Targeting AI Ethics: Data, Models, and People

Updated: Mar 23



Artificial Intelligence is a powerful technology that is shaping our world. Some AI

systems can deliver harmful, discriminatory outcomes that have further exacerbated

societal inequities. For example, some AI enabled systems make decisions about who

is approved for a loan, selected for a job interview or whether the police show up in

certain neighbourhoods. Institutions rely on these systems, but the technology is far

from neutral or objective. Systemic inequities are encoded into technology, further

amplifying unjust outcomes. The problem gets worse as our dependence on technology

increases.


But, how exactly, do ethical issues in AI arise? One way of thinking about this is to

break things down into three areas – data, models and people.


The Devil is in the Data

Data is key ingredient for AI systems. Some call data “the new oil” because without a lot

of data, AI systems that use machine learning would not be all that functional. Decisions

around gathering data have power structures baked into them. For example, if you

design a survey, you get to control how many questions are asked, what questions are asked and if there are drop down answers or free text boxes. You will also decide how

to reach people to complete your survey and you may choose to target a certain group

of people. In essence, you determine the purpose and method of data collection and

your choices become encoded in the process.


Historically, we may have more data about some people or things than others because

of these decision-making processes. That leads to gaps in available data. For example,

we’ve historically collected more medical research data for men than women. This

means that we often don’t have adequate medical datasets for women. If that historical

data is then used to power an AI system, it may generate biased outcomes. This is

known as algorithmic bias.


In our highly digital world, data is being collected all the time. Every online search, every

click, every keyboard stroke, we are contributing to a vast web of data collection. There

is also meta-data, or data about data, such as geo-location data for your phone. This

large volume of constant data collection raises many concerns about privacy and

consent, which also contributes to ethical issues in AI.


Mathematical models are not neutral.

Machine learning models can contain bias. Cathy O’Neil, a famous data scientist and

author of the book “Weapons of Math Destruction” says that “models are opinions

embedded in mathematics.” AI developers make many decisions during the design

process, such as deciding what techniques to apply to develop a model, engineering the

features that will be contained in a model, and determining the hyper parameters, the

over arching guidelines, for training a model. Each decision represents a value

judgment on the part of the person making it. For example, using deep neutral networks

in order improve technical accuracy may reduce how explainable the model is to

people. That trade off is a value judgement which prioritizes accuracy over

explainability. There are many decisions like this being made as models are

constructed.


People who make a technology shape a technology.

There is a body of work that demonstrates that the people who make a technology

shape that technology. We see this in game design. Female characters in a video game

are often scantily clad and sexualized, reflecting the fantasies of their mostly young,

male creators. Most AI developers are white or Asian men from a certain socio-

economic background. Their worldview and values inform the technology. In addition,

larger system, such as how funding is allocated, can also set the agenda for the types

of technologies that are developed. This was evident in the early days of AI, as much of

the funding came from the military during the 1950s and 1960s. Today, large players

like Google, Amazon, Facebook and Apple are driving the agenda.


This is not an exhaustive list, just a quick overview of how data, models and people

contribute to the ethical issues we see in AI systems.


To learn more, check out the PowerEd AI Ethics Micro-Credential at Athabasca University.


By Katrina Ingram, CEO, Ethically Aligned AI _______


Sign up for our newsletter to have new blog posts and other updates delivered to you each month!

Ethically Aligned AI is a social enterprise aimed at helping organizations make better choices about designing and deploying technology. Find out more at ethicallyalignedai.com © 2022 Ethically Aligned AI Inc. All right reserved.