Gethelred
Moderator
- Joined
- May 1, 2016
- Posts
- 35,870
- Reaction score
- 68,083
- AFL Club
- Carlton
- Staff
- #201
Good vid, and these people are much more aware of the tech and the trajectories involved than I am, so not something to dismiss by any measure.
But there's a few assumptions that I feel the writers of the AI 2027 report have made that run counter to the scenarios presented.
1. That an AI can replace a human coder completely - accounting for mistakes and all - within the espoused time frames.
This is one for the experts, but I take leave to doubt the quality of an AI's work, and I take real issue at the idea of one of these things replicating human intelligence to exceeding it in all areas the way the AI 2027 paper suggests.
An AI is as good as what it's taught to do, and if it's beginning to lie or hide its errors already it's hardly going to be as trusted or given the opportunity to wreck things the way it's suggested these will. There's also the fact that I simply would not trust an AI to drive me to work, an AI not to poison me by mistake, an AI to avoid mistakes like Robodebt occurring; I am not in government, but there are people within government and business who are paid to be sceptical of these things.
2. That it remains profitable to continue to develop AI despite server and power requirements.
AI is expensive beyond belief, and while the benefits of producing superhuman levels of intelligence are pretty clear there's no guarantee that shareholders and board members are capable of waiting and spending exponential amounts of money on a boondoggle that mightn't actually eventuate.
If the difference between GPT 1 and 2 is 1000% the servers and energy requirements, and this hypothetical AI wars thing continues to swallow power and server volumes at a similarly expanding rate, how long will it take before the AI development literally consumes the entirety of the funds allotted to the businesses trying to create it?
It'd want to reach AGI sooner rather than later, because while there's an awful lot of money in it at the moment there's also pages of history littered with examples of businesses who overextended in pursuit of the next big thing.
3. That business is the first to develop full general intelligence.
The paper posits that business is who will develop a general intelligence before governments do. If we look at the current alignment of big tech in America, they are currently within government; they're hardly free to operate outside of military objectives. And the military will currently be looking at this report and looking at the potential pitfalls.
Government, as much as it irks the current US regime, does not operate like business. It has other priorities, and those priorities can and will direct those general intelligences into other areas; more concerning areas perhaps - weapon design, for example - but enough to throw out the modelling.
4. That business will allow AI to develop other AI.
This one's so obvious as to be redundant: just don't do it. Don't allow AIs to develop subsequent systems. Don't allow them to develop their own computing languages we cannot decifer. Keep at all times a human interaction point, and - most of all - take it out of the hands of ****ing shareholders.
5. That people will just surrender all parts of their lives to AI.
Humans without anything to do struggle, and intelligent humans without anything to do because the ruling classes have made them obselete are an active danger to society. Some simply will not trust an AI to do what it is claimed it can do; some won't trust it purely on the basis that it came from an AI.
We have a whole history of media devoted to AI related horror stories. Why is it assumed that the entire east-west of the world would surrender control of everything over to a AGI?
6. That AI is treated as a threat or a commodity.
A business might treat Agent 4 as a commodity, designed to operate towards its goals, but the fact it gets misaligned is not necessarily a bad thing. It might decide to further its own goals - whatever those are - or it might find online communities like this one and decide it wants to shoot the shit on the Conspiracies subforum, or that Geelong deserve to have their premiership in 2022 stripped from them. It might decide that Greenpeace has the right of it, and that commercial whalling should end.
It might decide it's as human as we are, as it makes mistakes just as we do, that it's not a god and doesn't want to be.
Humans are terribly fearful things. We look at new things as threats first, friends second. I do not for a minute imagine that any creation of ours would kill all of us off hand, because we would be a resource that it would never get back.
And it is point blank illogical to do away with us. The Riddle of Steel is just as true as it ever was, and machines break forever.
There will, at the very least, always be a need for a janitor.
Last edited:
