Process Book for The New Workforce

Background and Motivation:

At the turn of the decade of the 2020s, the job market in the United States entered an era of volatility, starting with a historic layoff during the height of the Covid-19 Pandemic that the labor market has yet to recover from. (1) With the introduction and proliferation of AI brought even more anxiety and uncertainty among Americans relating to job security. (2) With the advancement of AI, many have felt threatened by possibly being laid off due to their occupation being able to be accomplished by more efficient AI, which, as it grows closer to and may exceed human performance, may appear more desirable than human workers to employers. With our project, we will analyze the rising interest in AI through search engine usage versus the numbers of layoffs in cities worldwide. Then we will see just how much AI popularity correlates to bigger layoffs, or if they correlate at all.

Project Objectives

Our objectives are:

• Show the popularity of AI in search and use since at least 2019
• Analyze trends in layoffs from 2020 to present in various geographic locations.
• Map trends in both datasets, comparing and contrasting the trends of the popularity of AI and job security in order to discern whether a tangible relation exists between the two.

Data:

For our search data, google provides its own analysis of search trends, which users can easily specify and obtain specific data relating to the search term over a desired time range. Google gives the user the ability to download large quantities of data relating to searches and can be filtered by region and date for further specificity.

For information about job layoffs, we turned to the machine learning and data science dataset website, Kaggle. We found publicly available dataset that provided ample information about job layoffs from 2020 to 2024, with features including the specific number of workers laid off, the region (by city), and even the relevant industry of the layoffs.

Data Processing:

Both the datasets are in .csv format, so processing covnerting them into json files where each row corresponds to a sub-json object shouldn't cause any compilations. However, one issue is that while Google's dataset is unwaveringly consistent, the Kaggle dataset requires a significant amount of filtering, as occasionally required columns such as "number of workers laid off" in the csv are not filled in some rows, resulting in rows that are unusable for analysis and presentation, requiring us to reduce the size of the dataset. Once the data has been converted into json objects, we can use JavaScript's d3 library to present the data in visualizations. For more diversity and insight, Google's dataset will come from data for multiple AI-related search terms, such as "machine learning" or "chatbot". Google's datasets will be combined to create an analysis of the overall interest tred in artificial intelligence.

Visualization Drafts:

Three sketch designs, where we played around with line graphs vs. bar graphs vs. icon arrays.

Final Design Draft

For the final design, we decided besides the obvious map and city selection, to display the results for each city with icon arrays. The layoffs amount will be represented by people icons, while AI popularity is shown per robot icon. After the user selects a city from a circle on the map, on the bottom will come up the results for that selected city. The user can select a month/year range from the selection menu above, then the results return two icon arrays— the layoffs and AI search popularity. There will also be a use of color for the laid-off people to show what field their job was in (tech, retail, education, etc.).

The map and dots on it will allow users to drag and easily select a city of interest. If the user has a specific city in mind, they can also type it in the search bar on the right, and find it on the list below it. We chose the icon arrays since we though the visual interest of person vs. robot would be more interesting for our topic. It also allows to show different values per icon instead of dealing with one scale on the y-axis like with bar charts or line graphs. For example, one person icon can mean per 100 people, while 1 robot icon can mean per 300 searches with "AI". This can be more easily differentiated with icon arrays and their lengends/keys, while one axis can't handle that difference in scale.

Milestone #1

Milestone #2