Can you think of a politician or a person of influence who is driven only by a goal? Do they make you feel uncomfortable, unsafe, or even angry? Why?
Unquestionably, these people can be ruthless in achieving their goals and are indifferent to the consequences of their actions. As a result people unjustly get hurt along the way.
These people and Machine Learning (ML) are very similar in the sense that they do not care about how they achieve their objectives unless there are measures put into place to monitor them.
However, ML has many advantages over people in achieving its objectives. Billions of dollars are spent on ML systems and every large company has a team of highly skilled individuals working on improving the ability of ML to achieve its objective.
First Get To Know How ML Works By Covering Some Terms
😎 Artificial intelligence (AI) — The trendsetting CEO
AI is like the popular CEO celebrity that everyone always talks about. It encompasses and represents the long term mission of the field: to build machines that emulate and then exceed the full range of human cognition.
Everyone wants to stay up to date with this trend-setting CEO and are constantly amazed at its intelligence and ability to improve.
🧠 Machine learning (ML) — The genius that sometimes lacks communicational skills
ML is like a crazy intelligent person that sometimes lacks to communicate how exactly things are done. It is a key part of how the CEO can be so clever.
It loves to use mathematics and statistics to give machines the ability to “learn” from data without being explicitly given the instructions for how to do so. However, it can not tell you exactly how and why the machines get smarter.
📈 Models — The business analyst
Just like business analysts, models become more accurate when they are given large volumes of high-quality data. Then, they can be used to make predictions.
Just like the Business Analyst would be trained to use data to make predictions, the ML algorithm trains a model with maths and stats. Then the model progressively improves performance on a specific task.
🎲 Reinforcement learning (RL) — The calculated experimenter
The calculated experimenter loves to learn by doing and trying new things, but they do not want to get fired. So they learn by trial and error; however, they make sure to listen to the policies and the responses from their boss.
If the boss rewards the calculated experimenter, they do the same or similar actions. However, if the calculated experimenter does something that the boss does not like they avoid doing it again.
🐈 Deep learning (DL) — The copycat
DL attempts to copy the activity in layers of neurons in the brain to learn how to recognize complex patterns in data. However, this copycat is like an XXL one as it uses a large number of layers of neurons. Then it makes models that help it to learn rich representations of data to achieve the performance it wants.
📣 Natural Language Processing (NLP) — The interpreter
An interpreter that translates between two languages so that both parties can understand. Similarly, NLP translates between the way humans talk so that the computer can understand interpret and manipulate human language.
In essence, NLP works by breaking down human language into shorter, elemental pieces, to understand relationships between the pieces and explore how they work together to create meaning.
Unethical ML Models Are A Problem
Whew — now that we got the jargon out of the way, let’s focus on the large problems that bad applications of ML cause and what it means for humanity.
The Information Apocalypse
“We’re entering an era in which our enemies can make anyone say anything at any point in time” — Jordan Peele
Deepfakes were created in 2017 and allow one to manipulate videos or other digital representations produced by sophisticated artificial intelligence. They make manufactured images and sounds that appear to be real.
Manipulating videos and photos through artificial intelligence could potentially be used to spread misinformation or damage one’s reputation. Furthermore, as technology progresses it will be hard for the public to understand what is real and what was created for propaganda.
Deepfakes have the potential to cause social unrest and division as everyone will be able to conduct “research” and prove whatever option they hold to be correct.
Facial Recognition & ML In Decision Making
Facial recognition is very common around the world as 50% of the world currently allows the use of it. And only 3 countries have partial bans on the technology!
The main application of facial recognition for the “benefit” of humanity is allowing police to find criminals.
One of the most notable companies in facial recognition is Clearview which scraped more than three billion images of people from platforms like Facebook, YouTube, Venmo, and millions of other websites. Then Clearview licensed their search engine for faces to over 600 law enforcement agencies.
Federal and state law enforcement offices said that Clearview had been used to help solve a variety of cases. Then a follow-up investigation from Buzzfeed revealed that Clearview’s technology had also been used by private individuals, banks, schools, the US Department of Justice, and retailers such as Best Buy.
Although this tool may sound like a superpower, it can cause wrongful arrests. For example, in May 2019, Detroit police arrested Michael Oliver after a facial recognition algorithm incorrectly matched him with a cellphone video. Then, in January 2020, Detroit police arrested Robert Williams after a similar algorithm incorrectly worked.
These are reported cases of the model not working; however, we don’t know exactly how many times the algorithms made mistakes. Furthermore, people may blindly trust Clearview’s technology as it is often assumed that technology, math, and science are always correct.
GPT-3 Making Biased Predictions
However, the GPT-3 model is trained on large volumes of language from the internet and thus, reflects the bias in those datasets. There is no getting around that AI needs data to work, and one of the fastest ways to get data is to look at papers from the past.
However, advancements in NLP that allow models to be trained are likely causing models to be more biased as everything a human writes is biased.
Researchers looked at the probability of the model following a profession with male or female, indicating words and found that professions demonstrating higher levels of education were heavily male leaning. As for the race, different races have different sentiments on a scale of 7.
The danger of GPT-3 being biased is that many companies use it as a base to build software that will together impact billions. For example, a company can use GPT-3 to respond to a text that a person types into the computer with a new piece of text that is appropriate to the context.
However, since GPT-3 is biased, its reply can favor some people instead of others. Now imagine if this was used as a test for the first round of hiring at a company — can you see why it could be an issue?
Reason For Unethical Models
Using Data That Is Biased
Our world is not fair. There’s no balanced representation of the world and so data will always have a lot of some categories and relatively little of others” — Olga Russakovsky
Machine learning models are only as good, or as bad, as the data fed into them during training.
And in the case of GPT-3, that data is massive. GPT-3 was trained on the Common Crawl dataset, a broad scrape of the 60 million domains on the internet along with a large subset of the sites to which they link. This means that GPT-3 ingested many of the internet’s more reputable outlets along with the less reputable ones.
AI Research Is Secretive
Another reason contributing to ML models being unethical is the fact that only 15% of AI papers publish their code. To understand why that is a problem imagine giving a 10-year-old homework and saying that they can tell the teacher if they have completed it and checked over their answers.
Without regulation, the 10-year-old would likely not do much of the homework, and won’t care if it is done correctly. Just like the 10-year-olds, without regulation, large corporations won’t be accountable by the public to make sure that they are eliminating bias from the models they create.
So, How Can You Help Keep ML Ethical?
One major way is to take steps to improve the decentralization of AI and to be educated upon what it is. SingularityNET is one big company that allows anyone to create, share, and monetize AI services at scale.
Another major player in the decentralization of AI is Ocean. It allows data providers to monetize data while preserving privacy and control while also allowing data consumers to access data that is needed to train models. Access to data is key as data samples are expensive to generate, which likely presents an advantage to large companies entering new fields with supervised learning-based models.
However, large companies don’t publish all of their code as is usually intertwined with proprietary scaling infrastructure that cannot be released.
- Bad applications of ML are a humanitarian crisis as they can cause an information apocalypse, cloud judgment, and make biased predictions.
- The reasons for unethical models is because models use data that is biased to get trained, and it is hard to get data that is not biased.
- The fact that AI research is secretive doesn’t force companies to regulate the way their AI works.
- One step everyone can take to improve unethical applications ML is to learn about the decentralization of AI.
So long as people are talking about the issues ML can cause, the world can avoid the issues they can bring to society.
To check out more of my work & to follow my journey of learning about emerging technology, sign up for my newsletter ☀️