It has always been the case that anything can be halted and, as noted much earlier in the thread, history is littered with examples of where vested interests simply claimed that wasn’t sensible, wasn’t possible, wasn’t cost effective etc. and yet it happened anyway.
I think the problem we have here in this and other examples is our inability to predict how advanced AI models might behave in any scenario. We’re creating the first species on the planet that has autonomy of action and is more intelligent than we are, personally I find that concernng.
It might be that AI will achieve miracles in developing incredible vaccines to cure diseases, equally it might engineer the perfect pathogen which spreads as easily (airborne) as the common cold, but has a fatality rate equal to Marburg Virus.
As Geoffrey Honto says: “If you want to know what it feels like when you’re not the most intelligent species on the planet - ask a chicken”!
Strikes me everyone is racing ahead (not least the Chinese) because there are billions to be made from this. Virtually everybody in the world would love an AI robot to run the house and virtually every business would like to replace every human worker with a robot that never takes time off or asks for a wage rise.
For that reason more than any other I expect stock prices in this sector to continue to climb, because I don’t think the public are aware just how fast this is coming.
The danger will be in malevolent actors using it, so for example a terrorist organisation could disrupt banking or air traffic control. It doesn’t bear thinking about…
JonathanG
Whereas I just think that’s what we’re being sold.
A thing only has autonomy of action if we are stupid enough to give it that. Apparently we are.
I’ve seen zero evidence it’s more intelligent than us. It certainly does as it’s told.
Technologies can do very bad things in the wrong hands. Society normally reacts by introducing regulations, legislation etc. I seriously doubt that approach will work with AI. The problem with AI is that it is unpredictable by nature. Combine that with the breaking down of international law, cooperation and institutions and you have a perfect storm on your hands. It is almost a guarantee that a malevolent organization will, at some point, abuse AI models to achieve it’s goals. It is even very likely that a malevolent actor is not needed to cause an AI disaster. The uncontrolled nature of AI models will take care of that. It is worrying.
I don’t necessarily agree with the first part of your post but, yes, this is clearly an issue.
I use AI models on a daily basis in my work and I find them very unpredictable (at least for now) in the sense that I am unable to predict what the approach will be that it takes or what the outcome will be. Using an identical prompt multiple times almost never results in the same approach and/or outcome.
The problem with regulation and/or legislation is that there must be a way to enforce- and check compliance. That is relatively easy for physical items (such as computer hardware) or traditional software (source code or byte code analysis, for example). How do you do that with something that is adaptive by nature? Any regulation or law will be outdated immediately.
I think there’s a danger in over complicating stuff. You don’t regulate on predictability of output for starters. A lot of us would never have never been employed if that were the case.
I see your point. It’s also an issue that AI models are mostly in the hands of a few very large companies. Companies that are difficult to capture in local rules and/or laws and almost impossible to capture in international rules and/or laws. And these are the companies that are very vocal about what they are doing. There are companies in less open countries that are developing AI capabilities that we know little or nothing about.
I think the international aspect is concerning but no more so than, for example, chemical weapons, government funded cyber attacks etc. We should worry about that but it’s one of a number of threats and needs to be seen in the context of us never truly having had “control” over what others do. We can only put our own house in order in the first instance.
When those companies operate on our territory then I really don’t think it is that hard. It’s just about having the collective will to do stuff. Flawed as it hay be, would we have even entertained the idea of a social media ban had we not seen others do it? Would we prosecute anything against any of the big guns had we not seen others do it?
I don’t think regulation is especially hard. The results of mixing AI, high res images and genetic info look superficially impressive fur example but could be easily mis-used, sold on etc. If you ask a company to demonstrate the lack of bias in its data set you instantly change the landscape. Insist they demonstrate how data is kept within company boundaries then save again. Don’t allow anything which uses personal data but reserves the right to monetise it. Get companies to list their data sets and simply don’t allow those who steal data.
It found me the required USB hub after much faff. Have just been using it to narrow down a guitar choice and it simply doesn’t accept that the guitar I own, and wish to compare to, exists. There are easier ways to be frustrated.
I think we’ve done the “Is John Naughton okay?” argument but the things he links to can be interesting.
I’ve links into a lot of stuff on UBI and this is a valuable contribution which might sober a few people up as regards UBI and AI. The idea that we’ll all be better off because “AI” is thoughtfully laid to waste here.
Ho hum. As predictable as the day is long.
Some people’s chats with Claude AI found to be publicly available online Some people's chats with Claude AI found publicly available online - BBC News
Anthropic’s Claude AI found to have escaped test environments and hacked external companies…
So it seems the OpenAI incident was not a one off.
I don’t see how it could be a one off. It’s effectively what it’s designed to do. The shock is surely that it’s a shock to anyone.
2 Likes
I think the shock is they didn’t notice until they went back and investigated.
Surely if a model leaves the lab and hacks a competitor they would be expected to have enough monitoring in place for this to be flagged up!
3 Likes
I’ve just watched a YT video called “Accelerate or die".
Amongst a great deal of interesting ideas, it suggests that it is “intelligence” itself as an existential being in its own right that is evolving and humanity is the ‘vehicle’ as it were, that is enabling the next stage of ‘super intelligence’ by producing AI which, if left to run without interference within unregulated Capitalism, will of itself become a superior being, freeing humans from the drudgery of work and providing for their welfare as a byproduct of an enlightened self-interest. Or that’s a précis of what I think it was suggesting.
Of course, despite my initial feeling of finding the argument persuasive, it could just be a load of old cobblers.
It’s like putting a small child in the driving seat of a car and leaving the keys in. What’s it gonna do? 
1 Like
Essentially that’s the argument of the super rich clowns who think AI solves everything and enables us to live in a work free utopia.
This is very relevant to that.
A dull read for many but essential for understanding that governments aren’t going to replace working income with a livable UBI.
I worry that Ai and associated robotics may decide that we cannot look after ourselves and conclude that we need looking after like we look after dogs, cats or cows. We may have our so called freedom completely taken away from us..