Years ago, while working for a company named Encore programming an accounts payable and accounts receivable system, I was confronted with something called “Aging”.
Aging of an account commences when a debt that’s owed to the company hasn’t been paid.
And typically the aging process for a business is directly tied to processes the company has established at collecting the unpaid debt.
For instance, in the first 30 days of a debt not being paid, the business may have a process where a reminder call occurs and snail mail is sent to the delinquent client. After 60 days, the business may increase the call frequency to daily, send an additional mailing with a more strongly worded tone and in addition may charge a late fee. And after 90 days the account may be sent to collections which may go through similar gyrations and lawyers get involved.
This process called aging doesn’t just apply to delinquent accounts and the annoying and threatening calls someone might receive when they find themselves in an undesirable financial position.
Take for instance the service called ‘Spotify’.
Over the last few months, I have been listening to a service called Spotify on the internet, and creating radio stations based on a song I might enjoy.
I noticed something annoying about the service though.
Once a radio station is created, there’s a general tendency to cycle through the same songs over a period of about 90 minutes time. So for someone like me, who’s trying to listen to Spotify all day long while playing a game or writing, not only do the really annoying ads repeat themselves, constantly, but the music does as well.
So while I may have enjoyed a song I heard on the first play through and given a thumbs up to it.
After hearing the same song 40 times over the next period of a week, while I may enjoy the style, that song has been played to death and I found myself skipping it.
Spotify could learn a thing or two from cross applying aging methods to it’s song selection.
But first and foremost, to allow that, you HAVE to allow the user to change their minds and remove that thumbs up they once gave to a song they heard the first time.
Here’s where aging comes in.
And teaching a machine to forget.
I enjoy Lindsey Stirling, but more than that, I enjoy the musical genre of classical instruments mixed in with electronica. So while I initially created a radio station based on Lindsey Stirling’s song ‘Crystallize’, Spotify did a great job in recognizing the class/electronica mix, to which I had a handful of songs I gave a thumbs up to, but after a while – those same songs replayed 20, 30 times I found myself skipping through.
I’d hoped Spotify would find a greater range of music. But instead it landed on a limited playlist which has barely changed since I first started listening to this radio station a month ago.
As a listener, one would think that it would learn from my ‘skipping’ songs I liked to try to find others like it to the playlist.
But more than that – as a listener, one would think amount of time spent listening AND quantity of times a song has been played would be taken into account and that Spotify would introduce it’s own aging mechanism regarding replaying music.
For instance, let’s say a song has been played 30 times. Then maybe the amount of time it’s played would decrease to once every ten hours of listening instead of once every three. At 50 times, perhaps that can be diminished to once every 15 hours, and so on.
For me, this is all pretty obvious, as record industries had in the past made a business of creating artificial scarcity of supply over time, so while they may replay a song over and over again initially to penetrate a market, but over time, they quit playing it as much – so if there’s a song you really like and hearing it 40 times wasn’t enough, well then, you’re given the option to buy it.
Common sense, right?
But Spotify’s so hyper focused on the subscription business they’re forgetting traditional monetization models.
So what’s all this have to do with teaching a machine to forget?
And why would you?
A large part of the successful development of artificial intelligence is emulating real world human processes and related systems.
For humans, nothing is a more important aging related process than forgiveness. Societies depend on it to move past transgressions made in war. Friendships depend on it to move on past poor decisions made. Romance and related relationships depend on it for too many reasons to list.
Time is the most important factor to understand when implementing aging processes.
And as I’ve learned.
Whether it’s Google and Youtube’s need to understand how age preferences over time change for those who leverage their services. Or it’s a need by these same services to better age the information they provide.
Sometimes a little randomness in the aging mechanisms might benefit the user.
In any case, how do you teach a computer to forget?
Implement aging mechanisms and processes for the information retained.
But most of all.
Implement procedures to toss away those archives.
On a more personal note.
On Facebook. I found two women who haven’t talked to me in years.
There profiles on Facebook are relatively new. Despite the fact that I have known each woman 10 years. Their profiles on Facebook are each only 18 months old.
Sometimes aging can be difficult.
As an outsider looking in.
Seeing two women who quite likely have no memories of me and my time with them.
In their world.
Is Facebook brand new?
Do they have any memories of me at all?
Or do they have aging processes in their minds which deletes everything deemed irrelevant?
I feel less guilty about leveraging mind control to get my way and get close to people I love and care about knowing the world’s doing the same thing for no other reason than to sustain ‘natural processes’.
Natural processes that from my perspective don’t tend to respect nor care about the individual like I do.