I’ve been working on building a personal internet. I have my own server with thousands of hours of content, 20,000 books, and enough music to run a pirate radio station.
The more time I spend consuming my own content, the less I feed the algorithms. One problem I’ve had is scrolling too much on Reddit and Instagram. Starting today I’m taking steps to further reduce my exposure to these apps.
I still like to peek at IG and I have made Reddit into a safe place where the content is just about watches and other hobbies of little consequence. I don’t need to look at it more than once a day for a few minutes. I decides to take these apps off of my phone, and installed them on a tablet.
In my quest to defeat the algorithm I’m taking steps to move most of my photos into plex and off the all seeing eye of iOS which uses machine learning to analyze the photos on our devices.
I’ve also been learning about HAR or Human Activity Recognition. This is a field study related to using the sensors on phones and wearables to understand everything people are doing throughout the day. Whether we’re sitting, standing, sleeping, eating, driving, etc.
Data from gyroscopes and accelerometers can be used to determine all kinds of things if the desire is there. I’ve seen papers where scientists used the data from a gyroscope and accelerometer to map a study participant’s commute by assertaining direction of travel and distance covered even without a constant gps signal.
When skimming some of these papers I learned about “side attacks”. According to Google’s Gemini:
some examples of accelerometer side-channel attacks:
- PIN and password cracking: By analyzing the phone’s movements while you enter a PIN or password, an attacker might be able to guess the sequence based on patterns in the acceleration data [2].
- Keystroke recognition: Similar to password cracking, the specific movements involved in typing on a touchscreen keyboard can be picked up by the accelerometer, potentially revealing what you’re typing [2].
- Speech recognition: Recent studies have even shown the possibility of using accelerometer data to reconstruct spoken words with some accuracy [3]
This is information that can be gained from just two sensors. In reviewing other sensors out phones contained, LiDar is another concerning one. It can be used to map things such as our faces and the interior of buildings we occupy.
Think about data collected from biometric sensors that analyze pulse, temperature, sleep quality, voice stress, and other factors. They can be used to discover deep insights about our physical health, mental heath, etc. People today willingly enter their menstrual cycle data, sexual activity and more into these apps that sync with out health apps.
We tend to dismiss this collection because we fail to understand how it can be used against us or to our disadvantage. Machine Learning algorithms and AI can understand the data we’re leaving behind and fheh can do it at scale, meaning that it can do it easily for everyone, on the fly.
We are being analyzed by these algorithms in ways we have never been studied before and as end users we have some access to our personal data, yet we have a limited understanding of what the tech companies and advertising networks doing this gathering want this data for. The worst part is that the reason these algorithms are built is to influence us into making decisions, usually about purchases but more commonly about what we believe.
The bottom line is, it is not our job to imagine how all of this data is being or could be used against us, but once we know that it is being used to “influence” or manipulate us, then the common sense thing to do is limit data collection. The reason for this post is to highlight that it’s not just our browsing data or likes on social platforma that are being collected. Data about our health data, movement, commute routes, mental / emotional state, life stage, location, co-location (locations we share with others), habits etc are being collected and combined into extremely detailed and comprehensive profiles right on our devices by companies who are known to have collaborated in secret domestic spying programs that were against users best interests, were likely illegal, and were definitely ethical violations and a breach of the social contract users entered into when giving these devices access to our lives.
One of the things I have learned in reading up on Human Activity Recognition is that hackers have even figured out a way to use a gyroscope as a microphone by using a machine learning algorithm to analyze the vibrations generated during human speech. They were able to detect speech with something like 64% accuracy, determine the gender of the speaker with 90+% accuracy, and several other qualities of human interaction. Other scientists have been able to use a camera as a microphone, or have used wifi signals emitted by ordinary routers to see through walls.
We don’t know if this data collection is happening, but we do know it is possible. we simply have no idea what can be gleaned from the data the sensors in our phones collect all day every day.
I’ve decided that I’m going to switch to carrying a dumb phone when I’m hanging out with people and will only use my smartphone at home. I have a second smartphone I can take with me in case I need phone for directions or something else.