliance complexities in

explainability of models

Short term

Integration efforts into

security infrastructure

Medium term

Industry-wide adoption &

increased reliance

Long term

Evolution of AI-driven

responses to emerging

fraud techniques

False negatives where AI

fails to detect new or

adaptive fraud patterns

Adversarial attacks

targeting AI models to

manipulate outcomes

Potential overfitting or

bias in AI models

affecting accuracy and/or

fairness of detection

AI-driven fraud detection represents a significant opportunity in fintech, providing real-time protection against sophisticated cyber threats. Advanced machine learning algorithms analyse vast

financial data to swiftly identify and prevent fraudulent activities, safeguarding financial assets and customer trust. Collaboration between fintech companies, financial institutions, and

cybersecurity experts is driving the development and adoption of robust fraud prevention solutions. The technology's impact is far-reaching, with macro network effects and improved

cybersecurity across the digital financial landscape. While the potential for financial benefits and competitive advantages is substantial, the implementation of AI-driven fraud detection requires

careful consideration of technical feasibility, frictions, and risks. By striking the right balance between innovation and responsible use, AI-driven fraud detection will continue to transform the

way financial entities combat financial crime, contributing to a more secure and trusted financial ecosystem.

Computer Vision

An AI technology that

enables machines to

interpret and understand

visual information.

In fintech, computer vision

revolutionizes various

processes, from automating

document verification and

identity recognition to

analysing financial charts and

visualizing data patterns.

By harnessing the power of

computer vision, fintech

companies streamline

operations, enhance user

experiences, and unlock

valuable insights from visual

data, driving efficiency and

innovation.

You can find out more about

this subject in our AI Chipset

Primer on the Zenith GitHub.

Increasing

availability of HQ

visual data

Growing demand

for automation &

efficiency

Rising interest in

data-driven FS

decision making

Competitive advantage

through visualising

complex data

Potential

improvements to

fraud detection

Need for fast &

accurate ID & docs

verification

More focus on visual

interaction with data

Combination of computer

vision with other AI tech

Improved accessibility &

inclusivity via automation

Obsolescence creates need

to adapt to automation

Conventional data analysis

methods challenged

Shift in the skill set required

within FS Operational tasks

Quick & seamless document processing

Enhanced security through adv. verification

Data-driven decision making for strategy

Automation of labour intensive doc verification

Higher customer engagement & satisfaction

Demonstration of technological leadership &

innovation of products with customer focus

Integration of end-to-end operations

Collaboration required to meet specific

industry needs

Streamlining of data collection & analysis

Real-world demos of

visual data

interpretation

Scalability to handle

large-scale real-time

data

Continuous improvement and refinement

required to improve accuracy & efficiency

Readily available

libraries, APIs and

tools

Growing ecosystem

of collaborating

entities

Need for continuous upskilling to keep up with

new data use cases

Data privacy

concerns over

processing

Ethical

considerations when

decision making

Potential biases affecting financial data analysis

& decision outcomes

Short term

Document verification

and basic data analysis

Medium term

Widespread POCs for

other FS operations

Long term

Augmented Reality based

financial interactions

Technical challenges

affecting decision making

Regulatory compliance in

ID verification &

customer data processing

Market resistance to

automated interpretation

of visual data

Computer vision technology holds immense potential in transforming fintech processes, enabling machines to interpret and understand visual information. Fintech companies are leveraging

computer vision to automate document verification, enhance fraud detection, and gain valuable insights from complex financial data. The technology's impact extends to improved user

experiences, data-driven decision-making, and increased efficiency in financial operations. While the opportunity is significant, challenges related to technical feasibility, ethical considerations,

and regulatory compliance must be addressed. The future outlook for computer vision in fintech is promising, with a gradual timeline for adoption and a need for careful risk management and

responsible implementation. By capitalizing on the benefits and addressing frictions, computer vision will play a pivotal role in driving innovation and efficiency in the evolving landscape of

fintech services.

Data Annotation

This involves labelling and

tagging data to train AI

models, providing the

foundation for robust and

accurate machine learning

applications in fintech.

High-quality annotated data

ensures the development of

AI models that accurately

recognize patterns, make

precise predictions, and drive

data-driven decision-making.

In fintech, data annotation

underpins critical processes

such as credit risk

assessment, fraud detection,

and personalized financial

recommendations.

Exponential

growth of data &

labelling

Advanced machine

learning relying on

HQ annotations

Increased demand

for personalised

financial services

Fintech investment into

data annotation services

and tools

Collaboration

required to create

HQ labelled data

Numerous

startups and

service providers

Already widespread

adoption in FinTech sector

Pre-existing labelled dataset

sharing in FinTech

Models gaining accuracy,

accessibility & inclusivity

Manual data labelling

processes facing

obsolescence

Traditional methods of risk

and fraud analysis being

challenged

Shifting job roles within

fintech with demand for data

professionals/experts

Precise recommendations giving competitive

edge for financial transactions

Enhanced customer trust & loyalty from higher

accuracy and relevant predictions

Cost savings through data annotation

automation & streamlining model development

Increased revenue through tailored services

Investments & partnerships from mutual

stakeholders impressed by HQ data

Reduction of financial losses from better risk &

fraud prevention

Integration to core infrastructure

of platforms & models

Industrial collaboration with annotation service

providers to ensure high quality datasets

Streamlining of data collection & AI model

creation through efficient data enrichment

Pre-existing success

of data annotation

for machine learning

Scalability of data

annotation to handle

larger volumes

Ongoing advancements needed in annotation

techniques to address evolving needs

High availability of

Fintech specific

tools & platforms

Growing ecosystem

of collaborating

experts

Need for upskilling of proficient data

annotators in FinTech-specific requirements

Privacy & security

when annotating

sensitive data

Regulatory

complexities in use

for risk assessment

Ethical considerations of using annotated data

for Decision Making

Short term

Identification of specific

use-cases for adoption

Medium term

Projections for wide-

spread integration of high

quality data in operations

Long term

Create evolution strategy

for data annotation

Risk of flawed

annotations impacting

machine learning models

Explanation of annotation

methods & responsible

use to regulators

Market resistance

(ethical) to automated

data annotation

Data annotation is a fundamental technology in fintech, providing the foundation for robust and accurate AI-driven financial services. Fintech companies leverage data annotation to improve

credit risk assessment, fraud detection, and personalized financial recommendations. The technology's impact extends to enhanced customer trust, competitive advantages, and cost savings.

While the opportunity is substantial, challenges related to technical feasibility, data privacy, and regulatory compliance must be addressed. The future outlook for data annotation in fintech is

promising, with a gradual timeline for adoption and a need for careful risk management and responsible implementation. By capitalizing on the benefits and addressing frictions, data annotation

will continue to play a pivotal role in driving innovation and efficiency in the evolving landscape of fintech services.

Data

De-Identification

A privacy-enhancing

technique that removes or

obscures personally

identifiable information from

data sets.

In fintech, where data

security and privacy are

paramount, data de-

identification enables the

safe and compliant use of

sensitive financial

information for AI research

and analysis.

By protecting customer

identities while allowing for

data-driven insights, fintech

companies can responsibly

leverage AI technology to

innovate and improve

financial services.

Increasing

regulation on data

privacy & PII

Advances enable

safe use of

sensitive data

Need to leverage

AI while working

to strict rules

Necessary to balance

innovation & privacy

Demonstrating

commitments to

data privacy

GDPR & CCPA

regulations

Data de-identification

practices becoming more

common in industry

Ongoing collaboration to

share techniques

Higher interest in sharing &

responsible data use

Outdated practices falling

behind regulation

Legacy methods challenged

by privacy-aware clients

Evolving job market in

fintech being driven by

demand

Competitive edge through data responsibility

Customer trust focused on reputation

Compliance adherence reducing legal exposure

Reduced impact of data breaches

Leveraging de-identified data for AI research &

analytics in applications

Attraction of investors impressed by

responsible AI use

Integration to core infrastructure

of platforms & models

Partnership models to maintain transparency &

consistency

Secure collaboration strengthening data

privacy standards in the chain

Demonstrated

success of de-

identification

Scalability of data

annotation to handle

larger volumes

Ongoing advancements needed in de-

identification to address evolving requirements

Privacy regulators

engage on tools &

services

Growing ecosystem

of privacy experts

Need for upskilling of proficiency in data

privacy needs and specialisms

Compliance vs utility

of data for research

Ethical

considerations

Ongoing regulatory complexities and changes in

data protection laws

Short term

Improve data privacy &

compliance requirements

Medium term

Develop initial data de-

identification automation

Long term

Collaborate on advanced

data privacy techniques

Unintended data de-

identification techniques

causing privacy breaches

Evolving data privacy

regulations affecting de-

identification processes

Scepticism of strength of

de-identification

practices from Customers

Data de-identification is a pivotal technology in fintech, ensuring the responsible use of sensitive financial data for AI research and analysis. By protecting customer identities while facilitating

data-driven insights, fintech companies can innovate and improve financial services while adhering to strict data privacy regulations. Data de-identification brings opportunities for competitive

advantages, cost savings, and enhanced customer trust. While the potential benefits are significant, challenges related to technical feasibility, regulatory compliance, and market acceptance

must be addressed. The future outlook for data de-identification in fintech is promising, with a gradual timeline for adoption and a need for careful risk management and responsible

implementation. By capitalizing on the benefits and addressing frictions, data de-identification will continue to play a crucial role in shaping the future of privacy-aware fintech services.

Data Quality

& Observability

This practice ensures that

data used in AI models is

accurate, complete, and

reliable.

In fintech, where data drives

critical decision-making,

these technologies play a

fundamental role in

maintaining trustworthy AI-

driven predictions and

outcomes.

By enhancing data

observability, fintech

companies gain valuable

insights into data pipelines,

ensuring transparency, and

reducing risks associated

with inaccurate or biased

data.

Increasing reliance

on AI-driven

decision making

Advances in data

pipeline

transparency

Higher demand for

data-driven

insights

Demand for accurate

and reliable data across

the industry

Customer trust

driven by accurate

predictions

Regulatory

mandates on

transparency

Simultaneous adoption of

data transparency processes

Established industry best

practices & data standards

Obsolescence penalised in

the marketplace

Legacy methods challenged

by evolving standards

More data literate & aware

talent in the workforce

Quantifiable trust & accuracy in predictions

Increased revenue through data-driven insights

Reduced potential for legal & reputational

damage caused by poor data quality

Reduced impact of data inaccuracies

Increased revenue from higher model accuracy

Attraction of investors impressed by

commitment to data quality

Integration or defined practices to core

platforms & business areas

Observability helping with overall strategic

planning of business

Interoperability strengthening data quality

standards through increased transparency

Demonstrations of

improvements to

accuracy

Already scalable to

large data volumes

Backdating of work to historic data sources

required for proper meshing

Commonplace tools

available for

Business Intelligence

Data Quality &

Observability Talent

is commonplace

Data Science requirements for strategic

planning & legacy solutions harder to source

Compliance priority

over quality-of-life

hygiene

Data quality

impacting decision

making to users

Observability may cause painful indicators to be

published until data quality is solved

Short term

Identify data repositories

and quality issues

Medium term

Publish & implement data

quality standards

Long term

Automate observation

and handle legacy data

Overly complex data

quality processes creating

inefficiencies

Stringent data quality

standards affecting

operational practices

Prioritisation slowing

down legacy data clean-

Data quality and observability are fundamental practices in fintech, ensuring the accuracy, completeness, and reliability of data used in AI models. By enhancing data observability, fintech

companies gain valuable insights into data pipelines, ensuring transparency, and reducing risks associated with inaccurate or biased data. Data quality and observability offer opportunities for

competitive advantages, cost savings, and enhanced customer trust. While the potential benefits are significant, challenges related to technical feasibility, data privacy, and regulatory

compliance must be addressed. The future outlook for data quality and observability in fintech is promising, with a gradual timeline for adoption and a need for careful risk management and

responsible implementation. By capitalizing on the benefits and addressing frictions, data quality and observability will continue to play a crucial role in shaping the future of trustworthy and

data-driven fintech services.