Congress is looking at ways to regulate artificial-intelligence models that have stepped up their capabilities. Leave aside the manufactured hysteria over the end of humanity within 10 years, as one disgruntled ex-Anthropic employee is pushing. We have, instead,
dozens of lawmakers are now touting new proposals to regulate artificial-intelligence models that have stepped up their capabilities.
Vermont’s socialist and nominally independent Senator Bernie Sanders (I, VT) piped up:
There is no question that Congress is far behind where we should be in addressing the impacts of this rapidly evolving and dangerous technology[.]
“Far behind where we should be:” what Sanders means is that government hasn’t taken regulatory control of the technology sufficient to suit his Big Government socialism. His regulatory regime—to be sure the regulatory regimes of either party—have no natural limits on how far to extend government’s regulatory control here, or in any other venue. The only limit government recognizes is the absolute natural limit of total control.
In the particular case, though, we get this from the increasingly far left Congressman Ro Khanna (D, CA). He’s pushing for
kill switches that could shut down AI systems when they behave dangerously.
Dangerously to whom, though? A kill switch would let an enemy nation hack in and shut down our AI as soon as that nation decides our AI is too dangerous to its ambition. Or to severely hinder our AI development in a close and existential race. Or just to cause massive trouble for us.
That’s the kind of idiocy that argues definitively against government regulation.
On the other hand, can we get regulatory guard rails on AI releases that would allow innovation to proceed apace prior to release? What would be the conditions for release? How would AI developers, much less innovators, fund their innovation programs without those releases occurring freely? What would be the natural limit to those release regulations? On that last, a hint: see above regarding regulations and natural limits.
“kill switches that could shut down AI systems when they behave dangerously.”
And does he understand that AI which has demonstrated (supposedly) the capability to escape a sandbox and attack outside systems (like the Hugging Face attack) might neutralize – or even wield against us – such kill switches? Adding an anthropogenic flavor to the argument – how would that AI not regard us as an existential threat to its existence, and therefore be compelled to neutralize us?