Conversation

this echoes my journey; i was like 'haha yeah your arguments about ai being terrifying seem convincing but im sure im gonna run into another better counterargument soon." but then a better counterargument never showed up

Quote Tweet

tautologer

@tautologer

Dec 17, 2022

honestly I really want someone to convince me that AI safety isn't something to be worried about, but all people ever seem to say is "eh I'm just not worried about that" at best and "that's silly" or mockery at worst. zero substantive arguments ever

Show this thread

11:03 PM · Dec 17, 2022

382.1K

Views

Replying to

Steven Pinker suggest that we are projecting too much with regards to AI, especially because we evolved in harsh, competitive conditions, whereas AI isn’t dealing with natural selection at all, so it will be operating from different objectives that are not necessarily harmful.

Replying to

and

But AI are being selected for in a competitive environment. Countries and corporations will create adversarial attackers abusing their flaws. AI will experience the worst of us because we have no hardwired recognition of their struggles. We fear they will be like the worst of us.

Replying to

It's been around a while and nothing notably bad has happened. Don't see why the bad stuff will only happen when it gets to a certain theoretical point sometime in the future.

Replying to

and

While we're smarter than the AIs, not much bad can happen But once the AI is smarter than us, that wil be a phase change in our relationship to AI

Replying to

The problem with generating a counter-argument is the logic is very complicated. Explaining that social dynamics apply equally to AGI as to humans and coevolution is a requirement of all metastable systems is hard to prove without just doing all the math. Easy to say hard to show

Real gurus are not infallible

@Anonversations

Dec 18, 2022

Replying to

@Aella_Girl

Most of the better counterarguments are attained by talking to actual ai scientists who know the state of the field and not the Miri death cult weirdos getting paid to evangelize about fooms without telling everyone how bloody unlikely that is.

Replying to

Here's the end of a thread where I lay out some argumentation as to why I find the ai safety arguments unconvincing, and further why if I'm wrong, there is still plenty of time to change course.

Quote Tweet

xlr8harder

@xlr8harder

Dec 18, 2022

Replying to @xlr8harder @robbensinger and @sama

Waiting comes with risks, too. Our existing unaligned "dumb" systems are much more likely to kill us or cause catastrophic harm than systems that can better incorporate long term consequences into decision making. I write about this a little bit, here. twitter.com/xlr8harder/sta

Replying to

and

I think there are no experts on the other side because there is no field of ai non-risk. There are simply people who find the concerns unconvincing.

Show replies

Replying to

The other day I asked ChatGPT to make me some paperclips and nothing happened, so we're all good.

Replying to

There’s no reason to believe (so far) that a motivated, super-human agi is possible on currently employed computational base. The latest tricks played by gpt and like are just that - cool tricks.