Data & Privacy
In this video, Hannah Fry explains what AI agents are and shares what happened when she built one.
“There’s this thing with AI called the lethal trifecta, which is if they’ve got access to private information, if they’ve got internet access, and if someone can give them an instruction that’s untrusted, then they’re not safe…”
Should you trust an AI agent with your sensitive and personal data?
This IBM web page explains how AI Agents…
Store and retrieve information and learn from past interactions
Use various sensors and data inputs to gather information from their environment
Use computer vision, natural language processing (NLP), speech recognition and motion sensors
Process information and make decisions using algorithms, machine learning, and neural networks to analyse data and determine the best course of action
Execute tasks like sending emails, generating reports, posting to social media, making purchases, responding to customers, managing and organising files, and even control physical entities like adjusting a thermostat or moving a robotic arm
-
A housekeeping agent, like DAN in Rogue Robot, uses real-time access to cameras and sensors all over the house to monitor the occupants activities and provide for their needs.
It also needs access to the family’s schedules, communications, messages, bank details, etc.
Despite all this data, the family in the scenario still ran out of protein supplies because DAN exists purely in the digital realm and there were no humans in the loop to conduct real-world physical checks.
The human in the loop is called the meatspace or meat layer. Marketplaces already exist where AI agents can hire humans to do real-world tasks.
-
The panopticon is a circular prison designed for constant, overbearing surveillance. The inmates’ cells are arranged around a central watchtower so that a prison inspector can look into the cells at any time, without being seen by the inmates.
This means the prisoners feel like they are being watched all the time. Constant surveillance like this can lead to hypervigilance, stress, fatigue and anxiety.
In this video, AI entrepreneur Connor Leahy draws parallels between panopticon and the effects of digital surveillance, algorithms, data tracking and social media.
-
Another concern is how secure your data is when it’s being used by an AI agent that also has access to external actors.
This research paper ‘Agents of Chaos’ explores some of the risks, including:
Compliance with non-owner instructions
Disclosure of sensitive information
Owner identity spoofing