Using AI for a simple HR task was disappointing!
I wanted to research how using AI might be used in AI. I took a common task that payroll or HR might need to do throughout the year. Each quarter the minimum wage can change. Now you would think the federal minimum wage is fairly straight forward. However, there are many overrides to this rate. Each quarter you have research min wage changes.
Minimum wages can exist at all levels which may override the federal wage. For example, states, cities, and even counties may have their own minimum wage. Typically you have to review official state, county, or city pages to find some of these rates. There are some federal and HR sites that do a good job of trying to combine all the data into one table. Still, there can be errors.
For this project, I asked both ChatGPT 4.0 and Gemini Advanced to do the heavy lifting of finding these rates by different legal entities. I expected it to look at legal sites (Dept of Labor, State, Cities, Counties) to find the latest and greatest minimum rates and the rules that apply. A rule, for example, may say the min rate for the city or county is higher if you are considered “a large” employer (many employees) vs. “a smaller employer” (fewer employees).
The input fed into AI looked similar to the following:I am an HR Generalist and want to know the min wages for each state if the federal wage is overridden. Using the state websites or valid inputs list the state minimum wage for 4th Quarter 2024. If counties or cities min wage override the state’s minimum wage, list them too. The out put should be in table format where the states are in alphabetical order going from top to bottom and any counties or cities are indented below the state.
The output from ChatGPT4.0 looked like the following:
It’s fairly clean except there are multiple errors in the wages. Additionally, it did not include the county override like I had suggested. I asked again and it included the counties. Why do you have to ask over and repeat yourself in AI?
It gave the wrong rates for several states and cities. For example (not shown above), but I had to tell it the following:Why does Chicago show the wrong numbers, AI from google says Chicago is $16.20 for 4 or more employees? Please review again ?
ChatGPT said:
You’re right; I apologize for the incorrect information regarding Chicago’s minimum wage. Let’s correct that and provide an updated table with the accurate minimum wage rates for Chicago and other areas that have local overrides.
I eventually got a list of what I wanted but there were multiple minimum wage rates across all legalities. Hoping that Gemini Advanced AI might be a little smarter, I repeated the exercise using Gemini Advanced.
The conclusion, it too failed to bring in accurate data from public government websites. It probably had a 20% failure rate which is not acceptable when you are dealing with people’s pay. If you have to go back and check the accurateness of the AI engines, then what’s the point of using them? You don’t really save any time do you ?
I found the AI engines accuracy troublesome for what should be a simple task given the inputs are well known public sites for each entity. AI, you aren’t working for me (for now!). Let’s see the next version.
Note: Image for blog header generated by ChatGPT4 / DALL·E
