Full chapter text
Responsibility
Artificial intelligence is the newest tool for computation, language,
graphics, decision-making and knowledge organization. But because
machines generate answers, moral responsibility does not automatically
shift to machines. The basic question is what kind of data The Prejudice
of Someone Who is the target? If a mistake is made, the person
responsible for What areas of human decision making are not capable of
full automation? Parallel philosophy holds that technical ability is not a
moral imperative.
63.1 Resources and Agents and Responsibilities
Artificial intelligence systems can produce complex decisions; however,
the structure of legal, institutional and ethical responsibilities is
determined by humans and institutions. The organization cannot end its
responsibility by saying that the algorithm made the decision. What data
is selected, what goals are set, what risks are accepted, and what human
reviews are kept are all human decisions.
63.2 Not a data neutral
The training data comes from the history of the society. If there are
discrepancies in history, the data can take their toll. But it 's also easy to
mistake all the statistical differences for bias in the data. The question is
whether the data is representative. The classification is a reasonable one.
The goal is to be morally right. Making a mistake can cause more damage.
How durable are the system 's results in independent testing?
63.3 Interpretation and appeal
Where artificial intelligence decisions affect jobs, credit, education,
health, justice or public benefit, affected persons should have the
opportunity to understand the reasons, correct errors and seek human
review. This is not a public enough reason for the model to say so.
Technical complexity is not a substitute for accountability.
63.4 Limitations of the scope of automation
Some functions may be suitable for routine computation, sorting, or
searching. But where decisions demand dignity, consent, punishment,
life-or-death, complex social contexts or moral conscience, full
automation demands special care. Efficiency is a value. fairness,
compassion, transparency and appeal are also values.
63.5 Artificial Intelligence and the Knowledge
Linguistic systems will give impressive answers, but impressive language
is not proof of truth. They can make up sources, twist events, or state
uncertain facts in a certain way. AI-generated content, therefore, needs
to be examined with independent evidence as a supporting proposition
rather than a source, particularly in history, law, medicine, science and
public policy.
63.6 Creativity and Authorship Responsibilities
Artificial intelligence will help with writing, drawing, music and
translation. But the writer 's responsibility does not end there. The
person or organization publishing the work will be responsible for the
facts, citations, rights, consents, and consequences. The means change or
there is a question of responsibility.
63.7 Language Inequality and the AI
Massive digital content in major languages can make the model more
efficient when Maithili-speaking languages are underrepresented due to
limited resources. It is not just a technical problem. it is a question of
linguistic justice. Local corpus building requires open licenses, quality
editing, diverse scripting support, and community involvement.
However, confidentiality and consent are also important in data
collection.
63.8 The Dignity of the Human Being
The criterion of this text is clear: the dignity and privacy of human
beings, autonomy and responsibility will be the criterion above technical
efficiency. Systems that are highly effective but reduce individuals to
mere statistics need ethical review.
Pūrvapakṣa (Prima Facie View): If AI makes better decisions than humans, why aren 't
human reviews unnecessary?
Uttarapakṣa (Response): On what criteria is the best choice? Average accuracy is
important, but rights, appeal, rare circumstances, price-trigger, and liability are
separate questions. High performance is not absolute proof of moral legitimacy.
63.14 Automatic Classification and Appeal
The system puts the person into a risk, ability, or priority category. If
there was a degree of probability, the authorities would have accepted
that as a definite fact. Misclassification can have serious consequences
for creditworthiness or independence.
The affected person will know what the main reasons are for using
automated tools, how to improve the data and where to place a human
appeal. Trade secrets are not absolute grounds for concealment of
substantive judicial reasons.
The success of the appeal must be measured. The process is symbolic if
only the email address is there, but no right to change the decision.
63.15 The Illusion of Giving the Human Form
By looking at the fluid language or faces, the user can perceive in the
system a sense of desire and compassion. Behavior being human-like and
having conscious experiences are different claims. Consciousness is not
automatically proved by the present demonstration.
Humanizing can lead to emotional dependence, sharing private
information, and false beliefs. The system must explain its limitations,
non-human nature, and risks.
Philosophically, the possibility of machine-awareness may remain an
open question, but policy is built on existing evidence and actual
capabilities, not fiction.
63.16 Efficiency and Humiliation
Automation can reduce wait times, but the lack of human hearing in the
context of bereavement, health or justice can be humiliating. Service is
not only about results, but also about behavior.
It is expensive and slow to make humans do all the work. High-impact,
easy-to-manage routines in case of unusual situations and disputes;
general tasks may be automated.
To correct errors in the calculation of efficiency, excluded users and trust
costs are added. Time savings are not the only value.
63.17 Precedent: The more capable system will be liable itself
The idea is to blame the agent if the system does the planning, language,
and optimization. But accountability requires the ability to reason, obey
rules, and act responsibly.
Legal personality may be a technicality; moral consciousness is not. The
responsibility of the manufacturer, the implementer, the owner and the
decision maker should not disappear in the name of the machine.
The question should be re-examined if the future ability changes. It is
now safer to specify the institutional responsibility.
63.9 Decision-Aidability and Decision-Right
Artificial intelligence can show patterns or give predictions based on
selection patterns. But the human decision maker will know what data
and error-value the recommendation is aimed at. The speed of the
instrument does not transfer responsibility.
In high-risk areas, the final decision demands meaningful human review
and not just a seal. The reviewer must have the time, authority, and
ability to change the system.
63.10 Multiple sources of bias
Bias can arise from chronic discrimination within the data, sample-
disparity, goal-selecting, labeling, measurement, or application context.
It is not enough just to state the data clearly.
Group-wise, the outcome of the experience and decisions affected by the
error should be examined. Absolute equality can sometimes conflict with
different measures; moral preferences must be publicly decided.
63.11 Interpretability and the reasons why
It is important to state what characteristics caused the model results,
especially in areas such as debt, employment, or health. General
technical details are not sufficient for the appeal of affected persons.
The explanation must be judgmental, understandable, and counter-
intuitive. It is risky to make policies based on predictive correlation.
Interpretability is not a guarantee of veracity; it is a condition of liability.
63.12 Labor and Creation and Crediting
Behind artificial intelligence systems are data-gathering, labeling,
writing, diagrams, and computational labor. The word automatic can
make this human contribution invisible.
Creative use raises questions of source-rights, consent, credit, and
livelihood. Limitations of innovation and unfair imitation should be clear
from the methods, techniques, and ethical dialogues; market forces alone
should not decide.
63.13 Maithili and Representation
Languages with less digital content may have more system errors and
dialects outside the standard form may be invisible. Incorrect translation
can cause actual harm to administration or health.
Community-accepted data, linguistic reviews, dialect-diversity and
public testing are necessary. The language-collection agreement and
benefit-sharing must be clear. Cultural content is not a free resource in
the name of digital presence.
Chapter Conclusion
Artificial intelligence is neither god nor devil. it is powerful human
technological system. Their ethical values depend on the use, targeting,
control of data, transparency and social consequences. The formula of
parallel philosophy finds special meaning in an age when technical
ability is no longer a moral imperative.
Chapter Bibliography
Luciano Floridi et al. are selected AI ethics articles.
Shannon Vallor — Technology and the Virtues.
Virginia Eubanks — Automating Inequality.
Cathy O’Neil — Weapons of Math Destruction.
UNESCO — Recommendation on the Ethics of Artificial Intelligence.