My first mind-blowing experience with AI
Playing with OpenAI's "Advanced Data Analysis"
Last night was the first time I have been truly floored by what I’ve seen AI be able to do. I upgraded to ChatGPT premium and tried out the “Advanced Data Analysis” feature. I uploaded a messy Excel file with data on alcohol consumption in countries around the world and asked the ambiguous question, “what are some interesting trends in Wine consumption based on this analysis?”
In ~5 seconds, GPT wrote Python code to read in the Excel, understood the structure of the spreadsheet (again, with multiple tabs and a cover sheet, not formatted well), and explained its process to me in clear step by step language.
This task alone — parsing a complicated spreadsheet and getting it prepared for analysis — is something that I have been paid to do, that lots of analysts are paid to do today.
I then asked it to make a chart of consumption by country, with different lines for different types of alcohol, and it did that immediately. This would have taken me at least an hour to do in Python and it did a better job than I would have in a matter of seconds:
What struck me about this is how fast this is moving and what the implications are for my career and lots of “knowledge work.” I will explore my thoughts on this in a later post.
Like you perhaps I used to be a knowledge worker in the 1980s, prior to the Apple Macintosh, Excel, and Powerpoint. I remember when I first saw a computerized spreadsheet, an Excel precursor called Visicalc. It was adding up rows and columns of data, it was copying and pasting data and formulas. I thought, "Oh my god, that is my job!" I was an accountant and at the time I got paid for making manual spreadsheets, charts and slides that would take far less time now, even without AI. What saved me, as the production of these outputs fell and fell, was that there was plenty of important 'knowledge work' to do BEYOND the grind of spreadsheet preparation. It seems clear that new AI tools will unburden us from the drudgery of data preparation, analysis, and reporting. But there will always be decisions to make on top of the analysis. There will always be a need to decide what should be the next prompt to the AI.
Or maybe you think the AI will take that over too?