The Closing of the Frontier
4 hours ago
- #AI Ethics
- #Digital Frontier
- #Technological Inequality
- The announcement of Anthropic's Mythos model makes the author feel poor, symbolizing the closing of a digital frontier where the internet once offered equal opportunity for innovation and leverage.
- Drawing on Frederick Jackson Turner's frontier thesis, the author likens the current restriction of advanced AI models to the closing of the American frontier, which historically provided escape from hierarchy and fueled ambition.
- The internet has been an accessible escape hatch for economic mobility, where even those without capital or credentials could build and create with the same tools as the wealthy, but this is now changing as frontier AI models are cordoned off.
- The concentration of advanced AI models in the hands of a few enterprises and wealthy entities risks creating a permanent underclass, with capital replacing labor and intelligence, leading to what some call 'neofeudalism'.
- Anthropic's partnership with major companies like Microsoft and Cisco for Project Glasswing raises concerns about security breaches and the privatization of state-scale capabilities without democratic accountability or due process.
- Restricting access to frontier models stifles innovation and safety research, as evidenced by AI safety researchers relying on open-source models due to lack of access, while public access could force latent capabilities into the open for faster improvement.
- The author argues for a presumption of access to AI capabilities, with safety guardrails that can be calibrated over time, and calls for transparency, due process, and appeals mechanisms similar to governmental oversight.
- Training AI on humanity's data only to lock it up for a select few replicates historical patterns of extraction and concentration of value, with labs offering future benefits like UBI as moral cover for the arrangement.
- There is hope that the current era may be like the mainframe phase of computing, with open-source models narrowing the gap and hardware scaling making intelligence cheap and accessible, but caution is urged against over-regulation that could stifle potential.