BIAS & TOXICITY PART 2

What we should do about them, how we should handle bias and toxicity, or adapt models to cope with these issues? 

In the fast-evolving world of artificial intelligence (AI), addressing bias and toxicity has emerged as a critical concern. These issues can perpetuate inequalities, reinforce stereotypes, and hinder progress toward a fair and just society. Mitigating bias and toxicity in AI models requires a multifaceted approach that encompasses data collection, transparency, collaboration, user engagement, and more. Here is how we can foster a safer and more ethical AI landscape:

  • Ensuring that the data used to train AI models is diverse and representative of various demographics, cultures, and perspectives can significantly mitigate bias. By providing a more accurate reflection of the real world, AI models can offer fairer outcomes.
  • Encouraging transparency in AI development processes is essential. This includes clearly stating the limitations, biases, and potential risks associated with the model. Adhering to ethical guidelines and principles is paramount to building trust in AI systems.
  • Implementing ongoing monitoring and evaluation of AI models in real-world scenarios is crucial to detect and rectify biases and toxic behavior. Regular audits and assessments should be conducted to ensure the model’s performance aligns with desired ethical standards.
  • Fostering collaboration among researchers, practitioners, policymakers, ethicists, and the public is vital to jointly address bias and toxicity in AI. Diverse perspectives and expertise are essential for developing comprehensive solutions.
  • Researching and implementing algorithms that specifically target bias reduction and fairness in AI models is critical. Techniques such as adversarial training, re-sampling, and fairness constraints can be employed to reduce disparities and ensure equitable outcomes.
  • Raising public awareness about the existence of bias and toxicity in AI models is fundamental. Educating users, developers, and decision-makers about the implications and potential harm caused by biased models is crucial to promoting responsible AI use.

By collectively embracing these steps and committing to ongoing improvements, we can steer AI towards a more ethical and inclusive future. Addressing bias and toxicity is not only a technological challenge but a societal imperative that demands our unwavering dedication and cooperation. Together, we can build a better, more equitable AI ecosystem.

BIAS & TOXICITY PART 1

What do you think?

Large Language Models are data sponges, soaking up vast amounts of text data from the internet. While this might seem like a great way to make them smarter, it comes with a troubling side effect. LLMs inadvertently inherit and propagate biases present in the sources they are trained on. In other words, the prejudices, stereotypes, and discrimination found in society’s digital niches and cracks get magnified and perpetuated by these models. The consequences of this bias amplification are far-reaching. It reinforces existing stereotypes, discriminates against marginalized groups, and perpetuates harmful practices. The ramifications of these biases ripple through every facet of our digitally intertwined lives.

The toxicity exhibited by Large Language Models is equally alarming. These models can generate harmful or offensive content with alarming ease. This toxic output, whether it be in the form of cyberbullying, hate speech, or misinformation, has real-world consequences. It doesn’t just stay on the screen; it seeps into our lives and affects individuals, communities, and societies. The widespread dissemination of such harmful content can create an environment that feels hostile and unwelcoming to many. It not only harms individuals but also erodes trust in the technology and the platforms that employ these models. People lose faith in digital spaces when they become breeding grounds for hate and misinformation.

As Large Language Models become increasingly integrated into various aspects of our lives, addressing their bias and toxicity becomes not just a choice but an imperative. To use these powerful tools responsibly and ethically, we must tackle these issues head-on. Ensuring that LLMs are free from bias and toxicity is the first step towards creating a digital space that is fair and inclusive for all users. It’s about reclaiming technology to serve humanity, rather than perpetuating its shortcomings. It’s about fostering an environment where diversity and equity are celebrated, not suppressed.

In conclusion, the bias and toxicity inherent in Large Language Models are not mere side issues. They are significant challenges that demand our attention and action. Recognizing the potential for harm that these biases and toxic outputs pose is the first step toward making a change. We must hold technology to a higher standard, one that promotes inclusivity, equality, and the betterment of society. Only then can we fully embrace the power of Large Language Models and their potential to enrich our digital world.

My phone Screen Time

Do I use my phone responsibly or addicted?

Research Question and Audience 

Ever since the smartphones have been introduced to us, they became an indispensable tool as they occupy a huge space in our daily life. Whether we want to connect with friends or strangers across the globe, to use them as a medium of entertainment or merely as a device of communication among other things, we rely heavily on them. It’s from this perspective that I decided to analyze my phone data to see if I use my phone responsibly or I’m deemed a phone addicted based on a predefined benchmark. This study can also serve to draw a line on what is considered phone addiction to alert smartphone users.   

What is a healthy amount of screen time for adults? A study conducted by RescueTime and eMarketer shown that people on average spend around 3 hours 15 minutes and 3 hours and 43 minutes respectively on their mobile devices. The top 20% of smartphone users spend more than 4.5 hours on their phones during weekdays (https://elitecontentmarketer.com/screen-time-statistics/).

I looked at my phone data for the month of October and realized that I spend on average 3 hours and 10 minutes per day on my phone. Apparently, I spend a reasonable amount of time on my phone based on those studies. But does it happen during the weekday or weekend or when I’m home, school or work? This inquisitiveness led me to this research question: “In which capacity, day, and the app category I’m spending the most time on my phone?

I wanted to dive even deeper to see if there is a relationship between the time spent on my phone and the other variables from the dataset, variables such as “Weather”, prompting to this research sub-question: “Can we also establish a relationship between the time spent on my phone and the other variables?”

Data Collection and Visualization

I collected 4-weeks of my phone screen time data to gauge the total time spent on each app and the day of week that happened the most. The most used apps from my phone were grouped into five (5) categories from my phone:

  • Creativity
  • Entertainment
  • Productivity and Finance
  • Social
  • Other

The variables included in my visualization are the following:

  • Time Spent (Time spent by apps category)
  • Apps Category (The five categories mentioned above)
  • Day (Weekday and weekend)
  • Capacity (where do I use my phone the most? At school, work, or home?)
  • Weather

The use of bar and line charts for the visualization was the best choice to represent the data and address the question. The bar charts showcase what apps category (discrete value) I spent the most time on in combination of where or in what capacity (discrete value) and which day (discrete value) of the week that occurs. The line charts on the other hand, intended to draw a connection between the weather and the time spent on my phone.

From the visualization, we can infer that there is no relationship between the weather and time spent on my phone. It’s fascinating how I was able to capture the answer to my research questions with just one click.

https://public.tableau.com/app/profile/prudence.brou3156/viz/PROJECT2_16675416983310/Dashboard1?publish=yes

Next Steps

I think, one of the shortcomings of this research is the lack of the computer screen time in the dataset. It would have been interesting to compare phone vs. computer screen time to gauge the overall time spent per day. In other words, the ability to broaden this study to include computer screen time data could take this project to the next level.        

Residential Noise Complaints in NYC

What is the most observed noise complaint across the 5 boroughs?

My visualization aims to inform the public about the number and type of residential noise complaints received in NYC from September 2021 – September 2022. The most observed noise complaint across the 5 boroughs was the loud music/party. One could ask the following question: “at what day and time of the week the loud music occurred”? The visualization didn’t include the time and day the incident was reported for further analyses.

  1. The bar chart showcases the type and most observed residential noise complaints across the 5 boroughs.
  2. The stacked chart gives to the audience more details regarding the status of the complaints, whether the city agency who handled the matter resolved it or not.
  3. The map pinpoints the location or street address where the incident occurred

https://public.tableau.com/app/profile/prudence.brou3156/viz/PROJECT1_UPDATE/Dashboard1?publish=yes