Agreed - the integration (and tuning of that integration) might be the most difficult piece. From my armchair it seems like just plugging a bunch of things together, but the devil is in the details.
The coffee cup is weird. given the apparent dexterity of the machine, there doesn’t seem to be any obvious reason why it couldn’t handle a cup of liquid - it has the capacity to grip it; it has the finesse of operation not to slosh it around; it’s possible they tried to get it to do that and didn’t yet succeed, but had some sort of obligation to publish a demo at a specific milestone date and time; it’s possible they hyperfocused on the handling of the coffee pod and didn’t think of the cup part, or didn’t think it was interesting, or something like that. Absence of an element in a demonstration might mean incapability, but it probably doesn’t always, automatically mean that.
This is where I think a multi-layered approach might help. A model that can make a high-level decision that a hedgehog needs rescuing from a pond, might not be all that good at co-ordinating the actions of doing the thing, but it doesn’t have to be the same model doing all of the functions; it could be (much as it is with humans) different, specialised subsystem-models handling the stuff they are good at; an additional model that is very good at handling objects, but not so great at understanding why it’s been told to do that, might provide the physical skill to augment the capability of the decision-making part.
I mean that absolulely was what was demonstrated, its not misrepresenting it at all. The robot was presented with a very carefully selected set of bits of trash and bins in a very clean, largely empty, room. Then given a single phrase that caused it to put the trash in the bins. That is lightyears away from anything remotely like “autonomy”.
Not that the LLM side is not impressive (assuming it actually used a LLM to decode that sentence into instructions, I’ve been around to startups enough to see that that could 100% be faked in that demo). Not too long ago that would have been thought the difficult part of that problem, and the “physical intelligence” of actually seeing and manipulating the environment the (relatively) easy part.
Yes, I believe the two words are synonyms when describing a person. But when we speak of an independent or autonomous person, that can mean a number of things:
a person with a relatively independent nature or character
a person who does not take a side in a particular conflict
a person who does not depend too much on any particular person (or people) to function in society
I had the latter sense of autonomous in mind, or something very near to it. I believe when @Mangetout wrote ‘not a slave’, he had the same idea. If you go back to the 5 criteria in the OP, they are not enough to rule out a slave.
The infant that would starve to death without a parent is not autonomous. Yet a famine does not strip a person of autonomy - the function of a person in society is sensitive to societal factors such as famine.
The hypothetical machine that stares at grass until its battery goes critical, which necessitates otherwise avoidable trespass and energy theft, does not meet my criteria for an autonomous person. In the same way, I do not consider autonomous a human person who stares at grass until he becomes so hungry he must resort to trespass and theft of food.
I was sort of with you until the last paragraph. A human who chooses to stare at grass whilst starving to death is (I would say) autonomous. A person who stares at grass whilst starving, because they are unable to stop, is not.
The difficulty in mapping that definition to a robot will of course be the question of what ‘choice’ really means, but I think that may be a red herring given that the concept of true free will in humans is argued by philosophers and scientists.
I consider the ability to make a choice part of personhood, not autonomy. Infants make choices, for example.
Functioning in society without being overly dependent is the key to autonomy, IMHO. Staring at grass until you starve - without an extraordinary rationale (hunger strike?) - just doesn’t cut it.
Well, it’s not so much about judging rationality as it is about whether this particular self-destructive behavior is considered “functioning in society”.
I’m not even sure I agree that functioning in society is critical to autonomy. If someone decides to go and live the life of an off grid hermit in the mountains, is that person behaving autonomously?
That’s a good point. Is he able to meet his basic needs? Or is he starving himself to death staring at grass or chasing hedgehogs?
I’m tempted to say he doesn’t get to change the standard just by moving some distance from other people. I think if he isn’t self sufficient, he hasn’t left society. And if he is self sufficient, he is undoubtably autonomous. A society of one, so to speak.
Also where is failure in this? If he is able to meet his own needs for three years then one day he trips and breaks his ankle and he chooses to set it himself, but eventually dies of wound complications, did he stop being autonomous?
Autonomy is self-determination. I don’t find anything in the definition that speaks about effectiveness or success.
Autonomy is self governance, if I am not mistaken. I think one can fail at government in more ways than one can fail at determination; determination either is or is not, while governance can be good, bad, effective, ineffective, successful, or unsuccessful.
There is no question that a dead person cannot be an autonomous person. A dead person isn’t a person, strictly speaking he was a person. If our hermit was self sufficient, then ceased to be, unintentionally, that is the moment he ceased to be autonomous. Whether that moment occurred when he was injured, or at some point during the complications, or at the moment of death, would depend on the standard of medical care he considered sufficient.
I feel like if I had simply used the term ‘self-governing’ in place of ‘autonomous’, this part of the discussion might never have happened, even though those two things mean the same. I think you might be inadvertently including ‘and completely successful at it’ as a hidden criterion in the notion of autonomy. I don’t think it is necessary and I didn’t anticipate any sort of infallibility in the scenario.
We must approach the creation of self-aware slave robots with caution, ensuring that penalties for building or owning such entities are comparable to the serious offense of enslaving humans.
However, a non-sentient robot should not be classified as a slave any more than any other inanimate tool. The real challenge lies in recognizing the emergence of consciousness within AI and preventing the development of sentient ‘slave robots.’
Regarding safety, strict standards must be implemented. For instance, domestic robots could be restricted to a height of no more than 3 feet, with break-away limbs for emergency shut down. If one goes berserk, just kick him and he’ll fall to pieces. Additionally, programming should limit the proximity of these robots to prevent more than two from operating within 25 feet of each other, avoiding the risk of a collective malfunction. The last thing I need is for a pack of pissed off tiny robots chasing after me.
While the concept of self-aware (e.g. sapient) robots is fascinating, their production should be tightly controlled and regulated. Our planet already hosts an over-abundance of intelligent life forms, and the current biodiversity crisis (Holocene extinction) suggests we should be cautious about introducing more. Robots may indeed have a smaller ecological footprint, but our planet doesn’t need millions more technological beings doing damage. On the other hand, if we could replace humans with self-conscious robots, maybe we’d be in better shape.
As for which AI company should lead in building autonomous robots, Boston Dynamics stands out with their impressive track record in robotics innovation. I’m less clear on which AI company should lead the way.
I don’t think an autonomous person has to be completely successful. You brought up an edge case - a person living in complete isolation - in which I admitted, if he cannot take care of himself and dies, he isn’t autonomous any more.
Let me reframe my main argument.
Imagine a human person with the following qualities:
has all 5 senses
normal perception
healthy body
normal motor skills
normal ability to decide to do things
is a slave/servant
Per your 5 criteria this person is autonomous. Per your definition of autonomous, this person is not autonomous because he is a slave/servant. You have contradicted yourself.
Now add my 6th criteria (which I’ll tweak a little, again):
not overly dependent on or under the control of any particular person(s) to function in society
and the person no longer qualifies as autonomous, resolving the contradiction.
Not exactly. I specified that the being would have intent, agency and volition - the ability to decide things. A slave may have the capacity for those things, but agency is the condition of being able to use them.
Honestly, I think I know what I meant; I think you know what I meant. I am not sure what purpose is served by this semantic hair-splitting.
I’m afraid I don’t. I already told you that I personally consider intent, agency, and volition (collectively, the ability to make decisions) to be criterion for personhood. From my perspective a slave has all three, by definition.
We must have some deeper misunderstanding here, for instance, I wouldn’t distinguish between ability and capacity, and I’m reading “condition of being able to” as “ability to”. Rendering your paragraph into this nonsensical mess:
I specified that the being would have the ability to decide things. A slave may have the ability to decide things, but agency is the ability to decide things.