Christmas shopping the IBM way: computerized gift selection in 1962 | Hacker News Reader

Christmas shopping the IBM way: computerized gift selection in 1962

righto.com

115 pointspicture15 comments

Informative. Thanks for sharing.

I smiled at

> "In a hurry?... the whole machine process takes only 10 seconds."

Not much slower than some websites today!

Was this the first machine-assisted shopping recommendation? The precursor to modern day google-analytics & social-media assisted shopping recommendations?

i wonder if they already had a database of items and attributes or if this was all built custom for the demo.

kind of amusing to think how similar it is to many modern day data science projects. it's entirely plausible to think that they may have had a roomful of people at one point, punching out punch cards describing which items were good for which demographics. like an actually mechanical mechanical turk.

The video said that Simpson's employee Frances Cook spent six weeks choosing the 3,000 gifts that could be suggested by the computer and deciding what sort of people those gifts would suit. So it sounds like one person did the data collection.

hah! i relistened to both clips trying to find this, not remembering it (and questioning my sanity!).

it's actually in the "Did you know?" text that accompanies the radio clip.

either way, i really enjoy reading your blog and twitter posts about computing history and geeky explorations and have for quite some time. thanks!

Sorry, I was sloppy with my notes.

i'll say it again. thank you for all that you do with the blog. from the innards of bitcoin to computer history and explorations of old hardware. it is incredibly inspiring, engaging and clearly done with great effort and care.

thanks again.

So is the source code available?

The problem is described as the customer answering basically 5 different criteria (age range, gender, interests, category, price range) about the gift recipient, and the computer returns a list of 10 gifts.

Based on the description of the problem being solved, I'd guess they would have used "FARGO", a sort of report generation language, which probably had some predefined way to "AND" those criteria together and print out 10 matches. FARGO may have morphed into RPG by 1962, but the idea would have been similar.

http://bitsavers.org/pdf/ibm/1401/C24-1464-3_1401_fargo.pdf

Unfortunately no source code. I've thought about how this might be implemented. The video says there are 3000 items; the IBM 1401 only has 4000 characters of memory (unless it has the 12,000 character expansion memory box, which I didn't see in the video). So most of the data must be on magnetic tape. At 15000 characters per second, ten seconds of processing would permit 50 characters per item, which seems reasonable.

So my theory is that the program runs through the 3000 records on tape and matches (ranks) them against the customer's input. It could keep track of the current top 10 matches as it goes through. Then it can print them out at the end. (I'd expect some pseudo-randomness so customers don't get the same results.)

Of course, they could do something simple, such as having a small number of pre-computed lists and just print one based on the customer's input.

So basically a full table scan, where the table is stored on tape instead of DASD. Sounds very plausible!

No, if 10 items are to be kept - that would be a select sort, heap one was invented in the same decade but later. With the amount of items to be printed kept in memory - just read the tape and grade the items vs. the current selection.

This sounds like an incredibly sensible approach.

We can probably assume each gift was assigned a single gender too (and possibly a single age category). Would that have helped as they could then pre-sort the tape for speed? I assume a seek is faster than a read, but don’t know enough about the architecture!

There's more than one tape drive, so they could have "sharded" on gender, for example.

Reply on news.ycombinator.com