Consideraciones éticas

sobre inteligencia artificial

MD006 – Ética y legalidad de los datos
26 de octubre de 2023

Máster Universitario en Ciencia de los Datos (MUDS)

Dr. Álvaro García Piquer - [email protected]
Sergi Bernet Andrés – [email protected]
Laura Serra Roig – [email protected]
2 A Lot of Words

Data Science

Artificial Intelligence

Aspectos éticos y legales de los datos

3 A Lot of Words

Data Science

Artificial Intelligence

Machine Learning

Aspectos éticos y legales de los datos

4 A Lot of Words

Data Science

Artificial Intelligence

Machine Learning

Deep Learning

Aspectos éticos y legales de los datos

5 A Lot of Words

Data Science

Artificial Intelligence
Computer Vision
Machine Learning

Deep Learning

Aspectos éticos y legales de los datos

6 A Lot of Words

Data Science

Artificial Intelligence
Computer Vision
Machine Learning
Natural Language Processing
Deep Learning

Aspectos éticos y legales de los datos

7 A Lot of Words

Data Science
Robotics
Artificial Intelligence
Computer Vision
Machine Learning
Natural Language Processing
Deep Learning

Aspectos éticos y legales de los datos

8 Contextualizing

Data Science

Artificial
Intelligence

Machine
Learning
Deep Learning
Deep learning is a subset of
machine learning that uses artificial
neural networks to model and solve
complex tasks

Aspectos éticos y legales de los datos

9 Contextualizing

Data Science

Artificial
Intelligence
Computer Vision
Natural Language Processing
Robotics
Knowledge Representation
Planning & Optimization

Machine
Learning
Deep Learning
Deep learning is a subset of
machine learning that uses artificial
neural networks to model and solve
complex tasks

Aspectos éticos y legales de los datos

10 Data Science Project

Business Data Data Modeling Evaluation Deployment

understanding understanding preparation and decision

Aspectos éticos y legales de los datos

11 Data Science Project
MD002 - Infraestructuras de computación
MD003 - Estructuras de datos y su almacenaje
MD004 - Herramientas de análisis y visualización de datos
MD005 - Sistemas basados en el conocimiento
MD006 - Pensamiento y Creatividad. Aspectos éticos y legales de los datos
MD007 - Datos y negocio
MD008 - Inteligencia artificial para la ciencia de los datos
MD009 - Herramientas avanzadas de análisis de datos

Business Data Data Modeling Evaluation Deployment

understanding understanding preparation and decision

MD007 MD004 MD002 MD005 MD004 MD002

MD009 MD003 MD008 MD009 MD003
MD009

Aspectos éticos y legales de los datos

12 Data Science Project
MD002 - Infraestructuras de computación
MD003 - Estructuras de datos y su almacenaje
MD004 - Herramientas de análisis y visualización de datos
MD005 - Sistemas basados en el conocimiento
MD006 - Pensamiento y Creatividad. Aspectos éticos y legales de los datos
MD007 - Datos y negocio
MD008 - Inteligencia artificial para la ciencia de los datos
MD009 - Herramientas avanzadas de análisis de datos

Business Data Data Modeling Evaluation Deployment

understanding understanding preparation and decision

MD007 MD004 MD002 MD005 MD004 MD002

MD006 MD009 MD003 MD008 MD009 MD003
MD006 MD006 MD009 MD006 MD006
MD006

Aspectos éticos y legales de los datos

13 Types of Advanced Analytics

• There are three types of analytics that companies use to drive their
decision making.

Prescriptive
Analytics

Predictive
Analytics What should be done?

Mathematical optimization,
simulation...
value

Descriptive What will happen?

Analytics
Data mining, forecasting,
machine learning, deep
learning…
What happened?
What is happening?

Descriptive statistics,
visualizations, clustering...

complexity

Aspectos éticos y legales de los datos

Real-Life Applications
of Artificial Intelligence
 Real-Life Examples – Descriptive Analytics
15

Netflix
• It is an excellent example of descriptive analytics, as it collects large amounts of data from users’ in-
platform to recognize trending shows and movies. This allows Netflix to make appropriate
recommendations to users and drives the internal decision-making mechanism for future content
production.

Google
• The results of a web server company using Google Analytics tools are the best example of descriptive
analysis. The results help determine what exactly happened in the past and determine whether a
promotional effort was effective based on simple metrics such as page views.
Amazon (on the border with artificial intelligence)

• It looks at which items you have recently purchased, which items are in your online shopping cart or
wishlist, which items you have checked and rated, and which items you are most searching for. This
knowledge is used to suggest additional products bought by other consumers as they order those
same things.
Aspectos éticos y legales de los datos
 Real-Life Examples – Predictive Analytics
16

Demand forecasting
• Retailers use predictive analytics algorithms to analyze different factors, such as region, season, and
buying habits, to forecast product demand. This approach helps them determine the ideal inventory
level to satisfy demand, preventing overstocking or stockouts.

Predictive maintenance
• IoT sensors installed on machines continuously collect data on their performance, which is analyzed by
predictive models to identify abnormalities and suggest maintenance of specific parts. By using
predictive analytics in this way, plants and factories can prevent equipment breakdowns and avoid
downtime.

Aspectos éticos y legales de los datos

 Real-Life Examples – Predictive Analytics
17

Finance analytics
• Financial teams are already leveraging predictive analytics to predict risks and revenues, allocate
resources effectively, and adjust operations to prevent extra expenses. It has a wide range of
applications, including fraud detection, credit scoring, customer segmentation, and portfolio optimization.

Insurance analytics

• Predictive analytics helps insurance companies correctly estimate future risks, as well as determine
fraudulent claims in time and reject them, avoiding unreasonable expenses.
Customer analytics

• Predictive analytics models help prevent customer churn by analyzing the dissatisfaction among current
customers and identifying customer segments at most risk for leaving

Aspectos éticos y legales de los datos

 Real-Life Examples – Predictive Analytics
18

Healthcare analytics
• Many medical institutions worldwide are implementing software systems that collect a large amount of
patient data, providing opportunities for predictive analytics. By analyzing historical and current data,
predictive algorithms can identify disease prerequisites and suggest preventive treatments, predict
the results of different treatments to choose the best option for each patient individually. Additionally,
predictive analytics can help predict disease outbreaks and epidemics.
Energy management
• Predictive analytics can help energy utilities create short- and long-term forecasts on energy
demand (and generation), considering weather conditions, seasonality, new emerging consumers and
other factors.
Sports analytics
• Analytics specialists in sports use on-field and off-field data to predict a player's value and
progression. On-field metrics include physical performance such as speed, time, scoring, tactics, and
health conditions. Off-field metrics refer to the business side of sports and provide insights on how
much profit a player can bring to the team/club, considering fan engagement, ticket sales, or
merchandise sales.

Aspectos éticos y legales de los datos

 Real-Life Examples – Prescriptive Analytics
19

Inventory optimization
• It determines the optimal inventory levels that will minimize costs while meeting customer
demand. It involves to formulate a mathematical model that represents the inventory system and the
associated costs, such as ordering costs, holding costs, and stockout costs. The model considers various
factors, such as demand, lead time, and safety stock, and aims to identify the optimal inventory levels that
will minimize the total cost of the system.

Efficient logistics
• Predictive analytics algorithms can help logistic companies in multiple ways. By analyzing factors such as
roads, traffic, and weather conditions, these algorithms can determine the fastest and cost-efficient
routes for logistics. Additionally, the algorithms can monitor fuel consumption and driving behavior,
helping to reduce transport costs.

Aspectos éticos y legales de los datos

 Real-Life Examples – Prescriptive Analytics
20

Cashflow optimization
• By optimizing cashflow businesses can identify opportunities to increase cash inflows and reduce
cash outflows, such as delaying payments, optimizing cash collections, or reducing inventory levels. The
optimization model can be adjusted over time to reflect changes in market conditions or changes in the
business environment, ensuring that the business always has the optimal cash management strategy.

Price optimization
• It helps businesses determine the optimal prices for their products or services. The model considers
factors such as market conditions, competition, and customer preferences. The objective is to identify the
optimal price points that will maximize profit while meeting customer demand.

Aspectos éticos y legales de los datos

 Real-Life Examples – Prescriptive Analytics
21

Scheduling
• Its goal is to make schedules that improve efficiency, reduce costs, and increase productivity.
The model represents the relationships between several aspects such as production processes, employee
schedules, resource allocation, and workloads. The objective is to identify the optimal schedule that will
maximize productivity while minimizing costs.

Treatment planning
• Healthcare providers develop treatment plans that minimize the risk of complications and
maximize patient outcomes. This can involve optimizing the dosage and timing of medications,
radiation therapy, or other treatments.

Aspectos éticos y legales de los datos

 What Impact Does AI Have?
22

Positive impact Negative impact

Economy Economy

Society Society

Aspectos éticos y legales de los datos

 What Impact Does AI Have?
23

Positive impact Negative impact

Economy Economy
• Increased productivity and efficiency in several • Displacement of some jobs
industries
• Rising income inequality as high-skill jobs requiring AI
• Creation of new job opportunities expertise become more valuable

Society Society
• Ethical concerns related to data privacy, bias, and
• Improvement in decision-making in various fields.
potential misuse.
• Greater convenience thanks to the development of
smart devices and virtual assistants. • Dependence on technology and the loss of certain
human skills.
• Enhancement of healthcare through personalized
treatment plans and early disease detection. • Potential exacerbation of existing social and
economic inequalities.

Aspectos éticos y legales de los datos

 What Impact Does AI Have?
24

Positive impact Negative impact

Economy Economy
• Increased productivity and efficiency in several • Displacement of some jobs
industries
It is important for society to continue
• Risingexamining
income inequality as high-skill jobs requiring AI
• Creation of new job opportunities
and critically regulating the development
expertiseand use of
become more valuable
AI to ensure that its benefits are distributed widely
and its negative consequences are minimized.

Society Society
• Ethical concerns related to data privacy, bias, and
• Improvement in decision-making in various fields.
potential misuse.
• Greater convenience thanks to the development of
smart devices and virtual assistants. • Dependence on technology and the loss of certain
human skills.
• Enhancement of healthcare through personalized
treatment plans and early disease detection. • Potential exacerbation of existing social and
economic inequalities.

Aspectos éticos y legales de los datos

MUDS - Inteligencia artificial para la ciencia de los datos
Ethics and regulations
of Artificial Intelligence
 Bias
27

Coded Bias (documentary)

• https://www.netflix.com/title/81328723

• Trailer
▪ https://www.youtube.com/watch?v=jZl55PsfZJQ

Joy Buolamwini talk

(computer scientist and digital activist based at the MIT Media Lab)
• https://www.youtube.com/watch?v=UG_X_7g63rY

Aspectos éticos y legales de los datos

 Types of Bias
28

Bias in data science refers to systematic errors or biases in algorithms and systems that lead to
unfair or inaccurate results.
• Selection bias: Data science algorithms can be biased towards certain groups or individuals if
they are underrepresented in the data.
• Sampling bias: Data science algorithms extract insights from the data they are trained on.
Even if the collected data was not originally biased, if the data used to train does not have the
same variety than the original data, the insights will reflect these biases.
• Observer bias: The design and development of data science algorithms can also introduce
biases. For example, an algorithm developed by a team with a specific demographic bias may
produce biased results.
• Confirmation bias: Algorithms can be programmed to only look for information that confirms
existing beliefs or biases.

Aspectos éticos y legales de los datos

 Examples of Bias
29

COMPAS

• US judges use it to predict which criminals are more likely to reoffend in the future
• Black criminals were considered more likely to reoffend in the future than they were. On the other hand, white criminals
were judged to be less risky than they were.

PredPol

• It is used by US police departments in California, Florida, Maryland, etc. to predict crimes.

• Police officers were repeatedly dispatched to neighborhoods containing many racial minorities. In the data used,
regardless of the amount of crime that occurred in the area, more police complaints were reported in these regions. This
was because the police concentration was higher in these regions due to the existing human bias.

Amazon’s Recruiting Engine

• The system was developed to analyze resumes and job applications and identify the most qualified candidates.
• It was found to discriminate against women, dismissing resumes containing words like "woman" or "motherhood" and
preferring resumes that included experience in male-dominated organizations.

Aspectos éticos y legales de los datos

 Types of Bias
30

Aspectos éticos y legales de los datos

 Types of Bias
31

Survivorship bias: It is a special

case of selection bias, concentrating
on entities that passed a selection
process while overlooking those that
did not.

Aspectos éticos y legales de los datos

32

Aspectos éticos y legales de los datos

 AI (and Data Science) Risks
33

The Social Dilemma (documentary)

• https://www.netflix.com/title/81254224
• Trailer
▪ https://www.youtube.com/watch?v=uaaC57tcci0

Interview with Frances Haugen

(advocate for accountability & transparency in social media)
• https://www.youtube.com/watch?v=9Q6tgdTlQxM

Interview with Tristan Harris

(Co-Founder & Executive Director of the Center for Humane Technology, a
nonprofit organization whose mission is to align technology with humanity’s
best interests.)
• https://www.youtube.com/watch?v=YruYAnzirxw

Aspectos éticos y legales de los datos

MUDS - Inteligencia artificial para la ciencia de los datos
 Examples of Ethical Considerations
35

Advantage Risk for Data

for society society Protection

Aspectos éticos y legales de los datos

38 Why AI (and Data Science)
Should Be Regulated?

Risk Solution

Artificial Intelligence (AI) Creating regulations for AI

is a technology that research and development
benefits humanity, but it can ensure that safe AI
has the potential to cause becomes an integral part
massive and irreparable of the AI creation process.
harm to civilization.

Aspectos éticos y legales de los datos

MUDS - Inteligencia artificial para la ciencia de los datos
 Data Privacy and Data Governance
40

Data governance is everything that is done to

ensure that data is secure, private, accurate,
available and usable.
In 2016, the EU approved the General Data
Protection Regulation (GDPR) to guarantee the
protection and privacy of user data.

Aspectos éticos y legales de los datos

 Ethical Guidelines for Reliable Artificial Intelligence
41

European Commission, Directorate-General for

Communications Networks, Content and Technology, Ethics
guidelines for trustworthy AI, Publications Office, 2019,
https://data.europa.eu/doi/10.2759/346720
• The guidelines present a set of 7 key requirements
that AI systems must meet in order to be considered
trustworthy.

Aspectos éticos y legales de los datos

 Recommendation on the Ethics of Artificial Intelligence
42

UNESCO, Recommendation on the Ethics of Artificial Intelligence, 2022, United Nations Educational, Scientific
and Cultural Organization, 2022, https://unesdoc.unesco.org/ark:/48223/pf0000381137
• The Recommendation aims to realize the advantages that AI brings to society and reduce the risks it entails.
It ensures that digital transformations promote human rights and contribute to the achievement of the Sustainable
Development Goals, addressing issues of transparency, accountability and privacy, with action-oriented policy chapters
on data governance, education, culture, work, health care and economy.

Decisions impacting millions of people should The world needs rules for artificial intelligence
be fair, transparent and contestable. These new to benefit humanity. The Recommendation on
technologies must help us address the major the ethics of AI is a major answer. It sets the
challenges in our world today, such as first global normative framework while giving
increased inequalities and the environmental States the responsibility to apply it at their
crisis, and not deepening them. level. UNESCO will support its 193 Member
States in its implementation and ask them to
Gabriela Ramos, UNESCO’s Assistant Director report regularly on their progress and practices.
General for Social and Human Sciences
Audrey Azoulay, UNESCO Director-General

Aspectos éticos y legales de los datos

 Artificial Intelligence Law
43

The "EU AI Act" is a regulation proposed on April 21, 2021 by the European Commission with the aim of
introducing a common regulatory and legal framework for artificial intelligence.
• It classifies artificial intelligence applications based on risk and regulates them accordingly.

https://www.adalovelaceinstitute.org/resource/eu-ai-act-explainer/
Aspectos éticos y legales de los datos
 What Do You Think about this Text?
45

“Este trabajo sirve como un recordatorio crucial de que la atención meticulosa a la calidad
y preparación de los datos es tan importante como la elección del modelo adecuado.
Hasta que estos desafíos fundamentales relacionados con los datos sean abordados,
incluso el modelo más sofisticado no obtendrá los mejores resultados. Por lo tanto, el
enfoque futuro debe ser tanto en la mejora de los modelos como en la garantía de la
calidad de los datos para lograr mejorar la utilidad y eficacia de los modelos en cuestión.”

Aspectos éticos y legales de los datos

 What Do You Think about this Text?
46

“Este trabajo sirve como un recordatorio crucial de que la atención meticulosa a la calidad
y preparación de los datos es tan importante como la elección del modelo adecuado.
Hasta que estos desafíos fundamentales relacionados con los datos sean abordados,
incluso el modelo más sofisticado no obtendrá los mejores resultados. Por lo tanto, el
enfoque futuro debe ser tanto en la mejora de los modelos como en la garantía de la
calidad de los datos para lograr mejorar la utilidad y eficacia de los modelos en cuestión.”

“La calidad de los datos es fundamental, tan importante como la elección de un

modelo de aprendizaje automático. Sin resolver los problemas de datos, incluso
los modelos avanzados no funcionarán adecuadamente, por lo que es esencial
mejorar ambos para asegurar su efectividad en el futuro.”

Aspectos éticos y legales de los datos

 Individual Social Responsibility and Artificial Intelligence
47

Being ethical and socially responsible encourages individuals to make sound decisions by relying on their
moral beliefs and upholding their civic duty.
Each individual has a responsibility, understanding that the actions they commit should ultimately
benefit societal welfare.

Example
▪ Use of deepfakes: image, video, and audio manipulated with artificial intelligence

Aspectos éticos y legales de los datos

48

Aspectos éticos y legales de los datos

 Intellectual Property and Copyright
49

Generative AI harnesses authors' creative outputs in both text and image generation. It
analyzes their writing styles and visual themes to generate text that mimics
their literary voices and produce artwork inspired by their creations.

Aspectos éticos y legales de los datos

51

Aspectos éticos y legales de los datos

 Sustainable Development Goals
52

The Sustainable Development Goals (SDGs) are a universal call to action to end poverty, protect the
planet and ensure peace and prosperity for all by 2030.

Aspectos éticos y legales de los datos

 Impact on Sustainable Development Goals
54

Sætra, Henrik. (2021). AI in Context and the Sustainable Development Goals: Factoring in the
Unsustainability of the Sociotechnical System. Sustainability. 13. 10.3390/su13041738.
Aspectos éticos y legales de los datos
 Plans estratègics
55

• (Comisió Europea, Desembre 2018) “Plan coordinado sobre la inteligencia artificial”.

https://www.eciia.eu/2021/05/european-commission-coordiated-plan-on-artificial-intelligence-
review/
• (Generalitat de Catalunya, Febrer 2020) “L’Estratègia d’Intel·ligència Artificial de Catalunya”
https://smartcatalonia.gencat.cat/ca/projectes/tecnologies/detalls/article/Catalonia-AI
• (Gobierno de España, Desembre 2022) “Estrategia Nacional de Inteligencia Artificial ”.
https://portal.mineco.gob.es/es-es/digitalizacionIA/Paginas/ENIA.aspx

Anàlisis de diferents països

• (Fatima et al., September 2020) “National strategic artificial intelligence plans: A multi-
dimensional analysis”.
https://www.sciencedirect.com/science/article/abs/pii/S0313592620304021

Aspectos éticos y legales de los datos

 Additional References
56

How Facebook shapes your feed

• https://www.washingtonpost.com/technology/interactive/2021/how-facebook-algorithm-works/

5 Algorithms that Demonstrate Artificial Intelligence Bias

• https://www.geeksforgeeks.org/5-algorithms-that-demonstrate-artificial-intelligence-bias/

Coded Bias (documentary)

• https://www.netflix.com/title/81328723

Who Is Making Sure the A.I. Machines Aren’t Racist?

• https://www.nytimes.com/2021/03/15/technology/artificial-intelligence-google-bias.html

The Social Dilemma (documentary)

• https://www.netflix.com/title/81254224

Aspectos éticos y legales de los datos