Data Carpentry for Biologists

AI - Introduction and Basic Prompting

Notes

Understand the concepts

  • Working effectively with AI requires that you understanding the concepts
  • But it reduces the need for remembering specific syntax
  • E.g., It’s crucial to understand the idea of filtering the data, what it does, and how combining multiple conditions works
  • But it matters less if you quickly remember the how to write %in%

Be able to read the code

  • As scientists we are responsible for making sure our code reflects the analysis we think we are doing
  • The code we produce is our reproducible documenation of what we’ve done
  • We should be able to look at the code to determine if this is true
  • We’ll talk about other ways of checking that it is true later in the semester

Should I use AI for this?

  • Ask the meta question - given my goals how useful is using AI for this task?
    • Do I want to learn this or not?
    • Is the process of implementing part of my thinking (e.g. writing)?
    • Is it critical that this is right/that I understand the details or not?
    • Would I enjoy doing this more myself?
    • Does my brain need a break from executive function?
    • Is it quicker, easier, cheaper to do it directly?

Providing context

  • Models need “context”
  • Details of what you want done, how you want it done, and information needed to do it
  • The most basic form of context is the “prompt”
  • The thing that you ask the model to do
  • Good prompts provide details
  • For programming include language and any specific tools you want to use or not use
  • Since we want to be able to understand the code that is generated tell the model to use the things you’re familiar with

  • R, dplyr

  • Initial prompt (w/o data info):

Using R and dplyr produce code that will take my data table, select the species, date, and weight columns, filter for weights >50 g, and remove any rows with null weights, add a new column showing weight in kg, and sort the data by weight in descending order.

  • Did we cover how this filter statement works?
  • No, we learned drop_na(), why didn’t the model use it?
  • Didn’t specify we are also using tidyr
  • Add that context now (or just include it from the start)

Use drop_na from tidyr instead of !is.na

Use base R pipes

  • Try, unsucessfully, to run the resulting code
  • The model also needs information about the data you are working with
  • Data format as context
  • Often more useful to provide the top few rows of the data file
  • One way to do this is by pasting the first few rows into the prompt
  • Copy the prompt, add tidyr & base R pipes if needed

  • Open surveys.csv in RStudio and copy-past the first few lines

  • Modified prompt (w/data info):

Using R and dplyr and tidyr produce code that will take my data table, select the species, date, and weight columns, filter for weights >50 g, and remove any rows with null weights, add a new column showing weight in kg, and sort the data by weight in descending order.

The data is stored in a csv file named surveys.csv and the top few rows of that file are:

record_id,month,day,year,plot_id,species_id,sex,hindfoot_length,weight 1,7,16,1977,2,NL,M,32, 2,7,16,1977,3,NL,M,33, 3,7,16,1977,2,DM,F,37, 4,7,16,1977,7,DM,M,36,

  • Try to run the code
  • Now the model has all of the context it needs to produce runable code, that we can also understand
  • Let’s read through the code