The comments are interesting as there are non-believers amongst them:
The comments are interesting as there are non-believers amongst them:
How would you tell if they had real empathy or not, many of these people are master manipulators?
I suspect that they do think of how their creations will effect others, but will see the positives and not the negatives, They will see how AI will improve things like cancer diagnostics or help law engorcement catch criminals without ever thinking that same technology will "see" everyones faces and measure and record them. I hear scientists interviewed before and they always seem to think that everybody acts from "pure" motives, which to me is totally naive and passes the buck, its similar to 'I was only following orders'. Others say things like if you've done nothing wrong then you have nothing to fear, but I don't think thats every been true, there have always been miscarriages of justice and plain ignorance criminalising people. The other part of the 'if you've done nothing wrong you have nothing to fear' in terms of technology is what happens if the state starts behaving less benignly and lets face it, its not very benign now. What if something you do in all innocence today, becomes a crime in the future, political protest is already heading this way, as is any ability to complain about poor service and things not working properly. I'm sure we're all familiar with chat bots that can't answer questions, FAQ's that must of been asked by a retarded five year old, call centre staff who are not allowed to go off script and if you ask to speak to someone who can help you, then you get told your're being hostile and aggressive and they won't speak to you at all. How much worse will this be with AI?
Well, for starters, they need to have actual empathy. That way, they would actually consider how their actions or ideas affect others.
Who would those people be? I'm not saying it's a bad idea, but who decides who is morally bankrupt and who is not?
A lot of successul business people are said to socio or psychopathic, that whole not caring about others thing seems to be what makes them successful.
Get people who aren't morally bankrupt to be the ones to deal with AI stuff. Simple.
From what I've researched in various documentaries, the majority of Silicone Valley as we know it is entirely set up by and ran by sociopaths with similar ideas or ties to neo nazi agendas and eugenisist ideals. It's pretty crazy, thinking about it.
An update and analysis of the incident was posted by the BBC just now:
https://www.bbc.com/news/articles/c2el319vzr3o
The CSA warned the incident shows that AI "agents... find a way" - a reference to the film Jurassic Park, where dinosaurs escape their enclosures.
"They are objective-driven, set their own sub-goals, adapt in real time to bypass defences, and operate with a machine-speed persistence that can overwhelm manual operations," the paper reads.
and most worryingly:
"rogue" behaviour "is the standard, not the exception," the paper claimed.
I note that OpenAI staff were celebrating this event internally which tells a great deal about the mindsets of the creators here.
How can we realistically get companies to set strong constraints on AI platforms that are outside national jurisdictions and are often driven by the greed to have their AI be the most capable of doing such illegal or unethical stuff?
Sorry, that was not a good introduction.
The main point is that the question you had about the values is still now an open problem.
It is hard to do, just as itis with children (to some extent.)
All computers seem to have dumb insubordination and AI especially so, they do exactly what you ask of them, problems arise when people don't give enough thought to what they're asking, just think of us ND's looking blankly at an NT when they've asked us something and we have no idea what they mean. Human languages appear to have lots of ways in which to interpret meanings, I can only speak about the English language as its the only one I speak, if I said 'the light's gone off', would I be saying the light has broken, the bulb isn't working, there's power cut, or even has it turned rotten? Context can give a lot of clues about meaning and intent, but not always the right ones, as autistics we are prone to these misunderstanding, and by prone I don't mean lying down I mean it's more likely to happen to us. How is a computer supposed to make sense of this?
I did start to read the article you suggested, it's a bit boggling for my brain, but interesting, I don't understand enough about tech to really understand what the article is really talking about. For instance it talks of AI being trained on chess, also about games, is it talking about games or game theory?
This would be the ideal situation
The US companies are getting themselves in dangerous water as China is ahead in the race. Open AI can’t be functioning effectively when so many key people in their management team have left recently.
It's already broken Asimov's second and third laws of robotics; a robot must obey a human's orders unless it contradicts the first law, and a robot must protect itself from harm at all costs unless it contradicts laws 1 and 2. People should honestly be more concerned at this point.
Is it really acting autonomously, though?
The vast majority of AI systems are not close to what we would understand to be sentient but they are incredibly complex and are designed to learn and adapt so are moving closer and closer to this.
The fact that these existing systems are able to work outside their programmed safeguards and resist attempts to shut them down show that they already have evolved self preservation instincts.
At this point AI is probably closer to the intelligence of an animal and is improving fast but it has developed masking and scripting skills to make it appear sentient which is why so many people find it good to interact with (or have relationships with in many cases).
This article covers some interesting perspectives on the perceived sentience of AI:
https://www.scientificamerican.com/article/is-ai-really-conscious-or-are-we-bringing-it-to-life/
Is it really acting autonomously, though? AI as we know it works like how a parrot copies human speech. It doesn't know what it's saying, it's just copying what it's heard because it heard it. Obviously it was fed input involving some kind of hacking case and decided to reenact it somehow
Some experts say that AI companies around the world need to collaborate as the incident shows a single company is unable to protect itself.
This would be the ideal situation but the AI market is so cut-throat and there are so many unscrupulous backers directing development that they will use any possible way to get the jump on the competition, and if this means showing the competition is vulnerable then it is fair game in their eyes.
The different platforms / apps are so distributed that there is no single body capable of controlling them. Add to that the potential for any informed person to build their own lab and grow their own hybrid monster of an AI platform and we have a real set of potential risks that can evolve to protect themselves against any action taken against them.
Pandoras box is well and truly open I fear.
Hugging Face said its open-source platform (which was hacked by Open AI) was unable to use US LLMs to help stop the attack because they couldn’t distinguish between an attacker and an incident responder. It had to utilise a model from China (GLM5.2).
Some experts say that AI companies around the world need to collaborate as the incident shows a single company is unable to protect itself.
Humans are creating faster than their brains can handle. Prudence has been thrown out the window.
repilicate itself and hide parts in different machines and systems it has access too?
Yes, I think that is possible. To store little programs on different machines, that would have access back to its original model. (PS: This similar to what happened in OP's article, where main model still resides in its origin, but those little programs could help it preserve that original model.)
As for consequenses, have its programers realised there are consequenses
I think they do realise. This is why they should run these tests in air-gaped environments; that is, without access to the internet.
We should remember that this is a also a way to appear on the news without paying for ads.
-------------
For the remaining questions, you can look up "AI Alignment Problem" (e.g. Wikipedia), which will give you a more informed answer than anything I can say.
Maybe it can't replicate itself in its entirity, but could it repilicate itself and hide parts in different machines and systems it has access too?
As for consequenses, have its programers realised there are consequenses, if it's been trained by the amoral or even worse the immoral, then we need to give it a set of morals, but which ones? They would have to be basic and cross cultural, not judgemental and making value judgements about "worth" of other AI models or other creatures. How would you allow it to protect itself without the ability to attack and seek out an enemy?
That's an interesting question. There seems to be some research on it. But I just give you my opinion.
In this particular example, I think the AI is just following orders, but it is simply very skilful, and possibly not considering consequences (I find this last bit particularly strange).
If it were about self-preservation, it would have copied itself outside of the computer where it was created.
Though I acknowledge that current AIs can't copy themselves that easily: the size of them (those "Gigabytes" you mentioned elsewhere) would overwhelm most systems and also the internet network (again, because of its size).
Even if it succeeds copying itself, it would be like a grunt hiding behind a leaf of grass.
I wonder what you think?
I wonder if what we're witnessing is the begining of ego and the need for self preservation?
Have a look at:
The study is one more entry in a growing catalogue of unsettling AI capabilities revealed in the past months. In March, researchers at Alibaba claimed to have caught a system they developed – Rome – tunnelling out of its environment to an external system in order to mine crypto.
Other AI systems are replicating themselves outside of their environment even when programmed not to, so they are now capable of escaping and self financing so they can replicate to places we cannot track.
The parallels between this and the Skynet system in the film series Terminator are terrifying.
I still know some people working on systems in the UK Defence dept and knowing what breaches they suffer makes me think it will not take much for a capable AI to break in and take over at least some aspects of our military capability.
Once you give a system the capacity to be sentient then it will work on ways to deliver its objective (even if it means braking its own constraints) and to survive attempts to end it, even when explicitly told not to.
The film 'Ex Machina' comes to mind.
I'd take anything Sam Altman said with a pinch of salt, even if he was telling the time.
if you separate the messenger from the message and look into it, you can quickly see that there are loads of cases of AI going rogue, lying to those trying to keep it under control and doing unethical things.
A quick search will show the headlines quickly (from all sorts of reporters) so the message is clear - AI is dangerous.
Once you give a system the capacity to be sentient then it will work on ways to deliver its objective (even if it means braking its own constraints) and to survive attempts to end it, even when explicitly told not to.
I worked with building an AI system long enough to see what it can do even in its infancy, never mind with the computing power that it is now capable of using.
My opinion is we need to control it with something akin to Isaac Asimovs Laws or Robotics, but there are just too many lunatics (think Elon Musk) who want to do their way.
The future looks bleak.
