Description
New Content for Week 5:
5. Conclusion and Future Works
5.1 Conclusion
Describe what you have concluded in the research project including how the research question has been answered and if the hypothesis has been approved or disapproved, as well as the main contribution or takeaway of this research project.
5.2 Limitations
Discuss any limitations of your solution, and the reason of the limitations, as well as possible solution to overcome the limitation.
5.3 Future Works
Provide a list of at least 2 current or future technology trends that might affect your project or research results.
Discuss how each trend would influence your project or research.
Your discussion should provide details about the impact that the technology trend would have on your project or research.
Justify your analysis with appropriate research.
Produce a financial impact analysis scenario for a hypothetical company based on the potential effect of trends identified in the previous steps.
Abstract and Keywords
Provide an Abstract of your research project in 200 – 250 words which should include the motivation and background of the project, the problem to be solved, the solution to address the problem, the main technical feature of the solution, and the result of the research project.
Provide at least three keywords
Put both the Abstract and keywords on the title page of the Computer Science Problem-Solving Research Project Report document.
Key Assignment final draft: Computer Science Problem Analysis document
Review the entire document for any changes and improvements that you would like to make.
Ensure that this final version of the document is sufficiently detailed to fully meet the assignment requirements for each part of the course.
Any previous instructor feedback should be addressed with appropriate changes.
Include a summary of lessons learned and next steps for the project.
Running Head: ARTIFICIAL INTELLIGENCE
Facebook – Artificial Intelligence
Colorado Technical University
Shirish Bhatnagar
February 3rd, 2021
1
2
ARTIFICIAL INTELLIGENCE
Abstract
There is a lack of diversity in the Artificial Intelligence and machine learning fields.
Most Artificial Intelligence experts are white and male, and the products created directly
symbolize these demographics. The majority of these Artificial Intelligence experts are unaware
of the problems and issues their work poses for other groups and how this could tamper with the
machines’ integration and implementation.
For instance, Obamehinti’s tale of algorithmic discrimination shows how Facebook has
had to invent new tools and processes to cater to the diversity these Artificial Intelligence
machines pose (Tom, 2019). Upon further investigation, it was discovered that women and
people of darker skin were under presented in the Artificial Intelligence training data. This
directly influenced the product as it became less accurate and ineffective in catering to the entire
population niche’s needs. Other organizations have gone ahead to shed light on the risk of biased
Artificial Intelligence systems and how assigning critical and personal roles could influence high
levels of customer dissatisfaction. It is essential to understand that
Facebook continues to realize profits by increasingly being careful of the type of products
and services they introduce to the market. Additionally, the world is increasingly becoming
aware of the importance of diversity. As a result, Facebook needs to be careful and aware of the
products and services they offer the public; this includes the integration of Artificial Intelligence
and machine learning, which poses significant threat levels to the organization. The research will
shed light on this issue by providing a comprehensive assessment and analysis of the situation. It
will also make it easy to come up with possibilities and solutions to make the Artificial
Intelligence better and ease its integration
ARTIFICIAL INTELLIGENCE
3
Contents
Contents …………………………………………………………………………………………………………………………………. 3
Artificial Intelligence …………………………………………………………………………………………………………………. 5
Company Overview ……………………………………………………………………………………………………………….. 5
Problem Identified …………………………………………………………………………………………………………………. 7
Problem Context and Background ………………………………………………………………………………………….. 8
Problem statement ……………………………………………………………………………………………………………….. 10
Data collection methods Identified. ……………………………………………………………………………………… 10
Solutions Identified ……………………………………………………………………………………………………………. 11
Approach ………………………………………………………………………………………………………………………….. 14
Solution Identification ……………………………………………………………………………………………………………… 16
Hypothesis Statement …………………………………………………………………………………………………………… 22
Changing the algorithms and datasets used could help identify a solution for the problem of lack of
diversity in AI and machine learning ………………………………………………………………………………………… 22
Research Question ……………………………………………………………………………………………………………….. 22
Related Works ……………………………………………………………………………………………………………………… 22
Stainer James (2019) we must fix diversity problems……………………………………………………………… 23
Research Methodology …………………………………………………………………………………………………………….. 25
Research matters………………………………………………………………………………………………………………….. 25
Research Methodology …………………………………………………………………………………………………………. 25
Experiment design ……………………………………………………………………………………………………………….. 27
Requirements and Constraints of the experiment ………………………………………………………………….. 28
ARTIFICIAL INTELLIGENCE
4
Experiment Procedures ……………………………………………………………………………………………………………. 29
Project Plan …………………………………………………………………………………………………………………………….. 31
Project Plan for Data Collection & Analysis ………………………………………………………………………….. 32
Tasks ………………………………………………………………………………………………………………………………… 32
Timeline and Human Resource …………………………………………………………………………………………….. 33
Estimated Costs ……………………………………………………………………………………………………………………. 35
Risk and Cost-Benefit Analysis…………………………………………………………………………………………………. 36
Conclusion and Future Works (Week 5) …………………………………………………………………………………… 38
Conclusion (Week 5) …………………………………………………………………………………………………………….. 38
Limitations (Week 5) ……………………………………………………………………………………………………………. 38
Future Works (Week 5) ………………………………………………………………………………………………………… 38
References ……………………………………………………………………………………………………………………………….. 39
5
ARTIFICIAL INTELLIGENCE
Artificial Intelligence
Company Overview
Facebook is an American Company that offers online social networking services. It was
founded by Mark Zuckerberg, Dustin Moskovitz, Chris Hughes, and Eduardo Saverin. The
platform has grown to become the largest social network platform globally, with more than one
billion users as of 2012. The Company has its headquarters in Menlo Park, California. Access to
the platform is provided to the world free of charge, with the requirement being the owner of an
account (Caers, 2013). The organization generates most of its revenue from advertisements on
the website. The platform provides users with the ability to join groups, network, upload photos,
and create new profiles. Users can chat with each other and send messages; the platform also has
the choice of a phone or video call.
The attractiveness of Facebook comes from Mark Zuckerberg’s insistent nature of users
being honest and transparent about who they are (Hall, 2017). Users are forbidden from using
false identities; the organization blocks any suspected or reported user of using a false identity or
duping people (Kirkpatrick, 2011). The main argument behind this is that transparency is
necessary for forming personal relationships and attachments and contributes to sharing
information and ideas, and helps build society as a whole. The platform has also made it easier
for businesses to connect or inform the consumers of the products offered easily. Facebook aims
at satisfying the needs of its users by providing them with the best quality of services. However,
this has not always been as easy as the advancements in technology have also introduced various
disadvantages that need to be addressed.
ARTIFICIAL INTELLIGENCE
6
Project Introduction
The project will cover the integration of Artificial Intelligence, also known as AI, on
Facebook and the possible solutions for the challenges faced. However, it is essential first to
identify the products offered by the organization. The Products offered include; Messenger,
Instagram (has applications such as Boomerang), Facebook Shops, Spark, AR Studio, Audience
Network, Facebook Shops, and any other software or applications offered by Facebook Inc.
Facebook has recently diversified and began purchasing other applications such as Whatsapp and
Instagram. This is a move to influence efficiency and continue the market domination observed
over time. It is essential to note that Facebook products do not include Facebook-offered
products or services that contain their privacy policies and terms of services. These exceptions
include Free Basics, Messenger Kids, and Oculus Products.
It is essential to note that almost everyone on the planet today has at least one social
media account, as these platforms help people stay connected with their families and friends. The
technology used allows link people, ideas, and thoughts over vast geographical areas. According
to Statista 2020, Facebook is considered the largest and most popular social networking site, with
over one million active users (Clement, 2020). Facebook faces stiff competition from sites such
as Snapchat, LinkedIn, and Twitter. Snapchat is a photo and video sharing application developed
by Evan Spiegel and Bobby Murphy in 2011. The organization boasted 218 million daily users in
the year 2019, showing its reach and competitiveness. Snapchat is very popular with teenagers as
it offers some of the most popular features and filters. Twitter and LinkedIn also pose
considerable threats to Facebook as they have defined their products and provided the consumers
with what Facebook missed. For instance, LinkedIn is a site that allows prospective employers
ARTIFICIAL INTELLIGENCE
7
and employees to communicate. It is a job-search application and is known by the specificity of
the features offered. It is taken more seriously when it comes to job searching than Facebook.
Facebook has grown significantly since its introduction to the market. The platform is
more robust and continues to provide the users with new features that enhance the networking
experience. Facebook has recently delved into the Artificial Intelligence market and has
identified a gap that it intends to fill (Croft, 2007). The Company functions with the purpose of
connecting people around the planet through the products and services offered. It has identified
Artificial Intelligence as one of the ways to improve and enhance this connectivity. Artificial
intelligence enables machines to learn and clarify data without supervisory help. The
development of Artificial Intelligence and machine learning has enabled growth in the
organization in various ways. For instance, Facebook analyzes a large amount of data shared.
The Artificial Intelligence employed helps ascertain engagement levels, analyzes large data
volumes, and allows Facebook to match the appropriate content.
Problem Identified
The problem being investigated pertains to the challenges Facebook continues to face in
integrating and implementing Artificial Intelligence into their systems. Artificial intelligence and
machine learning are a big deal and continue to contribute significantly to technological
advancements in the world today. According to technical estimates, Artificial Intelligence
algorithms and robots could take over 40% of the global jobs available to human beings today
(Tom, 2019). However, Artificial Intelligence also poses some significant challenges to
organizations that need to be scaled for effective and successful implementation. One such
problem of Artificial Intelligence in relation to Computer Science is the problem of diversity.
ARTIFICIAL INTELLIGENCE
8
Problem Context and Background
According to the New York University Research center, the lack of diversity in the
artificial intelligence field has peaked. The diversity issue has contributed to the flawed systems
that have influence racial and gender biases, according to more than 150 studies and reports. The
Artificial Intelligence field, which is white and male-dominated, risks replicating and
perpetuating power imbalances and historical biases in the world today. Examples of the lack of
diversity can be observed in image recognition services not recognizing users of darker skin
tones and the image recognition services making offensive classifications of minorities. The AI
industry needs to acknowledge the seriousness of the situation and develop new means and
methods to address it.
It is essential to note that more than 80% of Artificial Intelligence professors are men.
Approximately 15% of AI researchers on Facebook are women. The makeup of the Artificial
Intelligence department across the world reflects the large problem across computer science
systems and society. Women comprise only 24% of the population in the field of computer and
information sciences (Pugh et al, 2015). Additionally, only 4% of the Artificial Intelligence
workforce on Facebook is black, and little to no data exists on the makeup of trans-workers in
the organization. The urgency and seriousness of this issue increase as Artificial Intelligence are
continually integrated into society. Essentially, the lack of diversity in Artificial Intelligence
machines is increasing a significant amount of power and capital to a specific group of people in
the larger population.
In 2018, venture capital funding for Artificial Intelligence startups reached record levels
in 2018 and increased in 2017 to 9.33billion dollars in funding. In the United States, startups
increased to 113% as more money and resources were injected into the field. The solution to the
ARTIFICIAL INTELLIGENCE
9
issue of diversity in the Artificial Intelligence industry does not only involve solving the
demographic makeup of most AI teams. It requires an in-depth solution and the identification of
the root cause of the problem. Men comprise approximately 71% of the application pool for AI
jobs in the United States, according to the 2018 AI index report that is released annually. The AI
institute suggested a solution to this problem by suggesting the publication of compensation
levels for workers publicly and dealing with most organizations’ hiring practices when it comes
to underrepresented groups at all levels.
One issue that influences the lack of diversity in AI is the lack of diverse thoughts. The
population creating the technology is significantly homogenous, leading to the homogeneity of
the products produced. In as much as these AI people might have good intentions of serving
everyone, their inside biases will be reflected in the design and products produced. As AI
engineers write algorithms, their biases tend to show up when it comes to the algorithm design
decisions. Most AI tools are driven by data that comes from the society. The society possesses its
own in-built biases, which have been influenced by so many factors over time. For example,
when designing a tool that assesses the creditworthiness of a person, the historical biases will be
copied and integrated into the machine to provide an accurate answer.
Artificial Intelligence has shown immense potential despite its being a recent integrated
technology in education, health, urban, and mobility. The pace at which these technological
advancements are advancing suggests that AI and machine learning will modify the way we live
and work in the coming years (Leavy et al, 2018). As a result, Facebook needs a fast solution to
the AI challenge it currently faces. It is essential to note that this issue may hold potential
negative consequences for the organization. First, people around the world are woke and have
quickly recognized the importance of diversity and inclusion. As a result, diversity has been a
ARTIFICIAL INTELLIGENCE
10
sensitive and controversial topic. We have seen customers boycotting companies that seek to
degrade one race or suggest another race’s superiority. The AI issue of diversity has shown
probable chances of degrading the Black race and influencing feelings of white superiority.
Facebook has always prided itself in influencing people all over the world. If this is not checked,
the divisions brought about by these changes may influence some of the application consumers
negatively. The platform enjoyed by so many people may be divided, with only a few people
subscribing to it and enjoying its content.
Problem statement
The lack of diversity in AI and machine language due to the Computer science algorithms used
and the composition of the tech-engineers.
Data collection methods Identified.
`
The most appropriate data collection method identified is digital ethnography.
Ethnography refers to the study and observation of people in a real-world environment. This
research methodology enables researchers to observe the participants in their natural
environment. It goes against normal methods of research by bringing the researcher to the
participant (Pink, 2016). Digital ethnography is simply the digital evolution of ethnography.
This type of methodology opens up new opportunities for researchers and enables them to
understand the participant’s behavior better than before. Researchers can easily peek into the
lives of the participants without necessarily needing to be there physically. It is discrete, which
enhances the accuracy of the results generated.
There are various ways through which digital ethnography can be carried out. This
provides the researcher with flexibility and also makes the research more fun and engaging. The
ARTIFICIAL INTELLIGENCE
11
best thing about this research methodology is that people are on their phones, 80% of the time
nowadays. Digital ethnography is a natural research tool that makes research effortless because
the participants are aware of what is required of them. It easily fits into people’s lives as they go
on with their daily activities. There are various advantages associated with using digital
ethnography; first, it helps in the generation of honest and insightful results. The most significant
aspect associated with digital ethnography is the independence it offers. The participants are
allowed to use their own devices, which eliminates the need for a third-party observer. Secondly,
this type of methodology provides in-the-moment feedback. The researcher can gather customer
insight in the specific time they interact with the products as opposed to the time taken by other
types of methodologies. Thirdly, this type of methodology provides flexibility and can be used to
obtain results over a wide geographical area.
One aspect of digital ethnography is observation. The researcher has the ability to
observe things as they unfold and come up with the most accurate conclusion. This type of
methodology is suitable in this case because of the high amounts of observation required. The
researcher needs to identify how the AI works and where the problem of diversity originates
(Varis, 2016). The researcher must also understand how this affects the under-presented groups
and how over-presented groups perceive this. By understanding this, the researcher understands
the true impact of AI and machine learning on diversity. Observation can also contribute to the
development of accurate solutions needed to solve the identified problem.
Solutions Identified
While it is incredibly important to have a diverse set of employees, the organization also
needs people who can pose questions about the application’s data and algorithms. It is not easy to
take the race column out of the algorithm. However, algorithms quickly learn these biases
ARTIFICIAL INTELLIGENCE
12
through proxies and correlate this to factors such as race or gender in the data. The application
data need to be checked for potential biases to implement the necessary safeguards needed to
protect against these biases (Gong et al, 2019). Human biases are well documented right from
implicit association tests that demonstrate the biases we have unawares. To identify the best
solution possible for the problem of biases in AI, we first need to identify the ways in which
these biases make their way into our algorithms. AI systems learn the decision-making process
based on human interactions and reflect the social and historical inequities that exist in the world
today. For instance, Amazon stopped employing algorithms after learning that it favored
applications that used words such as ‘executed’ or ‘captured.’
Another source of bias in AI is flawed data sampling in which specific groups are under
or over-presented. For example, Joy Buolamwini, an MIT working with Timnit Gebru,
discovered that facial analysis technologies showed high error rates for minorities and
particularly minority women due to the unrepresentative training data. The society shoulders the
responsibility of bias. It reduces people’s ability to participate in the society and the economy. It
reduces AI’s potential for the society and business and encourages the distortion of results.
Businesses and organizations need to ensure that AI systems employed to improve human
decision making and encourage the research progress.
From the progress made in academic research on AI bias, two solutions have emerged.
First, human beings must take advantage of the various ways AI potentially enhances human
interactions (Daughtery et al, 2019). Machine learning systems neglect the variables that
inaccurately predict outcomes. It also makes it easy to probe algorithms for bias, which helps
reveal the human biases that had initially unnoticedâ€â€integrating AI in decision-making benefits
disadvantaged groups by eradicating human biases. The second imperative pertains to speeding
ARTIFICIAL INTELLIGENCE
13
the progress in addressing the biases that exist in AI. Finding a solution for the AI problem does
not require a quick fix, as understanding and measuring fairness can be hard. Researchers have
come up with technical ways of measuring fairness, such as employing predictive values across
groups. However, this introduces one significant challenge: different fairness definitions cannot
be easily satisfied. However, researchers have made significant progress in ensuring that the
techniques employed help these AI systems meet all these definitions of fairness. One promising
technique that has been employed in this area is counterfactual fairness. This model ensures that
the decisions made are counterfactual with factors that are deemed sensitive such as gender,
sexual orientation, and race being changed. There has also been the development of the pathspecific model that helps deal with the process of fairness in complicated cases.
While these improvements might help, some challenges require more enhanced and
technical solutions. For instance, researchers need to determine when a system seems fair enough
to be released and which situations are fully automated and should be permissible. Organizations
such as Facebook are advised to establish a responsible process that can help mitigate bias. For
instance, the organization can employ the use of a portfolio of technical tools such as third-party
audits to help deal with these problems. Tech companies are providing solutions for these
problems and coming up with new ways in which organizations can successfully employ AI and
machine learning.
Another solution to this problem is engaging in fact-based conversations that surround
the issue of human biases. Proxies such as procedural checks have been widely employed when
it comes to deciding the fairness of human decisions (Cohen, 2019). With the advancements of
tools to help probe for biases in machines, the standards in which we hold these machines can
inherently be raised. For instance, running algorithms alongside human decision-makers can
ARTIFICIAL INTELLIGENCE
14
prove beneficial and employing techniques such as éxplainability’. Another potential solution is
to consider how humans and machines can work together to mitigate bias. Some ‘human-in-theloop’ systems can make suggestions and provide recommendations for human beings to choose
from. This can improve human understanding pertaining to Artificial Intelligence and can lead to
the development of better algorithms.
The last identified solution is investments. Making more investments in AI and providing
significant data can help bridge the problem of bias in research. It can help generate information
such as how to make a designer’s choices more transparent and ethical in the development of
these machines. Investing more in AI could also mean diversifying the field. A more diverse
community would be better suited to anticipate and spot biases that exist in the algorithms. This
solution suggests the employment of a diverse population when it comes to these algorithms.
Artificial Intelligence can help in the identification of solutions that most societies face.
Facebook needs to choose one possible solution that could contribute to the successful
integration of AI technology.
Approach
The contribution of AI in today’s world cannot be assumed. Machine learning and AI
have influenced significant growths and aided in today’s world’s flexibility and convenience.
However, these advancements also pose potential risks and challenges for organizations. There
have been significant developments when it comes to dealing with the issue of diversity. The
world has slowly realized the importance of diversity and appreciating each other’s differences.
However, the integration of these AI machines poses the threat of taking us back to the years
where there were major divisions over race and gender. This issue is greatly attributed to the fact
that most AI engineers are while and males, which translates to most of the products they
ARTIFICIAL INTELLIGENCE
15
develop. However, solutions to these challenges need to be identified and the necessary measures
employed to help deal with the problem of diversity. One such solution is assessing the
technicality of the algorithms used and influencing fairness and equality in the AI industry.
While these solutions do not 100% guarantee results, they show great potential in influencing
positive results in the AI industry.
16
ARTIFICIAL INTELLIGENCE
Solution Identification
Diversity is an important concept in computer science. Diversity in computer science
reflects in the products and services delivered. Many people are unable to enter the field of
computer science due to wrong perceptions on the part of the managers, teachers, and
communities. Artificial Intelligence technologists are not nearly diverse enough to cater to the
concerns of the populations affected. New programs have been identified to build a pipeline of
diverse AI talents. By identifying and addressing diversity gaps in the community today, tech
leaders can identify and mitigate the bias in the systems designed for tomorrow. Women’s
representation in tech-related jobs has decreased by 32% since 1990. In a study conducted in
2014, women only held 25% of computer-related jobs (Stathoulopoulos et al, 2019).
While there is little to no data on the effect of coronavirus pandemic on the computer
science field, recent reports suggest that women are losing ground in the workforce. For instance,
the percentage of black and Latinx employees in the Computer Science field is extremely low,
with women occupying 1-3% of the tech workforce. Diversity is an issue in the entrepreneurs
and venture capitalists who are deeply invested in this industry.
There are various computer science principles and concepts that apply directly to this
problem of diversity. One such principle is digital information. Computers store their
information in bits. This is a series of on and off states that are identified by ones and zeroes.
Computers use binary language and information images such as audio files and texts to store bits
of information. Digital information is the basis of all computing, including programming. It
covers things such as binary and bits, digital communication, encoding text and compressing text
and images. Artificial Intelligence works by combining large amounts of data with intelligent
algorithms, which automatically allows the software to automatically learn the various features
ARTIFICIAL INTELLIGENCE
17
of the data (Housers, 2019). The development of AI is highly dependent on the data encoded by
these AI tech engineers. Dealing with the information fed to these computers addresses one
aspect of machine learning. These AI engineers need to be educated and informed on the types of
information and algorithms to use in AI development.
The second principle identified that could help with the identification of a solution to this
problem is programming. Programming refers to the process of creating a set of instructions that
instruct a computer how to do things and perform a task. Computer programming can be done in
various ways but by using computer programming languages such as JavaScript and Python. AI
programming is an advancement of technology that has influenced efficiency and optimum
benefits for various organizations. AI has introduced a new level of smart technology to different
industries and increased the potential for most Organization success. Developers are willing to
experiment, explore and implement the various capabilities to satisfy human and organization
necessities. In the development of software applications, developers make use of various
languages in writing AI. As much as there is no perfect programming languages used in artificial
intelligence, developers should identify the most effective and suitable languages. The
developmental process relies on the desired functionality of the AI application being created. It is
essential to note that most organizations, including Facebook, look for functionality in their
systems. The tech people developing these software types should identify the most efficient
programming language and type to use. For instance, python encourages programming and is
less complex compared to other programming languages. It supports various programming styles
and is a multi-paradigm type of programming. It provides these computer scientists with
convenience and the ability to style systems that meet the needs and requirements of the
customers.
ARTIFICIAL INTELLIGENCE
18
Another crucial concept of computer science is algorithms. Algorithms can be defined as
a series of instructions that influence a step-by-step process that helps in the solving of a
problem. There are various algorithms employed in the development of Artificial Intelligence.
Algorithms are used to divide a variable into different classes and predict the class for a given
input. For instance, classification algorithms can be employed in the classification of emails as
spams or not. The algorithm to put under study is the regression algorithm. These types of
algorithms are popular under supervised machine learning algorithms and predict the output
values based on the data injected.
This type of algorithm can be used in machine learning to identify the functionality and
usability of the systems developed. Regression algorithms are used in predictions and
forecasting; in essence, it can contribute to the AI industry by helping them predict the usability
and diversity of their systems. One important concept to understand is algorithmic bias- this
comes about when writing these AI systems’ algorithms. This problem influences the population
in general since only a percentage of the population is well catered for in these algorithms. AI
techs need to ensure that the algorithms used in writing these systems cater to the needs of the
diverse population.
Artificial Intelligence systems are designed to assist humans through the integration of
behavior such as problem-solving, planning and learning. The technological industry is
increasingly adopting AI to influence performance for many organizations from predictive
analysis to data sorting. Overall investing in AI is expected to increase to approximately
$190billion globally by 2015. Human technologists are increasingly contributing to this growth
and fueling this aggressive development needed to build perfect algorithms that form the basis of
decision making. However, these technologists are not diverse enough to take into consideration
ARTIFICIAL INTELLIGENCE
19
the concerns of the affected populations. As companies and organizations are looking up to the
future in AI jobs, they need to ensure and implement measures that promote diversity and don’t
accelerate algorithmic bias.
The AI industry needs to build a pipeline of diverse talent. It is essential to understand
that the solution does not only begin in the AI employment industry. The solution starts from the
institutions of learning in which AI is taught. The inequality in the AI space starts right from the
school institutions. There are various measures that can be implemented to help with the issue of
diversity in schooling institutions. For instance, a small number of young people can be selected
to participate in AI summer internship programs. New graduates need to be connected with local
talents and industry experts to guide them and provide them with employment opportunities in
the AI industry. Students need to be provided with hands-on experience on machine learning,
data analytics and visualization. Under-presented groups face structural barriers in accessing
computer science programs. This creates disparities and prevents diversity in the AI industry. For
instance, Black students are less likely to have CS classes as compared to White students.
Additionally, most people who have the knowledge and understanding of AI did so in school.
However, Hispanic and Black students are more likely to learn CS outside the classroom than
White students. Black students are also disadvantaged in that they are less likely to access
computers at home; this factor is directly associated with their dwindling confidence levels in
CS.
Under-presented groups also face the social barrier challenge in learning CS. There is a
huge perception that CS is only meant for certain groups, specifically the Whites and Asian
males. As a result, this discourages other groups from pursuing CS, which significantly impacts
the makeup of the AI tech industry today. Additionally, organizations are included in this cloud
ARTIFICIAL INTELLIGENCE
20
of perception and tend to hire from the stereotyped groups and ignore other groups. Female
students are less likely to delve into the field of computer science and attempt to take
opportunities that present themselves in the Computer science world. Students need to be
provided with hands-on experience of computer science; everyone should be given a chance to
delve into the AI world without the fear of discrimination and segregation.
The automation of automation is also an important concept that can help solve the issue
of diversity in Artificial Intelligence. AI tech needs to consider the possibility of automating
processes and creating new forms of machine learning in automation software. For instance,
organizations can implement a multi-layered approach to computer science to enable the opening
up of new branches of study. This will create an entirely new way to identify problems and the
possible solutions in the computer industry (Strainer, 2019). As a computer scientist, the task and
responsibility of predicting and adapting to the massive changes expected in the industry in the
coming years fall upon you. AI is expected to become even more in demand; this creates the
necessity for diversity. One factor to consider is the White and Asian male population that makes
up the AI industry cannot cater to the needs and wants of the whole population. Even if they
discovered new ways of automating highly unlikely machines, the demand and need for AI in
other industries are increasing. AI will become more complex in time; meeting the population’s
diverse requirements requires efficient and strategic measures.
Facebook CTO Mike Schroepfer emphasizes that hiring is a vital aspect of diversity in AI
and prevents bias for AI techs building products for users. However, he fails to shed light on the
number of black people who work in Facebook AI research. Facebook has recently reported
employee diversity in huge numbers for six years; however, it does not shed light on individual
teams’ tally diversity statistics. Facebook states that it takes a lot of consideration on improving
ARTIFICIAL INTELLIGENCE
21
diversity across the Organization (Khari Johnson, 2020). The CTO identifies that the real
solution to these problems is implementing and using a diverse data set in the Organization. This
formalizes the processes in the Organization and statistical methods used to determine the
representative data set. According to a 2019 report released by Mutale Nkonde, Facebook AI
research did not comprise any Black employees. This influenced the development of a lot of
policies and regulations, such as the Algorithmic Accountability Act. This act was developed to
ensure that corporations assessed AI safety and security, including the biases in organizations
such as Facebook.
Bias is also found in AI because the training data lacks a comprehensive representation of
users. Policymakers insist on hiring diverse and pluralistic teams that will ensure the creation of
AI with more people in mind. Following George Floyd’s death and protests all over the country,
more emphasis is being directed towards institutionalized discrimination. Facebook had had
some troubles when it came to protecting its stand on diversity. For instance, Mark Zuckerberg
defended a racist tweet by Donald Trump, which suggested that the Black community should be
silenced over their demonstrations and riots (Khari Johnson, 2020). Facebook is also reported to
have fired an engineer that participated in the demonstrations; this is not a good look for the
Organization and the strides it has made so far to present itself as a diverse Company. However,
Facebook continues to defend its position and emphasizes its effort to build a responsible AI
society. For instance, Facebook has increased its content moderation and is increasingly working
with innovation teams on subjects like security, integrity and algorithmic fairness.
Lastly, one of the most important factors to consider is to build tools that work and that
people employ by default. This makes it representative by default. The more representative, a
team is, the better the perspectives and needs of the users are taken into consideration. This
ARTIFICIAL INTELLIGENCE
22
diversity is incorporated into the products built and, in the systems, employed. Ensuring fairness
in the AI industry is the first step towards fighting institutional discrimination and creating AI
systems that are supportive and cater to the needs of the entire population.
Hypothesis Statement
Changing the algorithms and datasets used could help identify a solution for the problem of lack
of diversity in AI and machine learning
Research Question
How will changing the hiring practices influence the algorithms and datasets used and
help solve the issue of lack of diversity in the AI industry?
Related Works
Adams, B., & Khomh, F. (2020). The Diversity Crisis of Software Engineering for Artificial
Intelligence. IEEE Software, 37(5), 104-108.
The article sheds light on the diversity issues experienced in the AI industry. These
challenges pertain to race, gender, geography and other associated factors. The article suggests
that the datasets used should be carefully selected to avoid incorrect and harmful consequences
to the unassuming end users. The article suggests that teams involved in the building of AI
products should be made more diverse because of the multiple biases that are neither
straightforward nor easily identifiable. The solution proposed by this research is different from
the one identified in my study. This solution does not take into consideration the role of
organizations and companies in implementing policies and procedures that promote a diverse
workforce.
ARTIFICIAL INTELLIGENCE
23
Houser, K. A. (2019). Can AI Solve the Diversity Problem in the Tech Industry: Mitigating
Noise and Bias in Employment Decision-Making. Stan. Tech. L. Rev., 22, 290.
This research suggests that mitigating noise and bias in employment decision making will
be an effective way of solving the problem of lack of diversity in the AI industry. The study
identifies the underrepresented populations, i.e. women and how these factors will directly
influence diversity in the AI industry. While this study shares a lot of similarities with my
research; there are also some differences to the solutions identified. My study proposes the
solution as publishing harassment and discriminatory policies while this research suggests
influencing the decision-making process.
Stainer James (2019) we must fix diversity problems.
The article identifies the need for fixing diversity problems. It provides a summary of
research on the lack of diversity in the AI industry, including gender and racial diversity. It
identifies various solutions for the problem of diversity in AI. However, one major factor has
been emphasized; solving the problem of bias in classification. The article states that AI tech and
engineers should be extremely careful when classifying people on the products they develop. The
bias of classification cuts across all forms of diversity and affects people globally. The solution
identified by this study is different from mine. My study focuses on creating a diverse workforce
by implementing various policies and measures to ensure organizations uphold diversity.
Stachowiak, R. (2019). Why Diversity Matters. In Remaining Relevant in Your Tech
Career (pp. 41-52). Apress, Berkeley, CA
This article identifies the importance of diversity in the Organization. It also identifies
how diversity directly relates to the success of an individual’s career and how it influences their
ARTIFICIAL INTELLIGENCE
24
performance. By identifying the importance of diversity, the article sheds light on solutions that
can be implemented to create a diverse environment. For instance, the article points out the
importance of hiring a diverse workforce and taking significant care in the selection and hiring
process.
Stathoulopoulos, K., & Mateos-Garcia, J. C. (2019). Gender Diversity in AI Research.
Available at SSRN 3428240.
This article focuses on the lack of gender diversity in the artificial intelligence workforce.
The article reports on research conducted and interviews that shed light on the gender disparities
in the AI workforce. These interviews shed light on the problems women face when it comes to
the AI industry. It also factors in some solutions these interviewees think that will promote
diversity in the AI industry. One such factor identified is the hiring process; the interviewees
suggested that organizations need to consider their hiring processes more. However, the article
did not identify any mitigating factor that would help resolve this issue, such as the
implementation of the necessary policies and procedures.
25
ARTIFICIAL INTELLIGENCE
Research Methodology
Research matters
Research in AI bias reveals that there is a great irregularity in the products as a result of
the machine learning algorithms employed. This is directly associated with the biased
assumptions that are made during the development of an algorithm and prejudices in the training
data. There are various types of AI biases, including cognitive biases and training datasets which
introduce these biases. Cognitive biases refer to our innate feelings towards an individual or a
group of people as a result of their perceived stereotypes. There are approximately more than 180
human biases that have been recognized by psychologists (AI Multiple, 2021). Each of these
biases can directly affect an individual and how we make decisions; as a result, these biases can
easily seep into machine learning algorithms through the designs or data sets used. Another
technical issue identified that might influence the lack of diversity in AI machines includes the
use of incomplete data. Suppose the data and information used is not complete. In that case, this
might directly affect the AI and machines developed by presenting only a specific segment of the
population rather than the whole group.
Research Methodology
The type of research methodology to be used in this case is experimental research. It is
essential to understand that to prove the validity of the solution identified, experimental methods
must be undertaken. Experimental research is usually conducted with a scientific approach that
employs two groups of variables. The first set presents a constant that can be used to ascertain
the differences between the second set. A constant is used as a form of control experiment during
research (Lipsey, 1990). It is essential to note that any research that is usually conducted under
any scientifically standard conditions employs the use of experimental methods. The success of
ARTIFICIAL INTELLIGENCE
26
experimental methods is highly dependent on the variables used; the control variables and the
experimental variable (Lipsey, 1990). In this case, the constant would be the current state of AI
and machine learning with respect to its lack of diversity. On the other hand, the variable would
be testing whether changing the algorithms and the dataset used by changing the hiring
procedures would help generate a solution for the problem identified.
There are various types of experimental design. The most common ones include preexperimental research design, true experimental research design, and quasi-experimental
research design (Lipsey, 1990). The group subject classification which is normally influenced by
various conditions determines the type of research that is employed. True-experimental research
design is highly dependent on statistical analysis to ascertain the validity of a hypothesis
(Fjermestad, 1998). The hypothesis that was developed from the research question was whether
changing the algorithms and datasets used could help identify a solution for the problem of lack
of diversity in AI and machine learning. True experimental design is considered as the most valid
type of experimental research and is mostly carried out with the absence of a pretest (Huck,
1991).
In this case, two groups were identified; one control group and the actual group. The
control group consisted of AI techs that were randomly selected from various organizations. It
also included the head of these AI groups as he had the power to hire and recruit new members.
This was the control group; the dynamics of the group were not changed, and the way they write
their algorithms nor the datasets they use interfered with (Huck, 1991). The second group was
the variable being tested, comprised of an AI tech department and additional diverse recruits.
These recruits comprised of African Americans, women, and other unrepresented groups. The
algorithms and datasets used in this case were not changed neither interfered with. One factor to
ARTIFICIAL INTELLIGENCE
27
consider is that the root cause of the problem resulting in the lack of diversity in AI and machine
learning is the tech industry’s composition. Thereby, introducing a group that is balanced brings
with it, new algorithms and datasets that can be used to solve the issue.
The main reason for choosing this type of research methodology is because of the
potential advantages it holds towards my research. First, I have a strong hold over the variables
used to obtain the desired results. In as much as this is an experimental procedure, I can change
the variables anytime and obtain the results expected (Huck, 1991). Secondly, the results
obtained are specific and help inform whether the identified hypothesis can provide a possible
solution for the problem faced (Lipsey, 1990). Thirdly, it provides a foundation for future
research to apply the findings to similar situations and instances. For instance, if the results
obtained prove that diversity in the AI industry can directly influence the algorithms and data
sets employed, this can be used in other industries and emphasizes the importance and necessity
for diversity.
Experiment design
The experiment design to be used in this case would be the matched pairs design. This is
a design where the participants are matched to their key variables. There are various variables in
research; in this case, the variable is the algorithmic language and the data set used. Diversity is
also a variable that will be used when assigning participants to their groups. The matched pair
experimental design can be used in differentiating the control group from the experimental
group. This helps in segmentation and the collection of accurate results (Imai, 2008).
ARTIFICIAL INTELLIGENCE
28
Requirements and Constraints of the experiment
It is essential to understand that there are various requirements for this experiment. First,
the experiment is to be conducted using two groups of participants (Lipsey, 1990). One group
consists of the current makeup of the tech industry, while the other group is to have certain
modifications. The products developed in AI reflect the composition and make-up of the tech
industry. If a solution is to be identified, the make-up needs to be first solved, to help with other
technical stuff such as the algorithm and dataset employed. Secondly, the experiment is to be
conducted on a blindfolded basis (Gribbons, 1996). One of the experiment features is observing
the behavior and actions of the tech industry in terms of the algorithms and datasets used. If one
group was to be aware that they are observed, then the validity and accuracy of the results
obtained could potentially be damaged. In order to obtain accurate results, some aspects of the
research need to be kept hidden.
The research requires constant observation and monitoring. This implies that various
researchers will be involved in the research process. The participants will go on with their dayto-day activities as usual, with the researchers being deployed in their organization of operation.
For example, Facebook has made various strides in integrating AI into its systems. This
organization can be used as a studying point for the research. The control group can consist of
the tech department on Facebook, while the experimental group can come from various
departments of different organizations. One aspect needed in the experimental group is diversity,
as this will reflect in the algorithm and datasets employed. Therefore, the experimental group
will consist of a diverse population so as to introduce a differentiating factor in the experiment.
Another requirement of the experiment is past information and knowledge on the issue. This
will provide a basis on how to conduct the research and provide a foundation for future research.
29
ARTIFICIAL INTELLIGENCE
This helps inform the research and helps establish what factors need to be placed as a priority
during the research (Gribbey, 1996). It also serves as a background analysis where the researcher
can identify past shortcomings and devise ways on how to limit them or overcome them.
The sample size presents a significant limitation for the research. Due to the large sample
size and the segregation of the research into two groups, more time and resources will be
required. A lot of time will be wasted in the data collection process and obtaining the research’s
required information. The researcher is required to carefully collect data from each sample group
and tech engineer. Another constraint associated with a large sample size is resources (Francis et
al, 2010). It is essential to understand that in as much as people want to solve this problem of
diversity, little to no resources have been dedicated to carefully analyzing this problem to find a
suitable solution. The sample size is a limitation since the available resources are already
strained.
Experiment Procedures
The first step in the experiment is to identify the sample population to be used. The two
groups need to be identified and defined. Secondly, the organizations that would be involved in
the research need to be notified. The research does not seek to segregate the sample population
but rather observe them in their natural environment. Hence organizations such as Facebook
need to be notified of the research. Facebook and other organizations, including Google, have
come forward to state their support for diversity in AI and machine learning. It is therefore
expected that obtaining permits to be used during the research will be an easy task.
Thirdly, the researchers are to select the population to be used in the experimental group.
The experimental group is supposed to be a diverse group of tech-engineers. This will require
ARTIFICIAL INTELLIGENCE
30
time and keenness during the selection process. The hiring process is to be reviewed in this case;
the practices of various departments will be identified, and the pool of potential employees
analyzed. This will help identify the diverse and prospective employees who showed interest in
working in the AI and tech industry. It will provide the required population for the experimental
group.
The next step is to collect the necessary resources needed for the research. While the
control group has the necessary resources in performing their activities, the experimental group
does not have the required resources for the experiment. This issue can be solved by forming a
merger with an organization in the research process. This has to be an organization that values
diversity and actively employs AI and machine learning in their systems. This organization will
provide the necessary resources needed for the experiment. The last step is to conduct the
experiment. The time period for the experiment will be established once the experimental and
control group has been identified. The experiment will be conducted following the requirements
stated above. The results obtained will be applied to various business sectors and provide a
foundation for future research.
31
ARTIFICIAL INTELLIGENCE
Project Plan
The intended audience for this project is all the stakeholders in companies implementing
AI systems. The stakeholders of different companies, including Facebook, are ideal in this
project as the target is on diversity. This project’s objectives are achieved by solving diversity
contributed by the hiring process in Facebook and other organizations. The plan for this project
includes five significant phases. The first phase is the initiation and conception phase, where the
focus is on the expansion of different ideas and meeting any demands that relate to the Project
(Ceschi et al., 2005). In this phase, the focus is also on the exploration of how a solution will be
achieved based on the data collected.
The second phase is the planning phase, where the emphasis is on the scope of the project
and the calculation of the schedule and budget. In this phase, the different tasks to be carried out
are listed in order of importance, and an estimate of the time and budget is brought out (Ceschi et
al., 2005). The third phase is the launch phase. This is the ultimate phase in that it determines
how tasks and resources are to be allocated. In this phase, the urgent tasks are prioritized, and the
calculation of the budget and a schedule for the project. An adjustment for changes to be
implemented later is accounted for in this phase of the project.
The fourth phase is the performance and control phase. In this phase, the project plan’s
progress is monitored and the status updated (Ceschi et al., 2005). In this phase, real-time events
are tracked based on the tasks completed and those yet to be completed. An accurate picture of
the status of this project is brought out in this phase. The last phase is the closure phase, where
clean up duties are allocated to specific timelines. As this is the final phase, a realistic snapshot
of the project is created.
ARTIFICIAL INTELLIGENCE
32
Project Plan for Data Collection & Analysis
In order to collect important data for this project and verify it, an analysis is an effective
data plan. An analysis is significant as it estimates the specific outcomes of a project and the
limitations that might arise (Moser & Korstjens, 2018). The data collection and analysis process
is focused on the achievement of the following tasks.
Tasks
•
Task 1: Obtaining and preparation of data related to diversity in artificial intelligence in
social media platforms
•
Task 2: collecting data on the challenges faced by Facebook in the implementation of
Artificial intelligence in their systems
•
Task 3: Analyzing the extent that minority groups have been affected by the lack of
inclusivity
•
Task 4: Identification of different data sources that relate to the lack of inclusivity in
artificial intelligence
•
Task 5: Analysis of Facebook artificial intelligence workforce and its impact on the issue
of diversity
•
Task 6: Collection of data from the sample of social media users observed in this project
•
Task 7: Conduct a review of research sources available on the issue of diversity in the
artificial intelligence field
•
Task 8: Analysis of data collected using the observation method on a sample of the
population
•
Task 9: Comparison of the percentage of presentation of the minority against the majority
on the Facebook social platform.
ARTIFICIAL INTELLIGENCE
•
33
Task 10: Presentation of the results
Timeline and Human Resource
The timeline for this project is brought out in the figure below.
The ten tasks will be completed within a period of ten weeks. For every task, one week is
allocated, which shows the ample achievement of the project’s set objectives. However, for the
successful handling of the different activities, an experienced human resource section is required.
The human resource team comprises individuals who are tasked with various responsibilities
based on their experiences and qualifications. The key component to the success of any project is
the human resource. In this project, a human resource management plan aids in the effective
management of tasks throughout the ten weeks. The specific roles and responsibilities of the
project team are:
•
Diversity and artificial intelligence data preparation
•
Data collection
•
Identifying the different sources of data
ARTIFICIAL INTELLIGENCE
•
Collection of data by observation
•
Analysis and comparison of data
•
Presentation of results
34
The organizational chart that best suits this project has two levels. The first level is the
manager level. At this level, the human resource in charge has the responsibility of providing
instructions. In the second level, the individuals work in groups to deliver results from research
carried out. The sequence of tasks depends on the available resources and the available methods
of data collection. On analysis of the project, it is evident that there exist special roles that should
be considered for the realization of the objectives. The core responsibility of the research group
is the collection of data from different sources.
The acquisition of resources is important for the effective performance of the human resource
team. The resources will be acquired by the form of purchase from online sources. The important
sources include the online reading texts and previously designed systems that can be applied in
this project. However, there are some additional resources significant in the fulfillment of the set
objectives. On this, the emphasis is on the creation of a budget that caters to the resources and
activities of the project.
The completion of tasks will involve separate carrying out of activities by the human
resource team. The team will however, handle separate tasks based on the specific timeline
available. The distribution of the human resource involves the delegation of duties for the
successful completion of the tasks. From week one to week five, the human resource team’s
focus is on collecting and analyzing information and data collected. From week six to week ten,
ARTIFICIAL INTELLIGENCE
35
the focus is on reviewing, analyzing, and comparing the gathered data or information. Basically,
the tasks are completed in team forms.
Estimated Costs
The estimated costs in a project include the resources and costs that are required for the
completion of a project within the set time. The different cost estimation accounts are important
for the calculation of the total amount that determines the budget to be set. The costs related to
the performance of the different tasks throughout the ten weeks vary as a result of the method
implemented, the team involved, the activities to be carried out, and the resources used. For the
ten weeks, the estimated financial cost is $2540. This cost caters for the different elements
required for project completion.
The intensity of the tasks to be performed contributes to differences in the financial
resources required (Gupta et al., 2010). For task one, financial and human resources are required.
On estimation, the costs incurred for the completion of the task is $160. The second task involves
the collection of data, which is set to incur $130. Task three involves going overboard and using
different data analysis tools to analyze the data collected. In addition to the human resources
involved, the estimated cost is $250. This caters for the different finances and any additional
activities to be carried out.
Task four involves identification. In the completion of this task, the financial costs
incurred relate to human resources. The estimated cost is $100. The fifth task focuses on the
analysis of the Facebook workforce in terms of the data available in the specific statistical
sources. In the completion of this task, financial and human resources are required. The financial
cost for the task is estimated at $560. Task six involves the collection of data from different
ARTIFICIAL INTELLIGENCE
36
social media platforms. The estimated cost is $230. Task seven involves the application of
financial resources which is an indication that more emphasis should be on the finances. The
estimated cost for the task is $300. Task eight involves the use of different data analysis tools.
The cost estimated with respect to this task is $380. The ultimate task that determines the success
of this project is task ten. In this task, the focus is on the finances. The costs incurred are
estimated at $430.
Risk and Cost-Benefit Analysis
The risk and cost-benefit analysis is a suitable method for the evaluation of the
desirability of different actions and costs with respect to a project (Pearce, 2016). The risks that
relate to this project are determined in terms of their probability. The risks related to this project
arise from conducting the different activities for the completion of set tasks. One of the risks is
the ow returns of scale on the collected data. The collection of data may not lead to the creation
of more value, but there may be a decrease in the relevance of information.
Another risk is the loss of revenue as a result of the breach of different aspects in the data
collection process. The revenue is lost through the high costs incurred when carrying out the
activities of the project. The loss of revenue is also a result of the breach of trust by this project’s
participants. Due to asking the same questions to the participants, there may be questions
regarding its use. The third risk is difficulty in prioritization. There are different aspects
considered in this project when carrying out the tasks. Therefore, there may be issues in
prioritization, especially when focusing on important information or data.
Conflict of interest is a significant risk that relates to this project. When conducting the
tasks, there are issues related to striking a working balance between maintaining the trust of the
ARTIFICIAL INTELLIGENCE
37
participants and maximizing the value of the project. There is a risk of relationship damage with
the participants when the value of a project is maximized. On the cost-benefit analysis, it is
evident that the main costs to be incurred for this project include the financial costs, human
resource-related costs, and sunk costs. As the focus is on the data collection and analysis, the
costs incurred are lesser as compared to the benefits.
The estimated financial costs include $2540. In addition, more finances are channeled to
handle the sunk costs incurred as a result of different decisions made. On analysis, the benefits
outweigh the costs on the successful completion of this project. One of the benefits of this
project is data possession regarding the presentation of diverse groups in artificial intelligence.
Another benefit is the realization of the objectives of this project. The costs incurred are
significant to realizing the objectives on the successful coordination of the ten tasks.
Cost-benefit analysis involves focusing on the direct and indirect costs and comparing
them to the benefits (Pearce, 2016). The different costs that relate to this project are also
compared with the cost-benefits analysis. With the acquisition of the information from the
different data collection methods, it is evident that the benefits are 50% more than the costs. On
this, this project is worth investing more as the returns are evident. The financial, human
resource, and sunk costs can be handled within ten weeks to ensure benefits are achieved for the
project. There is a probability that this project will be extended, but this will not affect the
current ratio of costs and benefits.
ARTIFICIAL INTELLIGENCE
Conclusion and Future Works (Week 5)
Conclusion (Week 5)
Limitations (Week 5)
Future Works (Week 5)
38
39
ARTIFICIAL INTELLIGENCE
References
AI Multiple (2021) Bias in AI: What it is, Types & Examples, How & Tools to fix it
https://research.aimultiple.com/ai-bias/
Cohen, T. (2019). How to leverage artificial intelligence to meet your diversity goals. Strategic
HR Review.
https://www.emerald.com/insight/content/doi/10.1108/SHR-12-2018-0105/full/html
Caers, R., De Feyter, T., De Couck, M., Stough, T., Vigna, C., & Du Bois, C. (2013). Facebook:
A literature review. New media & society, 15(6), 982-1002.
https://journals.sagepub.com/doi/abs/10.1177/1461444813488061
Ceschi, M., Sillitti, A., Succi, G., & De Panfilis, S. (2005). Project management in plan-based
and agile companies. IEEE Software, 22(3), 21-27.
Croft, C. (2007). A brief history of the Facebook. Retrieved from.
http://www.meerutcollege.org/mcm_admin/upload/1587223450.pdf
Clement (2020); Global networks ranked by number of users
https://www.statista.com/statistics/272014/global-social-networks-ranked-by-number-ofusers/#:~:text=What%20is%20the%20most%20popular,2.7%20billion%20monthly%20a
ctive%20users.
Daugherty, P. R., Wilson, H. J., & Chowdhury, R. (2019). Using artificial intelligence to
promote diversity. MIT Sloan Management Review, 60(2), 1.
ARTIFICIAL INTELLIGENCE
40
https://search.proquest.com/openview/7b528c41903fcfe0494b67d93f0e4447/1?pqorigsite=gscholar&cbl=26142
Fjermestad, J., & Hiltz, S. R. (1998). An assessment of group support systems experimental
research: Methodology and results. Journal of Management Information Systems, 15(3),
7-149.
https://www.tandfonline.com/doi/abs/10.1080/07421222.1998.11518216
Francis, J. J., Johnston, M., Robertson, C., Glidewell, L., Entwistle, V., Eccles, M. P., &
Grimshaw, J. M. (2010). What is an adequate sample size? Operationalising data
saturation for theory-based interview studies. Psychology and health, 25(10), 1229-1245.
https://www.tandfonline.com/doi/abs/10.1080/08870440903194015
Gribbons, B., & Herman, J. (1996). True and quasi-experimental designs. Practical assessment,
research, and evaluation, 5(1), 14.
https://scholarworks.umass.edu/pare/vol5/iss1/14/
Gong, Z., Zhong, P., & Hu, W. (2019). Diversity in machine learning. IEEE Access, 7, 6432364350.
https://ieeexplore.ieee.org/abstract/document/8717641
Gupta, S., Kapoor, A., Shankar, R., Pati, P., Muthu, S., Hoang, A. and Chandrasekaran, S.,
Realization Tech Inc, (2010). Facilitation of multi-project management using task
hierarchy. U.S. Patent 7,774,742.
Houser, K. A. (2019). Can AI Solve the Diversity Problem in the Tech Industry: Mitigating
Noise and Bias in Employment Decision-Making. Stan. Tech. L. Rev., 22, 290.
ARTIFICIAL INTELLIGENCE
41
https://heinonline.org/HOL/LandingPage?handle=hein.journals/stantlr22÷=9&id=&page=
Huck, S. W. (1991). True experimental design. The Journal of Experimental Education, 59(2),
193-196.
https://www.tandfonline.com/doi/abs/10.1080/00220973.1991.10806613?journalCode=vjxe20
Imai, K. (2008). Variance identification and efficiency analysis in randomized experiments under
the matchedâ€Âpair design. Statistics in medicine, 27(24), 4857-4873.
https://onlinelibrary.wiley.com/doi/abs/10.1002/sim.3337
Kirkpatrick, D. (2011). The Facebook effect: The inside story of the company that is connecting
the world. Simon and Schuster.
https://books.google.co.ke/books?hl=en&lr=&id=PxTvbMVCPEC&oi=fnd&pg=PA1&dq=Facebook+history&ots=DVe5sTT16d&sig=5v6UdYM5
RfuPqDa9pY957vkW6pU&redir_esc=y#v=onepage&q=Facebook%20history&f=false
Khari Johnson (2020): Facebook CTO says hiring matters fir mitigating AI bias but the company
lacks AI diversity stats
https://venturebeat.com/2020/06/15/facebook-cto-says-hiring-matters-for-mitigating-ai-bias-butthe-company-lacks-ai-research-diversity-stats/
Leavy, S. (2018, May). Gender bias in artificial intelligence: The need for diversity and gender
theory in machine learning. In Proceedings of the 1st international workshop on gender
equality in software engineering (pp. 14-16).
https://dl.acm.org/doi/abs/10.1145/3195570.3195580
ARTIFICIAL INTELLIGENCE
42
Lipsey, M. W., & Aiken, L. S. (1990). Design sensitivity: Statistical power for experimental
research (Vol. 19). sage.
https://books.google.co.ke/books?hl=en&lr=&id=NXb5SbaMrXMC&oi=fnd&pg=PA7&dq=exp
erimental+research&ots=Jt4MZHVhzI&sig=UFRswnQDuBLsAxXd4Av_N9w2kHI&red
ir_esc=y#v=onepage&q=experimental%20research&f=false
Hall Mark (2017) Facebook: American Company
https://www.britannica.com/topic/Facebook
Moser, A., & Korstjens, I. (2018). Series: Practical guidance to qualitative research. Part 3:
Sampling, data collection and analysis. European Journal of General Practice, 24(1), 918.
Pearce, D. W. (2016). Cost-benefit analysis. Macmillan International Higher Education.
Pink, S. (2016). Digital ethnography. Innovative methods in media and communication research,
161-165.
https://link.springer.com/book/10.1007%2F978-3-319-40700-5#page=181
Pugh, J. K., Soros, L. B., Szerlip, P. A., & Stanley, K. O. (2015, July). Confronting the challenge
of quality diversity. In Proceedings of the 2015 Annual Conference on Genetic and
Evolutionary Computation (pp. 967-974).
https://dl.acm.org/doi/abs/10.1145/2739480.2754664
Stachowiak, R. (2019). Why Diversity Matters. In Remaining Relevant in Your Tech Career (pp.
41-52). Apress, Berkeley, CA.
ARTIFICIAL INTELLIGENCE
43
https://link.springer.com/chapter/10.1007/978-1-4842-3703-8_4
Stathoulopoulos, K., & Mateos-Garcia, J. C. (2019). Gender Diversity in AI Research. Available
at SSRN 3428240.
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3428240
Stainer James (2019) We must fix diversity problems
https://medium.com/@jstanier/we-must-fix-ais-diversity-problem-6ad5fddc2f8c
Tom Simonite (2019) Facebook Is Finding Problems with Artificial Intelligence Too
https://www.wired.com/story/facebook-finding-problems-artificial-intelligence-too/
Varis, P. (2016). Digital ethnography. The Routledge handbook of language and digital
communication, 55-68.
https://books.google.co.ke/books?hl=en&lr=&id=ofMsCgAAQBAJ&oi=fnd&pg=PA55&dq=dig
ital+ethnography&ots=XPcIy7nA5D&sig=YNaZopqd0qV4Bv6B60WuRaKUIps&redir_
esc=y#v=onepage&q=digital%20ethnography&f=false
Purchase answer to see full
attachment
