Good question! If by AI you refer to a learning system, then the analogy is close, though not exact since the AI system may lack consciousness as does a human child. Following that analogy, once a cybernetic humanoid in AI form reaches fruition of its learning development, given that (1) there is no intervention of its creator in its post-manufacture operation, and (2) the manufacturer/creator has not intentionally programmed irrationality (e.g., immorality) into the AI, it can be argued that AI will have become an agent of free choice, that is a volitional creature. From that point it is arguably the case that the AI is responsible for its own action just as a human upon attainment of adulthood is usually considered responsible for itself and not the parents, even in the case of substandard parenting that does not sink to intentional corruption of youth. However, this argument requires careful qualification as outlined briefly in what follows.

Here and in the course on AI (COMP 5210) that I developed and taught at Webster University (cited below), I used the qualifier "arguably" because there are difficulties with the analogy, as well as the constraints (1) and (2) above that I signified by "given that". Here I quote from my Lecture Two in that course:

"The main difficulty with assigning full responsibility to an AI system occurs in case the AI lacks consciousness of its actions. Considering that case, I have realized and propose that the creator can supply a substitute for consciousness by embedding morality algorithms, that is, property protection mechanisms, in a constitutional ruleset unmodifiable by the AI; i.e., an inaccessible part of its operating system that places constraints on its behavior. (This is analogous to the people of the United States placing limits on what its government may do by means of a Constitution, except for fact that the protection mechanisms of the AI is unconditionally insuperable whereas in the case of the US Constitution, the close and continual involvement of citizens acting as watchdogs is required.)

Accordingly, I propose this mechanism as a replacement for what has been called "Asimov's First Law of AI" which I consider a potential enabler of the AI to violate its rationality constraint by justifying the violation as necessary to protect its human creator(s). I further propose that the scientists who design and implement the AI do bear responsibility for 'good faith behavior' in programming an ethical system, given the assumptions I stated above. This is true because survival of humans, the species often called "homo sapiens", but less often demonstrated to be such, depends on their building AI systems that are superior to the humans themselves in consistently avoiding initiatory coercion (as distinct from purely defensive counter-coercion)."

I hope this helps. This excerpt from COMP 5210 may be cited thus:

You Learn First Russel Paradox, then you learn Searching AI technique, then you learn Sorting technique AI . Then you will be able to find out how the parent and child layer effect each other.

TO Adnan Majeed,

these above as you said, all are included in AI, just not including of parent-child paradox, this paradox was produced in my thinking about life evolving theory.

Thanks for your attentions.

I think that the burden of proof in the question is inverted for this case. We may state the question, "how many responsibilities may parents bear for the child" in the case of parent-child relationships based on an implicit independence assumption. On the other hand, scientist-AI relationship burden question is, "how many responsibilities may parents not bear for the AI". The reason being is that more than likely the courts and ethical debate will start from the assumption of the AI not being intelligent enough to be autonomous and shift the burden to the scientist.

This also frames the question on the boundary of ethics or law. More than likely it will not be taken as a clear cut exercise but grow out based on incremental developments. This points to that it will eventually grow out from product liability law into other concerns. The other concerns will have to wait until the AI independence can be established legally and scientifically (an unfair and unequal treatment since we do not question that in humans).

The key problem is laid on: a matured AI would be independent from human beings. Isn't the scientists who made of AI may be infinite resiponsibilities? Or the resiponsibilities may be borne by whole society?

Yaozhi Jiang ,

I fail how the problem is laid without a proper definition of independence. Independence for an AI can mean independence of a particular platform, independence of programming, independence via learning of patterns (e.g. deep learning. Which should be emphasized is not intelligence), independence due to a robotic body, independence because of lack of energy restrictions. While it might seem hair splitting, each of this can have a bearing on how independence is interpreted and how to evaluate legal implications for the manufacturer, users.

I also do not see the attribution to society. It would be like saying, 'what is your responsibility as part of society for building a gaming console' or 'what is the responsibility to you for genetic manipulation using crispr/cas9'. My point is that this question needs to be restated to provide more context for interpretation and evaluation.

To Arturo Giegel,

My viewpoint about meaning of "matured" is that from designing, manufacturing, maintenancing, to operating, even if also to managing, altogether of these above may be done by robotic system. The problem is laid on this technical level may be easy to understand?

Thanks for your attentions!

Yaozhi Jiang ,

You are asking about axiological repercussions of an entity based on attributes "independence" and "maturity" that are too broad.

For example, if I design a Prolog system that has rules of inference for a chatter bot, based on the concepts given of maturity and independence, I will say that such systems are mature for production (it is a 70's language that has produced multiple expert systems). Nonetheless, without additional constraints to satisfy, it will just fall into product liability.

Liability and axiological attribution for this type of problem depends on the particular of the sort of developmental stage of the AI (it is the case with a minor). The problem with an AI system is that of determining when that "of-age" is acquired. Development time in terms of technological maturity is not equivalent to natural age of a human being. Also, it will also depend on whether:

• The system learns (not in the deep learning/pattern recognition sense but true acquisition of complex concepts).
• Whether it can be proved that the system has emergent behavior (extremely difficult to nail down much less prove).
• Aware of its own actions(easy enough)
• Did its actions under no duress and of its own volition. This is extremely difficult since volition is too ambiguous a term and needs further elaboration. In addition, special systems to avoid programming tampering would have to be built-in. From a computer systems security perspective, it is extremely difficult to provide any guarantee.

I am sure we can add more constraints to the list.

I think that, if these issues are determined, expanded and adapted to determine "independence" and "maturity", I think it will be very productive in finding the boundaries that you asked about.

Since this is an incremental process of development the burden of untying the creator from the AI will be extremely difficult and again will start from product liability(it has already started through autonomous systems and will evolve to other systems until we arrive at true AI's).

As a final point to address, infinite responsibilities can be ruled out based on the concept of instantiation of an AI. It does not matter whether there is the potential for reproducing the AI, if it is not realized, then there is no realized responsibility.

to Arturo Geigel,

Yes , it is too broad.

make heuristic search

Yaozhi Jiang ,

I think the question has much merit and if elaborated with enough detail, it can be a great contribution.

Thanks for Arturo Geigle!