CORE139:
LARGE LANGUAGE
MODELS
ETHICS
AND
LLMs
NEWS
.
• https://towardsdatascience.com/the-name-that-broke-chatgpt-who-i
s-david-mayer-f03f0dc74877
• https://sora.com/
• https://openai.com/index/video-generation-models-as-world-simulators/
4
ETHIC AND AI
REFLECTION AND VISION OF THE
FUTURE IN LITERATURE
● 1818 - Frankenstein - Mary Shelley (explores the ethics of
creating intelligent beings and the responsibility of creators)
● 1872 - Erewhon - Samuel Butler (speculates about the
development of machines with consciousness and their possible
domination)
● 1920 - R.U.R. - Karel Čapek (introduces the word "robot" and
describes the rebellion of artificial beings against humanity)
● 1950 - I, robot - Isaac Asimov (formulates the three laws of
robotics and explores the relationship between humans and
robots)
(by Vít Šebestík)
6
ETHIC CHALLENGES AND LLMs
models have become an integral part of applications ranging from
customer service automation to content generation. Here are the key
ethical considerations for their use:
● Accountability and transparency
● Bias and discrimination
● Privacy and security
● Misinformation and manipulation
● Impact on employment
7
ACCOUNTABILITY AND TRANSPARENCY
Transparency in how LLMs operate is crucial for building trust among
users. Stakeholders must understand how these models make
decisions and generate outputs. This includes providing clear
explanations for model behavior and ensuring accountability for the
consequences of their use. Developing ethical frameworks that outline
the responsibilities of AI developers can help foster accountability
within the industry.
8
BIAS AND DISCRIMINATION
Large language models can inadvertently reproduce or amplify
existing biases in society:
Identifying Bias: It is essential to analyze training data and model
outputs for the presence of stereotypes and discriminatory views.
Impact mitigation: organisations should implement strategies to
minimise bias and ensure fair treatment of all users.
9
PRIVACY AND SECURITY
The use of LLM may have privacy implications:
Data protection: it is necessary to ensure that models are not trained
on sensitive personal data without users' consent.
Information security: It is important to protect data from misuse or
unauthorized access.
10
MISINFORMATION AND MANIPULATION
LLM can be misused to create misinformation:
Preventing misinformation: organizations should have clear policies
against using AI to spread false information.
User education: it is important to educate the public on how to
recognize AI-generated misinformation.
11
IMPACT ON EMPLOYMENT
Automation through LLM can impact the labour market:
Transition to new roles: It is essential to support workers in
transitioning to new roles created by technological advances.
Training and retraining: Organisations should invest in training and
retraining programmes for employees.
12
PAIR TASK
ACTIVITY
● 1: Content Generation: A language model generates fake news.
What are the risks, and who should bear responsibility?
● 2: Bias in Models: The model favors a specific culture or
religion. How would you address this issue?
● 3: Use in Education/Research: Students/Researches use the
model to write essays/papers. Is this ethical?
● 4: Data Privacy: The model is trained on users' sensitive data.
How can privacy be ensured?
14
DATA
WHAT TO KNOW ABOUT DATA?
● Data provenance: refers to the documentation of the origin of
data, including its source, context, and how it has been
transformed over time. Understanding data provenance is vital for
ensuring ethical data use, as it provides insights into the quality
and reliability of the data used to train LLMs.
● Data lineage: refers to the documentation of the origin of data,
including its source, context, and how it has been transformed
over time. Understanding data provenance is vital for ensuring
ethical data use, as it provides insights into the quality and
reliability of the data used to train LLMs.
16
DISCUSSION
WHAT ARE YOU AFRAID ABOUT?
18
AI REGULATIONS
WORLDWIDE
The global approach to AI regulation reflects a diverse range of
strategies aimed at balancing innovation with ethical considerations
and public safety. As countries continue to develop their regulatory
frameworks, ongoing collaboration and dialogue will be essential in
establishing effective governance that can adapt to the rapidly
changing landscape of AI technology.
20
WORLDWIDE
The global approach to AI regulation reflects a diverse range of
strategies aimed at balancing innovation with ethical considerations
and public safety. As countries continue to develop their regulatory
frameworks, ongoing collaboration and dialogue will be essential in
establishing effective governance that can adapt to the rapidly
changing landscape of AI technology.
21
EUROPEAN UNION
● AI Act: The EU is pioneering comprehensive AI regulation with
the AI Act, which establishes a legal framework addressing the
risks associated with AI. It categorizes AI applications into four
risk levels, sets requirements for high-risk applications, and
mandates conformity assessments before deployment. The Act
aims to ensure safety and protect fundamental rights while
promoting innovation within the EU. It is expected to be fully
applicable by June 2026 after a two-year grace period.
● AI Liability Directive: This proposed directive aims to clarify civil
liability related to damages caused by AI systems, addressing
gaps in existing product liability laws.
22
UNITED STATES OF AMERICA
● Artificial Intelligence Environmental Impacts Act of 2024:
Introduced in February 2024, this legislation directs the National
Institute of Standards and Technology (NIST) to develop
standards for measuring AI's environmental impacts, promoting
transparency in reporting these effects.
● State-Level Initiatives: Various states are exploring their own
regulations, leading to a patchwork of laws across the country.
This includes guidelines on data privacy and ethical AI use in
sectors such as healthcare and finance.
23
UNITED KINGDOM
● AI Security Guidelines: Developed by the UK National Cyber
Security Centre, these guidelines focus on enhancing
cybersecurity measures for AI systems. They cover secure design
practices and have been endorsed by multiple countries..
● ICO Consultation: The Information Commissioner’s Office (ICO)
is consulting on how existing data protection laws apply to
generative AI models, emphasizing the need for transparency in
data usage.
24
ASIA
● China: In March 2024, China published a draft Artificial
Intelligence Law, which aims to establish strict governance over
AI technologies. This law emphasizes safety and ethical
considerations in AI development.
● Singapore: The Cyber Security Agency launched guidelines to
secure AI systems across their lifecycle, focusing on best
practices for mitigating cybersecurity risks..
● Hong Kong: The Hong Kong Monetary Authority issued guiding
principles for using generative AI in customer-facing applications,
emphasizing governance, fairness, and transparency.
25
MIDDLE EAST
● Qatar Central Bank Guidelines: These guidelines mandate
licensed entities to develop an AI strategy that includes
accountability measures and risk management protocol.
26
UNESCO IDEA
UNITED NATIONS, UNITED IDEAS
AI regulation should not be viewed from the perspective of national
interests, but from the perspective of humanity as a whole.
28
29
QUESTIONS
30