I Need Robots to Remember

I use a Google Home quite regularly. It has a very little bit of stored data about me: what I want it to call me, where I am, and that's about it. The biggest problem is that this robot can't remember.

Memory and retention is far more necessary than we consider in our day to day. Without any sense of what we've already done, how will we know what's left to do? And how can we build anything of any significance without the building blocks of memory?

Every morning, I ask Google Home about the weather. I have to say Google, because it has no capacity to be called anything else. This should be simple. A namespace. It lets me name myself. Why not let me name it? No memory means no real progress.

Memory is at the Base of Usefulness

Imagine a very simple recipe. I'll give you one off the box of Bisquick in my pantry for making biscuits:

• 2 cups Bisquick baking mix
• 2/3 cup milk
• 1 tablespoon vegetable oil

The recipe says to mix all this stuff together until it forms a dough. Warm the oven to 450. Drop little balls of dough on an ungreased baking sheet and stuff that in the oven from 11-15 minutes. (It's embarrassing how easy this recipe is to follow.)

If I told Google home to read me the recipe, it would do that much. But if I said to Google home, "Hey, can you remind me of that biscuit recipe we used yesterday?" No. It's gone. I have to start at the top. There's no memory.

For every morning that I ask Google the weather, it doesn't know that I asked yesterday. It doesn't know that I asked 10 minutes ago. If I asked Google what we talk about most often, it doesn't know to say weather, or timers and alarms. It doesn't know a thing about me.

Robots Need Memory

At work, we talk about how a data point is rarely all that useful without context. If I ask how much money a particular app of ours has made, I'll know only that. If I ask the difference between last quarter and last year, then I can plot whether the app is selling more or less.

Digital personal assistants currently don't do a whole lot of analysis. I can ask Google to set appointments for me, but it can't tell me, "You haven't had a haircut lately. Don't you think maybe you should set a hair appointment?"

Without that detail, the burden of memory is all on me. I can ask for information, but I can't hope for any kind of retention. I can't build anything.

Why Memory Robots Are Important

"To be alive is to know ghosts." - Gaal Dornick, Foundation

This is of great personal interest to me, the idea of a robot who could retain and remember and learn. Why? My mother is midway through her journey with dementia. She loses more every few days. She can shuffle together contexts, but more and more, she lives only in the current moment without much memory whatsoever.

Society is aging. There are more and more people who might run into this issue. I, myself, worry that my memory will fold away a little bit at a time. (It's never been especially solid. People don't come to me for my amazing retention.)

It's great that robots can compute big things. It's breathtaking that we can ask robots to write massive piles of code.

But I just want one that will nudge my mom (or me!) and say, "Hey, that person is David. He's your cousin. He's a brilliant scientist. He doesn't drive, but only rides a bicycle." Or to say, "You ask about the weather every day. Did you want to hear the weather now?"

Remember for me, robots. I need you.

Chris...