Track Record
The thesis behind Unsupervised Learning has a paper trail. Digital assistants that know you, everything exposing an API, the assistant acting on your behalf, and AI reshaping work: first stated in 2006 and restated in dated pieces across the twenty years since.
The arc
The assistant and the daemon
In 2006 the blog describes "an application that spends 24 hours a day optimizing your life." 2008 names the Personal Daemon, a device you carry that broadcasts who you are and what you want. 2010 says the point of social technology is to present each of us as an API.
Universal daemonization
The jobs are going and not coming back (2013). Daemons act on your behalf, managed by your personal assistant (2014). Everything gets a daemon, and Siri sends the request to a venue’s API for you while acting as your 24/7 advocate (2015).
Assistants are the new applications
"Before: humans use company apps. After: our AI assistants call company APIs." The Real Internet of Things ships in December 2016, alongside a post predicting that tricking our AI assistants into doing things they shouldn’t would be one of the coolest new attack surfaces. 2017 adds the Life Dashboard and personal sensors as the data source.
Still on thesis
Sensors plus models will read human activity and intent. Machine learning will capture the je ne sais quoi of human existence. AI art in late 2022 opens the threat to knowledge work, and AI is about to feel like AGI.
Context, then the digital assistant
Custom models are AI’s killer app because they carry your context (Feb 2023). SPQA is the architecture that replaces most software (Mar 2023). AI’s next big thing is digital assistants (May 2023). By 2026: we are all building a single digital assistant, and products are becoming APIs.
The receipts
Every dated statement, with a link to the original, lives on the full timeline: more than a thousand statements from 2006 to today, most of them blog posts and tweets, plus newsletter and LinkedIn posts, organized by thesis.
Open the Predictions TimelineThe scoreboard with confidence levels and outcomes is at danielmiessler.com/predictions.