Technology advancements throughout history

Is job replacement something that people should really be worried about when it comes to AI? While it has the potential to change the nature of our work, it’s unlikely to completely replace human workers. The role of AI in our society is the same as any other big invention we’ve had in the past — to make us evolve and create new opportunities.

If we look back at the 18th century, something similar happened during the Industrial Revolution when automatic looms were invented to weave fabric without the need for manual labor. At the time, these looms were also seen as a potential threat to traditional hand weavers and word spread that this new technology would replace human workers entirely, but time proved otherwise!

While it brought significant changes to the textile industry, it also lead to the creation of new jobs related to machine maintenance, operation and design. Skilled workers were still needed to oversee the machines, perform maintenance and repairs, and adapt the technology to changing demands.

There is a pattern when it comes to technology advancements throughout history — they tend to reshape industries and job roles, evolving the workforce instead of causing complete elimination. That’s why we need to be adaptable, open to change and willing to step out of our comfort zone to walk alongside progress instead of staying behind.

Real problems AI brings that we must worry about

Okay, now that we know AI is unlikely to replace us all, this brings us to the core topic of this article — the real risks and problems that come with the boom of AI-powered tools.

One of the biggest risks is the perpetuation of biases and discrimination, which we have been working hard as a society to fight against. Minority groups are already suffering with the consequences in tools that are meant to be smarter than us. Let’s look at some examples!

Image generating AI tools

These types of tools have become a huge trend since they started to appear, and often reflect societal prejudices and stereorypes. If you explore enough of them, you will quickly find a pattern:

• Lighter skin tones appear more frequently than darker skin tones
• Women are more likely to appear in domestic settings than men
• Men are more likely to appear in professional roles than women
• Certain body types or appearances are favoured over others
• Disabled people are not accurately represented

A good illustration of that can be seen in the images below. When a new feature to create AI images of people in different scenarios of an app called Remini became famous among Instagram users, Cacai Bauer, a brazilian influencer with Down syndrome, decided to join in the fun and try it out herself. As a result, she got unrealistic images that made her thinner and washed away some of the features of her face.

Amazon’s recruiting tool

In 2018, it became public that Amazon had tried to develop a recruiting tool that used artificial intelligence to assist in the hiring process by automatically reviewing job applications and ranking candidates, but they had to discontinue it due to the identification of gender bias.

The tool was meant to review resumes and applications, assigning scores to candidates based on certain keywords and patterns found. This way they would be able to automate the initial screening and save time for recruiters.

However, during the testing phase of the tool, it was discovered that it was discriminating against female candidates. This discrimination was a result of the data used to train the machine learning model, more specifically the historical hiring patterns it learned, where the majority of the company’s past hires were male, making the AI downgrade resumes containing terms that were more commonly associated with women.

When the issue was spotted, there was an attempt to modify the algorithm and eliminate the bias, but they were not able to succeed and had to ultimately discontinue the tool.

Root causes

There are several reasons why discrimination is such a big problem when it comes to AI, let’s explore a few of them:

• Biased Training Data — If the training data for the machine learning model contains biased or discriminatory patterns, the AI will learn and reproduce them. Historical data is a big source of societal prejudices, as it plays an important role in building an AI’s understanding of the world.
• Underrepresentation — If certain groups are underrepresented in the training data, the AI may not have sufficient exposure to them, which can lead to misinterpretations and reinforce existing biases.
• Algorithmic Bias — The algorithms used in AI models might inherently have biases, even if the training data does not. For example, certain algorithms may be more likely to generate discriminatory outputs based on their design.
• Lack of Contextual Understanding — AI models may lack a deep understanding of cultural, historical or contextual factors that influence language and human interactions, leading to misunderstandings and biased responses.
• Human-AI Interaction — Users may unknowingly guide AI towards discriminatory responses through their queries or prompts.

Potential solutions

When it comes to addressing and mitigating bias in AI, complete elimination is still a challenge. However, there are several steps that can be taken to reduce the risk of introducing it and promote more equitable outcomes, here are a few of them:

• Diverse and Representative Data — Make sure your training data includes a wide range of perspectives, demographics, and sources.
• Data Preprocessing — Preprocess and clean the data to remove any unintentional biases that may exist in the dataset. This can involve techniques like data augmentation, debiasing algorithms, and balancing data representation.
• Bias Assessment and Monitoring — Regularly assess and monitor the model. Use metrics and techniques to quantify and identify bias in model predictions across different groups.
• Human Reviewers and Guidelines — When fine-tuning models, provide clear guidelines to human reviewers to avoid favoring any specific group, promoting hate speech or taking positions on controversial topics.
• Diverse Team and Stakeholder Involvement — Ensure that your AI development team includes diverse perspectives, including domain experts and individuals from different backgrounds.
• Transparency Reports — Publish transparency reports that detail the efforts you’re making to address the problem, the methods you’re using and the outcomes you’re achieving.

Is AI a friend or a foe?

As highlighted in this article, AI has been bringing back biases. However, while it can be seen as a threat, it is also a tool that we can use to our advantage. Many of the bullet points explored here were actually generated from conversations with ChatGPT, which is a clear indication of the potential benefits of AI.

Also, most of the data used to train machine learning models comes from humans and our history. This raises an important question, who is the real enemy? Is it the technology that is being developed, or is it the humans who are providing the data? Perhaps, it is a combination of both.

Therefore, it is important to approach the development and use of artificial intelligences with caution, taking everything we discussed in this article into consideration. With the right approach and understanding, AI can be a powerful tool to help us tackle some of the biggest challenges we face as a society in modern times.