Apologies for the long delay since I last wrote. I’ve been obsessively working on my Notebook Index project, and when I get wrapped up in writing code it’s hard to switch my brain to writing English. I recently reached the milestone I discussed last time; I’ve scanned the majority of my diaries (8 of 12 volumes) and built the “On This Day” feature I wanted. The software is good enough for my personal use. So, it’s time to pause programming and catch up on writing. My main lesson from this experience so far? While there is much worthy of debate about AI these days, one thing we should all agree on: AI is breakthrough technology for digitizing handwritten text. The tracings of ink on paper can now gain abilities we take for granted in our digital lives.
Recognizing text in images isn’t new. Your iPhone camera has had optical character recognition (OCR) since 2021. While it’s quite accurate for printed text, it struggles with handwriting. Consider this example. It’s an image from my diary, written in 1995 when I was traveling in Belfast, Northern Ireland:
This is the text that the iPhone’s built-in OCR software extracts from the image:
24 Many Hostel 1:15 am Tow a’ writing. I might take tomorrow off and try to catch op an writing, and other wite relax. I speat the early afternoon a the Than kill. I chopped into the Proguensive Unionist Party ffice and picked up 3 interviews sell a follow up for Friday. The I trucked back to the hostel and almost missed Pand, who took me to meet Mre. Evehyn Dickuson, Belfust midut 153 years. I havent recopied that intevienz yet. Going ave my notes is time-consuming and boring. Plus, theme’s us good place to work. Sir Aside from my excellent and expensive dinner at the strand, mpevening has han copying interre notes.
Now, given how bad my handwriting is, this isn’t terrible. However, it’s also barely readable! While I haven’t counted, I’m guessing there are more words with errors than words recognized accurately. Contrast this with the results of feeding this image to OpenAI’s gpt-4o-mini model:
24 May 1:15 am Hostel
Ton o’ writing. I might take tomorrow off and try to catch up on writing, and otherwise relax. I spent the early afternoon on the Tunkill. I dropped into the Progressive Unionist Party office and picked up interviews and a follow-up for Friday. Then I trucked back to the hostel and almost missed Paul, who took me to meet Mrs. Evelyn Dickerson, Belfast resident of 53 years. I haven’t recorded that interview yet.
Going over my notes is time-consuming and boring. Plus, there’s no good place to work. Aside from my excellent and expensive dinner at the Strand, my evening has been copying interview notes.
While the AI made three mistakes, for me, it reaches an important quality bar: I prefer to read the AI transcription to my own handwriting. I refer back to the scanned image only when it looks like the AI garbled something. It also makes me confident that full-text search will be more accurate, increasing the value of this digitization project.
One of my biggest goals was to add an “on this day” feature to my paper diaries. AI made this easy. I instructed the AI that it was scanning a diary composed of dated & ordered entries, and then I asked it to group the pages into entries. Just like with handwriting recognition, there are some mistakes, but I estimate over 90% of the entries are correctly identified. With that done, it was easy to write a feature that shows me all entries written on this day in prior years.
“On this Day” came just in time! The day after I finished that code, it reminded me that my first trip to Rangeley, Maine was 30 years ago that day. We went there — my granddad’s childhood hometown — to celebrate my grandparents’ 50th wedding anniversary. While I’ve been back to Rangeley many times since then, that first trip was special. Getting that reminder was awesome.
My UI, with details redacted because who wants to share their high school musings with the world? Note how things are grouped by date. gpt-4o is pretty accurate at determining the boundaries between entries.
As I mentioned at the beginning, now that I’m almost done scanning my diaries and have “On The Day” finished, I could stop here. But the progress I’ve made so far makes me excited about future possibilities, mostly about using pen & paper in more places in my life knowing that it’s getting easier to digitize what I write. Maybe I should start keeping my reading notes in a physical notebook and scan the notes later! Or keep paper project logs!
Perhaps a little-foreseen consequence of the Age of AI is it will make pen & paper newly practical in the digital world. If so, that’s a future I will enjoy.