July 31, 2023

People Just Really Want to Say "Enhance"

Firstly, let me just say thank you to all of you who came and met me at CONvergance last month or who generously invite me into your inboxes every month. This the 2 year anniversary of the very first Shed Letters, and I'm still having a blast sharing with you all! If you haven’t been with us from the start, let me invite you to peruse the archive at your leisure.

I don’t have any free book newsletter swaps to offer you today, because I quit the service out of frustration that all the young adult book giveaways were for “clean” “Christian” books only, and, well, Secondhand Origin Stories doesn’t make the cut, because it has queer people existing and isn’t Christian Evangelical propaganda. But that’s a whole other subject for another day.

I do have a fun little project going on over on my Tumblr, where I’m posting 1 chapter a week from Secondhand Origin Stories, in case you haven’t read it and think that would be a fun way to have it delivered to you. You can check that out here.

But today, I have a different subject in mind.

Picture me, at 5 or 6 years old, dressed as a flower and wandering the unfamiliar 80’s beige house of the unfamiliar adults hosting a Halloween party. There are other children around, somewhere, but they hadn’t held my funny little mind’s attention.

I find a room with no-one in it, and there’s a perfect shaft of light illuminating a button on the wall. It’s shining, the color of cherry candy, and the perfect size for my little palm. There are words near the button, but I can’t read, so they barely encode on my memory as present.

You know immediately what I wanted to do. You are, perhaps, not surprised that I stacked some objects from around the room in order to build a platform to let me to reach my prize.

I succeed in pressing the button, with no specific idea what would happen if I did. In my world, bright red buttons were usually toy related. But, no delights suddenly appeared. I tried it a few more times for good measure, then more or less accepted my loss and moved on.

There was excitement not long after though as a whole fleet of firetrucks and police cars encircled the Halloween party.

But there was no fire and no intruder. The host was fined, and all the exciting alarms and sirens left.

Later on in the party I tried to hit the beautiful button again, to a similarly anticlimactic response. By the end of the evening our host had racked up some notable fines for false alarms, and somehow I was discovered as the illiterate and innocent perpetrator.

And that was my introduction to the concept of the home security system.

I’m currently writing a sweet, low stakes, short novella, tentatively titled Doll’s Eye View, exclusively for all of you lovely newsletter subscribers. It’s from the point of view of Martin. Those of you who’ve read Secondhand Origin Stories will remember Martin as the sentient AI security system at the heart of the superhero home/base of operations. Between that and all the AI in the news, AI and smart homes/security systems have been on my mind.

To prepare for writing Doll’s Eye View, I finally went and read the Murderbot Diaries books, by Martha Wells, which were every bit as fun as I’d been promised by the many people telling me to read them over the years. The books are told mainly from the point of view of a Security construct who I love, and there’s also a sentient AI spaceship that, like Martin, considers itself to be a home and cares deeply for the people it shelters.

So it got me thinking a little bit deeper about what the point of view of a non-humanoid AI would be like. Now, my wife, Ty, works for a company that provides software and hardware for the myriad of large and small security systems all over the world. Since I write near-future scifi, it seemed like a no-brainer to ask Ty to tell me about the future of home security and home automation, and how they think AI will show up.

The result is going to be a series of newsletters talking about home, AI, and the future and present of home security and home automation, and what I think this would subjectively be like, if you were the system itself.

I’m starting off what Ty says is the greatest demand and the area where the most current development is focused.

Firstly, she says that AI applications right now are focused on making cameras better at detecting useful footage.

Seems sensible. You don't want false alerts or false alarms.

If, for example, you have a camera with a motion sensor aimed at your front door, and it sends a notice to your phone every time it’s activated, you want it to focus on people, not, say, leaves blowing in the wind.

Of course, we know that AI can generally identify the shape of humans, now. Even Midjourney almost knows what a human hand looks like, and it’s improving all of the time. My feelings on AI and the arts are, again, a separate rant for another day.

For now, humans are still smarter.

I particularly enjoyed this story of Marines immediately defeating an AI they’d been helping train to identify approaching people through the ever-so-high-tech method of putting boxes on their heads.

All that money. All that technology, and a solution Bugs Bunny would have come up with defeats it instantly. It’s just magnificent.

Funnily enough, the biggest request that they get at my wife’s work- and they get it all the time- doesn’t seem like a request for an AI application at all. People are calling for 4k cameras for their home security systems. Their company doesn’t make that, because it’s absurd and the size of your video storage would need to be ridiculous. That’s theater level video quality for an image that’s going to be primarily displayed in the app on your phone.

I suggested to her that people just really want to be able to say “enhance” and have the security system know what they want to see and suddenly give them a crystal clear image of the face of the perpetrator they imagine catching with their security system. Like on Star Trek, or any of the billions of cop shows on TV. Then, victorious, they will present this image of the badguy to the police and be vindicated and safe.

I suspect at least somebody is going to try a potentially profitable shortcut.

There’s an AI image enhancing tool created by Duke university that advertises up to 64% better clarity at refining human faces from a blurry batch of pixels. It uses “generative adversarial networks” in which two AI networks are pitted against each other- one developing images and the other accepting or rejecting the work to push the image to greater clarity, using a database of human faces to compare and pull from. Eventually the gatekeeping AI says that the image is good enough, and provides the result.

Ty’s opinion is that the first security company to offer this as an integrated part of their system will have anxious but authority-trusting demographics absolutely flocking to their service.

This, to be clear, is bad.

The images produced by Duke’s system are indeed sharp, plausible, and look good. But the developers are clear that the program “imagines” or “hallucinates” details like wrinkles, eyelashes, stubble, etc. It looks like it’s taking a blurry photo and making it clear via the magic of technology, but that’s not actually what’s happening. It is explicitly making stuff up, based on other faces it's seen.

The creators of the tool have published a paper that discusses biases in their datasets and how those biases influence the results, but it hasn’t been well covered and it’s not very accessible to the average anxious home owener. (https://arxiv.org/pdf/2003.03808.pdf ) Like most AI of it’s kind, it has a significant racial bias and struggles with people of color.

But I feel confident that, eventually, someone will put forth a feature claiming to enhance images just like on CSI, and use generative models to create fake images of people photographed by home security systems. What will the legal system do with those generative images in the court of law? I have no idea. I expect it depends an awful lot on who and where the early cases arise, and how much the people involved understand AI.

The thing that really kills me about all of this is that, due to some of my previous jobs, I have known an awful lot of burglars. I’ve spoken to them about times they’ve broken into people’s homes- what went through their minds, how they did it, what happened, etc.

I think people imagine criminal masterminds, casing the joint and plotting for days, with malice in their hearts and mustaches to twirl.

By far the story I heard the most often went, generally speaking “I was super high, and it looked like nobody was home, so I checked the door handle and then broke a window to see if I could pick up their TV and try to make a sandwich cause I was kind of hungry.”

Genuinely, the thing about just wanting to go in and get something to eat because they were severely high came up a lot more often than you would expect.

But that’s not how TV or movies work, and most people haven’t chatted with burglars, so that’s not the story most of us expect. And security systems get sold by projecting what people fear and providing an apparent safety measure against that fear. It’s going to be a hell of a thing to see how this gets used and how many lawsuits come out of it.

Please join me next time as we take a look at smart homes, what they actually do, how they can be used, and how they’re actually being used today. Then we can get to the insider experience of writing a fully sentient AI home system.

Till Next Time,

Lee Brontide

If you know someone who you think would like to join us, please feel personally invited to share any of these emails, or send them an invitation to sign up here. And remember that Secondhand Origin Stories is available for free as an ebook here, or in paperback form from your local independent book shop or online retailer.