AI Blog Series
What are some of the ethical concerns surrounding AI and machine learning?
Introduction
AI and machine learning are two of the most promising
technologies to come along in decades. They're poised to change
how we do everything from finding a romantic partner, to
detecting cancer and predicting earthquakes. But there's also a
dark side: with every new application of AI comes new ethical
concerns that can't be ignored.
Tim Davies
5 min read
Ethics cannot be understood by today's Artificial Intelligence
Artificial Intelligence is not a replacement for ethics. AI
can only be used to make decisions within a specific set of
rules, and these rules must be programmed into the system.
AI cannot think outside the box or understand ethics or
morality. For example, if you have an ethical dilemma that
requires a decision between saving one life versus saving
many lives (such as in medical treatment), then it would be
impossible for an AI program to choose between those two
options because it was not designed with this information in
mind.
AI has to be trained using data, and that data is biased by
its creators.
As you may know, artificial intelligence (AI) is trained on
data. In order to build an AI that can recognize images or
understand human speech, you need massive amounts of
training data—preferably real-world examples of the thing
your program intends to do.
Unfortunately, the world we live in is far from unbiased or objective. Racism and sexism are still extremely prevalent throughout society and have a huge impact on how we think and behave. These biases are reflected in written language (think gender pronouns or racial stereotypes), visual representations (like photos of white men as leaders), and behavioural patterns (like hiring decisions). If all this bias were removed from our collective consciousness today, it would be impossible for an AI system to understand how humans work without first being exposed to years' worth of "clean" data about human behaviour.
As an example, consider the following scenario: You’re a developer working on an AI that helps doctors diagnose diseases. You want to train your system by feeding it large amounts of medical data, but you find that most of the data from a specific region is biased toward older white women because they make up most of the population and therefore are more likely to be sick. If you don't have enough "clean" data available, how will your AI know how to properly recognize illnesses?
Unfortunately, the world we live in is far from unbiased or objective. Racism and sexism are still extremely prevalent throughout society and have a huge impact on how we think and behave. These biases are reflected in written language (think gender pronouns or racial stereotypes), visual representations (like photos of white men as leaders), and behavioural patterns (like hiring decisions). If all this bias were removed from our collective consciousness today, it would be impossible for an AI system to understand how humans work without first being exposed to years' worth of "clean" data about human behaviour.
As an example, consider the following scenario: You’re a developer working on an AI that helps doctors diagnose diseases. You want to train your system by feeding it large amounts of medical data, but you find that most of the data from a specific region is biased toward older white women because they make up most of the population and therefore are more likely to be sick. If you don't have enough "clean" data available, how will your AI know how to properly recognize illnesses?
Machine learning systems are not General Artificial
Intelligence — they gain knowledge as they process more data.
● Machine learning systems are not
General Artificial Intelligence — they gain knowledge as
they process more data.
● They don't possess a sense of empathy or a moral compass, and they have no understanding of what it means to be human.
● ML systems are also unable to learn new concepts that aren't already embedded in their training data. For example, an AI program can recognize faces but not understand what makes a face beautiful or ugly; it will simply identify faces based on its programming.
● They don't possess a sense of empathy or a moral compass, and they have no understanding of what it means to be human.
● ML systems are also unable to learn new concepts that aren't already embedded in their training data. For example, an AI program can recognize faces but not understand what makes a face beautiful or ugly; it will simply identify faces based on its programming.
The consequences of a machine learning error can influence
society on a large scale.
When you think of ethical concerns in the world of machine
learning and artificial intelligence, what comes to mind?
The first thing that might come to mind is some kind of AI
gone rogue, or maybe a robot uprising. That's certainly an
interesting plot for a film—but what about areas where
ethics play a role in the real world?
One example is the potential for mistakes made during the training phase of machine learning algorithms to negatively impact society on a large scale. Machine learning can be used for all kinds of things, from detecting diseases to processing loan applications at banks. For example: imagine if your bank's financial records were accidentally destroyed by an algorithm that was supposed to help them process payments faster but didn't work as intended due to poor design and testing practices (a scenario which has actually happened). What would happen then? How would this affect your trust in the system? These are just some examples of how mistakes made during training can lead up having far-reaching consequences throughout society.
One example is the potential for mistakes made during the training phase of machine learning algorithms to negatively impact society on a large scale. Machine learning can be used for all kinds of things, from detecting diseases to processing loan applications at banks. For example: imagine if your bank's financial records were accidentally destroyed by an algorithm that was supposed to help them process payments faster but didn't work as intended due to poor design and testing practices (a scenario which has actually happened). What would happen then? How would this affect your trust in the system? These are just some examples of how mistakes made during training can lead up having far-reaching consequences throughout society.
Bad actors can use AI capabilities to manipulate and persuade
individuals and populations
There are a number of ethical concerns surrounding AI and
machine learning that companies, governments and individuals
should consider in order to mitigate potential negative
effects. One such concern is that bad actors can use AI
capabilities to manipulate and persuade individuals and
populations. For example, an individual could be targeted
with an ad tailored specifically to their interests or
demographic information; similarly, a political party could
target an entire nation with ads aimed at influencing the
outcome of elections or public opinion on important issues
such as healthcare reform or climate change mitigation
efforts.
Some researchers have suggested it may be possible for these types of techniques to be used in non-political contexts as well—such as advertising products or services on social media platforms like Facebook, Twitter or TikTok—which raises additional questions about how this type of technology might impact privacy rights around personal data collection.
Some researchers have suggested it may be possible for these types of techniques to be used in non-political contexts as well—such as advertising products or services on social media platforms like Facebook, Twitter or TikTok—which raises additional questions about how this type of technology might impact privacy rights around personal data collection.
It’s difficult to determine whether or not what appears to be
fair is actually fair.
The first issue that arises when we consider the ethics of
AI is that it’s difficult to determine whether or not what
appears to be fair is actually fair.
It may be easy enough for us to recognize certain acts as morally wrong—for example, theft or murder. But deciding how much money someone should be paid based on their gender or race isn't as straightforward, because these are complex issues with many variables at play. And even if you could create an algorithm to determine this for you, it would still only have a limited understanding of social dynamics and cultural context which might require further analysis from humans before making a decision about fairness. For example, maybe one group has more experience in a particular field than another due to historical factors that are hard for machines alone to understand; this could result in bias towards one group over another even though there was no intent behind it whatsoever (which raises another question: does intent matter?). As such we cannot know if decisions made by machines will always lead them down an ethical path without having human oversight first - but given how quickly AI is evolving already makes this prospect unlikely until much later down the line at best!
It may be easy enough for us to recognize certain acts as morally wrong—for example, theft or murder. But deciding how much money someone should be paid based on their gender or race isn't as straightforward, because these are complex issues with many variables at play. And even if you could create an algorithm to determine this for you, it would still only have a limited understanding of social dynamics and cultural context which might require further analysis from humans before making a decision about fairness. For example, maybe one group has more experience in a particular field than another due to historical factors that are hard for machines alone to understand; this could result in bias towards one group over another even though there was no intent behind it whatsoever (which raises another question: does intent matter?). As such we cannot know if decisions made by machines will always lead them down an ethical path without having human oversight first - but given how quickly AI is evolving already makes this prospect unlikely until much later down the line at best!
The creators of AI do not always understand how a decision or
recommendation is made by the AI
One of the biggest ethical concerns surrounding AI and
machine learning is that a lot of people don't understand
how AI makes decisions or recommendations. In fact, it's
even potentially possible that you could be using an AI
system without knowing it.
A lot of times when we see AI in action, we just accept whatever output it gives us. We may not even realize there was a decision-making process that went on behind the scenes to get there. But this lack of transparency can be really problematic because:
● It's impossible to know if and how your personal data has been used by an algorithm without access to its source code (which isn't always easy).
● If an automated decision-making process isn't transparent, then there's no way for someone who disagrees with their results from exercising due process rights under law like filing lawsuits or seeking administrative review by regulators if necessary—and those rights are important if you feel wronged by an injustice caused by faulty AIs.
A lot of times when we see AI in action, we just accept whatever output it gives us. We may not even realize there was a decision-making process that went on behind the scenes to get there. But this lack of transparency can be really problematic because:
● It's impossible to know if and how your personal data has been used by an algorithm without access to its source code (which isn't always easy).
● If an automated decision-making process isn't transparent, then there's no way for someone who disagrees with their results from exercising due process rights under law like filing lawsuits or seeking administrative review by regulators if necessary—and those rights are important if you feel wronged by an injustice caused by faulty AIs.
Companies with the most access to data will have the most
power.
Data is the new oil. In other words, data is valuable and
companies with the most access to it will have the most power.
It's not just about collecting more data, but being able to
analyze and interpret it; this is where companies can gain a
competitive advantage. If you're looking for examples of this
in practice, look no further than Google and Facebook: both
are infamous for having access to massive amounts of
information about their users that they use to make their
products better (for example by using your search history to
suggest relevant ads).
Legislation is coming, but it's too early to know whether it
can force ethical logic into AI recommendations and decision
making
The need for legislation is clear. As AI grows in popularity
and power, so too will its potential for harm. However, it’s
unclear whether current laws can keep up with the pace at
which AI technology is developing. The industry itself needs
to take responsibility for ensuring that ethics are built into
the development process and applied in real-world scenarios,
but until then we can expect some bumps along the way as
legislators try to keep pace with this rapidly changing
landscape.
Conclusion
As you can see, there are many ethical concerns surrounding
artificial intelligence and machine learning. Technology has a
long history of being used for good and bad, but the stakes
are higher now than ever before. If we don’t take these issues
seriously now, we risk having to deal with the consequences
later on down the road — and we don’t want that! So please do
your part by educating yourself about what is going on in this
space so that when legislation does come into effect you know
how best to support ethical AI development.
Tim Davies