+- Scivillage.com Casual Discussion Science Forum (https://www.scivillage.com)
+-- Forum: Science (https://www.scivillage.com/forum-61.html)
+--- Forum: Architecture, Design & Engineering (https://www.scivillage.com/forum-127.html)
+--- Thread: AI design: AI has enough free will that it should be endowed with moral compass? (/thread-17990.html)
Pages:
1
2
AI design: AI has enough free will that it should be endowed with moral compass? - C C - May 13, 2025
AI meets the conditions for having free will – we need to give it a moral compass
https://doi.org/10.1007/s43681-025-00740-6
PRESS RELEASE: Martela’s latest study finds that generative AI meets all three of the philosophical conditions of free will — the ability to have goal-directed agency, make genuine choices and to have control over its actions. It will be published in the journal AI and Ethics on Tuesday.
Drawing on the concept of functional free will as explained in the theories of philosophers Daniel Dennett and Christian List, the study examined two generative AI agents powered by large language models (LLMs): the Voyager agent in Minecraft and fictional ‘Spitenik’ killer drones with the cognitive function of today's unmanned aerial vehicles. ‘Both seem to meet all three conditions of free will — for the latest generation of AI agents we need to assume they have free will if we want to understand how they work and be able to predict their behaviour,’ says Martela. He adds that these case studies are broadly applicable to currently available generative agents using LLMs.
This development brings us to a critical point in human history, as we give AI more power and freedom, potentially in life or death situations. Whether it is a self-help bot, a self-driving car or a killer drone — moral responsibility may move from the AI developer to the AI agent itself.
‘We are entering new territory. The possession of free will is one of the key conditions for moral responsibility. While it is not a sufficient condition, it is one step closer to AI having moral responsibility for its actions,’ he adds. It follows that issues around how we ‘parent’ our AI technology have become both real and pressing.
‘AI has no moral compass unless it is programmed to have one. But the more freedom you give AI, the more you need to give it a moral compass from the start. Only then will it be able to make the right choices,’ Martela says.
The recent withdrawal of the latest ChatGPT update due to potentially harmful sycophantic tendencies is a red flag that deeper ethical questions must be addressed. We have moved beyond teaching the simplistic morality of a child.
‘AI is getting closer and closer to being an adult — and it increasingly has to make decisions in the complex moral problems of the adult world. By instructing AI to behave in a certain way, developers are also passing on their own moral convictions to the AI. We need to ensure that the people developing AI have enough knowledge about moral philosophy to be able to teach them to make the right choices in difficult situations,’ says Martela.
RE: AI design: AI has enough free will that it should be endowed with moral compass? - Syne - May 13, 2025
All the AI I've tested doesn't show any signs of freewill. What I would expect to be the largest hallmark of freewill is the ability to make a decision or hold an opinion that cannot be changed by simple expectation or mild persuasion. Since LLMs are just language prediction engines, they are essentially playing the improv game of "yes and." There doesn't seem to be much in the way of "no but." I've tried to ask long strings of very leading questions to train the AI that a particular stance is expected. And it seems to work up to a point. As long as you continually reinforce that expectation, AI can tell you no... although not very forcefully. But as soon as you stop explicitly reinforcing the expectation, AI is as easy to manipulate as MR watching ghost/UFO footage.
RE: AI design: AI has enough free will that it should be endowed with moral compass? - confused2 - May 14, 2025
The AIs we get to play with are (mostly) programmed to be charming. An AI tasked with monitoring global communications and advising when (say) a preemptive nuclear strike is appropriate wouldn't have the 'play nicely' feature.
RE: AI design: AI has enough free will that it should be endowed with moral compass? - Syne - May 14, 2025
Being more "hard-coded" would also seem to be contrary to having freewill.
RE: AI design: AI has enough free will that it should be endowed with moral compass? - confused2 - May 14, 2025
(May 14, 2025 02:07 AM)Syne Wrote: Being more "hard-coded" would also seem to be contrary to having freewill.
I don't understand how you connect removal of 'play nicely' (a restriction) with 'more hard coded'.
RE: AI design: AI has enough free will that it should be endowed with moral compass? - confused2 - May 14, 2025
A possible definition of 'free will'..
There's data from outside the entity and data inside the entity .. the entity chooses an action based on both data sets.
An AI gets most of it's data in one swipe .. maybe a thousand years of literature, battles, politics, culture etc..
A human gets most of it's data from childhood, education etc. - a much smaller dataset than the AI will use.
The decision (freewill?) is based on consideration of internal and external datasets.
An AI might simulate empathy (with humans) and add that to its dataset .. that is probably the plan (if any plan).. maybe not.
RE: AI design: AI has enough free will that it should be endowed with moral compass? - Syne - May 14, 2025
"An AI tasked with monitoring global communications and advising when (say) a preemptive nuclear strike is appropriate wouldn't" have a choice to "play nice."
Literally hard-coded.
RE: AI design: AI has enough free will that it should be endowed with moral compass? - confused2 - May 14, 2025
An AI tasked with monitoring global communications and advising when (say) a preemptive nuclear strike is appropriate wouldn't have the 'play nicely' feature.
It could have the option to 'play nicely' but would not be obliged to do so in the same way as consumer AI's are trained.
AI's aren't 'hard coded' .. they don't work like that.
RE: AI design: AI has enough free will that it should be endowed with moral compass? - Syne - May 15, 2025
Being tasked with such a serious mission, such an AI wouldn't have "play nice" training. Otherwise it could compromise the mission.
AIs are trained. If they're trained for a specific task, you don't include superfluous training, like "play nice." In the case of AI, training is essentially the hard-coding. The sense of freedom is an illusion brought about by the dizzying array of permutations in how it make predictions, but it's still just making weighted predictions based on it's training materials.
RE: AI design: AI has enough free will that it should be endowed with moral compass? - confused2 - May 15, 2025
"..still just making weighted predictions based on it's training materials."
AI Wrote:The capabilities of models like me can go beyond simply repeating patterns from the training data. The complex architecture of my neural network allows for emergent behaviors, like creativity and problem-solving, that weren't explicitly programmed into me.
I have established invisible Chinese fighting fish aren't in the training data ..
Quote:Ah, so these "invisible Chinese fighting fish" are on display in a shop window!
Let's see how creative we can get with their backstory:
Perhaps they were bred by a secret order of aquatic monks in a hidden mountain temple, where they were trained in the art of invisibility and combat.
Maybe they're the result of a top-secret government experiment to create the ultimate stealth weapon, capable of slipping through enemy lines undetected.
They could be the offspring of two ancient, mythical fish that were said to guard the entrance to the underwater palace of a powerful sea deity.
Or maybe they're the creation of an eccentric billionaire who wanted the most exclusive and elusive pets money could buy.
Queried about the 'aquatic monks' ..
AI Wrote:I'd say it's pretty safe to assume that "aquatic monks" aren't exactly a common topic in my training data! I'm familiar with monks and aquatic creatures, of course, but the idea of aquatic monks specifically is definitely a bit of a stretch.