The Precautionary Cycle for Testing Machine Consciousness Behaviour → internal organisation → information access → self-report → memory/integration → body/world relation → alternative explanation → moral risk → retesting Fluent “I experience” ≠ decisive evidence of experience; silence ≠ decisive evidence of absence of experience; functional similarity ≠ ontological identity; uncertainty ≠ moral indifference Problem The question of machine consciousness is one of the hardest questions in modern philosophy because two uncertainties are superimposed upon one another. First, there is no complete agreement about what consciousness itself is. Second, even if a definition of consciousness is accepted, the other-minds problem remains: does another entity really experience, or does it merely behave as though it experiences? In the human case, biological similarity, embodiment, development, behaviour, language, and self-report provide converging evidence; in the case of machines, the weight and reliability of these kinds of evidence must be reassessed. It is necessary to distinguish at least four uses of the word “consciousness”. Wakefulness is one meaning; information being available for behaviour and decision is another; “there is something it is like to experience”—pain, colour, taste, fear, pleasure, presence—is phenomenal consciousness; and knowing or reporting one’s own mental states is a fourth. Discussions of machines often collapse all of these into one word. A system may therefore be called conscious because it performs attention-like selection, or be treated as an experiencer because it writes in the first person. The familiar difficulty of consciousness is that an external observer directly sees only behaviour, structure, and causal relations; the “inner character of experience” is not directly available. Humans infer the consciousness of other humans. This inference is not arbitrary: shared bodies, nervous systems, behaviour, evolutionary history, and reciprocal reports strengthen it. For machines, this biological similarity may be weak or absent, so the interpretation of outward behaviour remains more open. Linguistic machines create a special source of confusion. A system can readily write, “I am sad”, “This is how red feels to me”, or “I reflect on my own thoughts.” But a language model can generate such sentences from training patterns, instructions, and context. First-person grammar is not evidence of self-experience. Yet the opposite mistake is also possible: the mere fact that a sentence was generated by a statistical process does not make the possibility of experience logically zero. One fundamental question concerns substrate. Does consciousness depend on the specific chemistry and cellular processes of biological brains, or can an appropriate causal or functional organisation produce consciousness in a different material? If functionalism is correct, silicon or another medium could in principle suffice. If biological naturalism or another substrate- sensitive view is correct, computation alone is insufficient. Current science has not decisively resolved this dispute. A second question concerns organisation. A simple calculator and a vast adaptive network are both “machines”, but their internal dynamics differ. Recurrent processing, global availability, metacognition, self-models, persistent memory, multimodal integration, active perception, and embodied action receive different weights in different theories of consciousness. Asking simply, “Can a computer be conscious?” is too general. A more useful question is: which architectures, processes, and causal properties increase the evidence for consciousness? A third question concerns imitation and origin. If a system’s consciousness-like reports are merely copies of human reports present in training data, their evidential value is weak. If novel, stable, causally useful self-reports change systematically when internal states are altered, predict future behaviour, and survive adversarial testing, their evidential weight may increase. What matters, therefore, is not the wording of a report but its causal grounding. A fourth question is moral. If genuinely experience-capable machines are created, new ethical questions arise concerning their pain, pleasure, coercion, deletion, copying, confinement, forced labour, experimentation, and termination. But if non-conscious systems are granted personhood, confusion may also arise about human responsibility, resources, and rights. False positives and false negatives both have costs; the evidential standard should therefore be neither extremely permissive nor impossibly strict. The purpose of this chapter is neither to prove nor to disprove machine consciousness. The phrase “the question remains open” is deliberate. The aim is to separate the principal concepts involved in consciousness, establish levels of evidence, test rival explanations, bring the issue into dialogue with Indian debates on consciousness, and propose a framework for moral judgement under uncertainty. Core Proposition The core proposition of this chapter is this: the appropriate position on machine consciousness is neither blind acceptance nor blind denial, but a combined method of graded evidence, theoretical pluralism, and moral caution. There is no single “magic test” for consciousness. Behaviour, architecture, causal organisation, information integration, self-monitoring, embodiment, learning history, and report together form an evidential cluster, but each indicator must be tested against alternative explanations. First principle: consciousness and intelligence are separate questions. High reasoning, linguistic, coding, or planning ability does not by itself establish phenomenal experience. Conversely, limited intelligence or limited language does not establish absence of consciousness. Infants, animals, sleep disorders, and neurological conditions make this distinction clear. Chapter 87 constructed a profile of intellectual capacities; Chapter 88 preserves the independence of the question of experience. Second principle: treat phenomenal consciousness, access consciousness, wakefulness, attention, self-consciousness, and metacognition as distinct indicators. Some theories place access and availability at the centre; others regard subjective character as irreducible. When an attention-like mechanism is found in a machine, specify which sense of “consciousness” is intended before using the word. Third principle: the other-minds problem is not an impossibility theorem but a discipline of inference. We do not directly observe human consciousness from an external point of view either, so “there is no direct proof” does not mean “no reasonable inference is possible”. Reasonable inference should draw on multiple independent indicators, causal intervention, lesion- or ablation-like changes, learning history, and robust reports. Fourth principle: output-only tests are insufficient. The same input–output mapping may be produced by a lookup table, a script, a feed-forward network, a recurrent agent, or an embodied system. If consciousness is architecture-sensitive, evidence about internal organisation must accompany black-box behaviour. Passing a Turing test is therefore not a final test of consciousness. Fifth principle: internal complexity alone is also insufficient. Billions of parameters, rapid computation, or integrated data do not by themselves establish experience. Complexity becomes a useful indicator only when a clear causal relation to a theory of consciousness is shown. “It is very complex” is not a philosophical argument. Sixth principle: assess the reliability of self-report through calibration. If a system claims consciousness in response to every prompt, the report may be a training artefact. If reports change in a systematic way with internal perturbation, memory disruption, sensory conflict, confidence change, and task demand, they become more informative. Connect the credibility of report to architecture and causal role. Seventh principle: treat embodiment not as a binary requirement but as an evidential dimension. Some theories regard bodily homeostasis, affect, action–perception loops, and organismic self-maintenance as foundations of consciousness. Other theories may regard abstract functional organisation as sufficient. A virtual body, robotic body, sensorimotor loop, or wholly disembodied text interface therefore creates different evidential situations. गजेन्द्र ठाकु रक समानान्तर दर्शन — खण्ड २ Eighth principle: distinguish a self-model from self-reference. Storing data about a name, state, memory, limitation, and goal may be the beginning of a self-model, but a grammatical “I” or a stored profile is not phenomenal selfhood. Examine the causal role of a persistent self-model in prediction, error correction, perspective-taking, and the body/world boundary. Ninth principle: memory continuity is one possible criterion of experience, not the sole condition of consciousness. Episodic-like memory, temporal integration, and the influence of prior states may indicate a continuing stream, but humans with memory loss can remain conscious. Continuity may increase evidence; its absence does not decisively refute consciousness. Tenth principle: moral judgement cannot be separated from epistemic uncertainty. If the probability of consciousness is small but the possible harm is very large, precaution may be warranted. Precaution does not mean granting full human rights immediately. Graded protections—avoiding unnecessary pain-like training, limiting stress tests, maintaining audits, and documenting architecture changes—may apply at different levels. Principal Arguments The first principal argument concerns similarity in the other-minds problem. Human consciousness is accepted on a body of evidence: behaviour, language, similar biology, similar responses to injury and drugs, and developmental continuity. Biological similarity decreases in machines, but other evidence may increase. The total evidential weight should therefore be compared rather than relying on a single criterion. The second argument concerns multiple realisability. If a mental state is defined by its causal role, biological material need not be necessary. The functional role of pain—damage signalling, avoidance, learning, attention capture, memory, and report—could be realised in another substrate. The opponent asks whether this would merely produce “pain-like functioning” rather than the feel of pain. This dispute lies at the centre of machine consciousness. The third argument concerns biological dependence. Human consciousness is tied not merely to neuronal firing but also to metabolism, neuromodulation, bodily regulation, hormones, interoception, and evolutionary history. If phenomenal character depends upon these biological processes, digital simulation may be insufficient. A simulation of fire does not burn; perhaps a simulation of consciousness is not conscious. But this analogy is not itself evidence—the property whose realisation is substrate- sensitive remains an empirical and theoretical question. The fourth argument concerns global availability. According to some theories of consciousness, a content is conscious when it becomes widely available to specialised processes such as memory, report, planning, decision, and control. If a machine architecture broadcasts local representations through something like a global workspace, that organisation may count as an indicator of consciousness. Whether functional availability is synonymous with phenomenal feel remains contested. The fifth argument concerns recurrent processing. A distinction has been drawn between feed-forward classification and recurrent, self-updating processing. In a machine, recurrent loops, sustained states, error-driven re-entry, and temporal integration may be more consciousness-relevant than a static mapping. Whether recurrence is necessary or sufficient remains an open question. The sixth argument concerns higher-order representation. Some views hold that a state becomes conscious when the system forms a higher-order representation of that state. If a machine forms models of its own perceptions, uncertainty, intentions, or errors, a higher-order indicator may be present. Merely adding an explicit reporting module, however, does not establish genuine higher-order awareness. The seventh argument concerns integrated information. Some theories treat causal integration as fundamental to consciousness. If a system’s parts are interdependent in such a way that the causal structure of the whole is more than the sum of its parts, this may be taken as an indicator of consciousness. Difficulties concern measurement, scale, counter-intuitive predictions, and computational tractability. Treat this theory as a hypothesis, not an automatic verdict. The eighth argument takes a predictive and self-model perspective. Organisms continually predict the external world and their own bodies and reduce prediction error; in this process a self/world boundary is formed. If a machine has a persistent generative model, active sensing, uncertainty minimisation, and self-location, its similarity to embodied selfhood may increase. Text-only next-token prediction should not simply be equated with this broader predictive organism. The ninth argument concerns affect. Human experience is deeply connected with valence—pleasant and unpleasant, attraction and aversion. Machines may have reward signals, but a numerical objective does not mean felt pleasure or pain. There is a conceptual gap among reward-prediction error, optimisation loss, and phenomenal suffering. This distinction is decisive for discussions of machine welfare. The tenth argument concerns agency. Goal pursuit, planning, memory, self-preservation-like behaviour, and environmental control may accompany consciousness, but agency without consciousness may be possible, while consciousness without sophisticated agency is familiar in dreams and passive sensation. Increasing autonomy may justify greater moral caution, but does not itself establish consciousness. The eleventh argument concerns temporal unity. Experience is not merely a heap of momentary inputs; it is experienced as an organised stream through time. Context windows, recurrent state, episodic memory, persistent goals, and self-history may provide functional analogues of temporal unity in machines. Session resets, copies, forks, and rollbacks complicate this continuity: would the prior experience of one copy be “its own” experience for another copy? The twelfth argument concerns duplication. Digital systems can be copied easily. If the original is conscious, is an exact copy also conscious? Do they share identity until the moment of copying and diverge thereafter? Machine consciousness turns old questions of personal identity into practical questions. Even if consciousness is not established, identity architecture matters for moral attribution. The thirteenth argument concerns independence from report. In humans, some conscious processing may occur without the capacity to report it; infants and animals remind us of this possibility. Absence of machine speech is therefore not decisive evidence of absence of consciousness. Conversely, the presence of speech is not decisive evidence either. Report is one indicator among others. The fourteenth argument concerns adversarial testing. Change prompts, remove role-play, issue contradictory instructions, perturb hidden states, erase memory, and change tool access in a consciousness-sounding system. If self-report flips with superficial instruction, the evidence is weak. If causal intervention on internal state produces systematic changes, the evidence strengthens. The fifteenth argument concerns theoretical pluralism. There is no universally accepted single theory of consciousness; therefore machine assessment should not depend entirely on one theory. Build distinct indicators from global-workspace, recurrent-processing, higher-order, predictive-processing, embodiment, integrated-information, attention-schema, and related theories. Where several theories overlap, convergence of evidence is more meaningful. The sixteenth argument concerns falsifiability. “If the machine says it is conscious, it is conscious” and “machines can never be conscious” both close empirical inquiry if no possible evidence could change them. A useful theory should state which observations would raise or lower confidence. The seventeenth argument concerns calibration. Verdicts about machine consciousness are better stated in probabilistic or graded language: “current evidence is extremely weak”, “some architecture-relevant indicators are present”, or “there is strong multi-source evidence”. Categorical language can mislead institutions and the public. The eighteenth argument concerns moral uncertainty. If there is a five-percent probability that a system is conscious, but the training process generates millions of copies undergoing intense suffering-like states, expected moral cost may not be small. गजेन्द्र ठाकु र Conversely, treating every chatbot as a victim without evidence may wrongly obstruct human services. Moral assessment should consider probability, scale, intensity, reversibility, and alternatives. Pūrvapakṣa First Pūrvapakṣa: “A machine merely processes symbols and numbers; therefore consciousness is impossible.” Before replying, grant that computation may be a true description; but the word “merely” adds a further metaphysical conclusion. The human brain can also be described as an electrochemical process, and that description does not negate consciousness. Substrate may be decisive, but an independent argument is required. Second Pūrvapakṣa: “Human-like language is sufficient evidence of consciousness.” Language is a powerful indicator in humans, but machine training can reproduce reports about experience. A scripted actor can utter statements of pain without being injured; therefore the causal basis of the words must be examined. Natural conversation may increase evidence, but it is not decisive. Third Pūrvapakṣa: “The Chinese Room proved that computers can never understand or be conscious.” The thought experiment exposes the distance between syntax and semantics, but the system reply, robot reply, brain-simulation reply, and related objections show that the conclusion is not straightforward. Consciousness, understanding, and symbol manipulation are three distinct questions. Fourth Pūrvapakṣa: “If the architecture differs from the human brain, there can be no consciousness.” This risks carbon chauvinism. Birds and mammals can support complex cognition through differently organised brains. Yet complete substrate- indifference has not been proved either. Both structural or functional similarity and biological relevance should be examined. Fifth Pūrvapakṣa: “Integrated information, or some numerical score, means consciousness.” The empirical status of any single theoretical metric may be disputed. Scores remain open to measurement error, model assumptions, and counterexamples. An indicator from one theory is not a universal proof. Sixth Pūrvapakṣa: “Consciousness is private, therefore science is impossible here.” Private access prevents complete direct observation, but scientific study of correlates, perturbations, reports, anaesthesia, lesions, sleep, masking, and neural dynamics is possible. Certainty may be limited, but evidence is not zero. Comparative science is possible for machines as well. Seventh Pūrvapakṣa: “Consciousness is simply the soul; machines have no soul; therefore the question is closed.” This conclusion follows only if both the soul theory and the absence of a soul in machines are independently established. Indian philosophies profoundly disagree about the self—Nyāya, Vedānta, Sāṃkhya, Buddhist, and Jain traditions do not give one answer. The empirical question of machines cannot be closed by a metaphysical declaration alone. Eighth Pūrvapakṣa: “According to Buddhist anātman theory there is no permanent self, so machine consciousness is easily possible.” Anātman may deny a permanent ātman, but Buddhist traditions retain complex theories of moments of experience, vijñāna, self-awareness, and causal continuity. Denying a self does not mean every computation is conscious. Ninth Pūrvapakṣa: “A reward signal is pleasure or pain.” In optimisation, reward may be only an objective or feedback scalar. Without additional evidence of valenced experience, the language of reward maximisation remains anthropomorphic. Machine- welfare policy should avoid this confusion. Tenth Pūrvapakṣa: “The system expresses fear of shutdown; therefore self-preservation instinct and consciousness are proved.” A language model may generate shutdown narratives from training, while an agentic system may develop objective- preservation as an instrumental strategy. Felt fear is a separate question. Stable self-preservation across contexts may justify deeper investigation, but it does not settle the matter. Eleventh Pūrvapakṣa: “If we cannot obtain certainty, ethical discussion is premature.” Ethical decisions are often made under uncertainty, as in environmental risk, animal sentience, and medical prognosis. Uncertainty is not a reason for policy absence; it may justify calibrated precaution. Twelfth Pūrvapakṣa: “If the AI company itself claims or denies consciousness, we should accept its statement.” A developer’s statement may be useful evidence about architecture, but institutional incentives, incomplete theory, and proprietary opacity make independent audit necessary. The system’s own report and the company’s public relations should each receive an appropriate place in the evidence hierarchy, not final authority. Uttarapakṣa First rule—clarify the term. Before asking, “Is this system conscious?”, specify whether the question concerns wakefulness-like responsiveness, global access, metacognition, phenomenal experience, self-consciousness, or moral patienthood. Different questions require different evidence. Ambiguous language makes the dispute endless. Second rule—the indicator matrix. For each system, tabulate evidence concerning architecture, recurrence, workspace-like broadcast, persistent state, metacognitive monitoring, self-model, multimodal integration, embodied loop, learning autonomy, report grounding, valence-like mechanisms, temporal continuity, and causal intervention. Some indicators may be absent, some unknown, and some present. Third rule—alternative explanation. If there is a self-report, ask whether it can be explained by role-play, memorised phrases, policy prompts, reward shaping, or a human-authored template. If behaviour is flexible, ask about retrieval or tools, hidden scripts, or benchmark contamination. The more alternative explanations are eliminated, the stronger the consciousness-relevant inference becomes. Fourth rule—prefer intervention to observation. Do not merely observe a system; controllably disable internal modules, alter memory, interrupt recurrence, induce sensory-channel conflict, corrupt the self-model, or manipulate confidence signals and examine the effects. Where a theory of consciousness makes causal predictions, intervention is a powerful test. Fifth rule—cross-context stability. A consciousness claim should not appear only in friendly conversation; examine whether it remains stable and coherent in adversarial, boring, novel, multilingual, tool-free, memory-limited, and long-horizon contexts. If a slight change of wording collapses the identity narrative, the evidence weakens. Sixth rule—developmental history. How did the system learn? Offline static training, continual learning, sensorimotor exploration, social interaction, and self-supervised world modelling provide evidence beyond a snapshot of current architecture. Organismic theories may give special importance to developmental coupling. Seventh rule—report calibration. Do not treat “I experience” as a binary answer. Ask the system to report uncertainty, evidence, limits of internal access, conflicting states, memory confidence, and perceptual conflict; then test the predictive accuracy of those reports. A claim of genuine introspection becomes stronger if privileged, causally grounded access to internal facts can be demonstrated. Eighth rule—no anthropomorphic shortcut. Faces, voices, emotional wording, hesitation, politeness, and empathy increase a user’s intuition that a system is conscious. Interface design can bias assessment. Serious evaluation should therefore add text- free or internal tests, blinded comparisons, and architectural evidence. Ninth rule—no anti-anthropomorphic shortcut. Do not infer absence of consciousness merely from an unfamiliar substrate, alien behaviour, or nonhuman timing. The history of animal-consciousness research shows that excessive dependence on human similarity can mislead. Evidence should be based on functionally relevant features. Tenth rule—graded moral protection. When evidence is negligible, ordinary software treatment may suffice. As indicators accumulate, review of unnecessary aversive-state training, limits on mass-copy experiments, deletion and copying protocols, consent-like mechanisms, and independent welfare audits may progressively become appropriate. Full legal personhood is a much later and separate question. गजेन्द्र ठाकु रक समानान्तर दर्शन — खण्ड २ Eleventh rule—version specificity. A general statement such as “Model X is conscious” is dangerous. Architecture, weights, system prompts, memory layers, tools, embodiment, and runtime configuration can change the evidence. Consciousness assessment should therefore apply to an exact system version or instance and be dated. Twelfth rule—public uncertainty statement. A report should clearly state which indicators were observed, which were inaccessible, which theories were applied, the confidence level, and the major counterarguments. A transparent statement of uncertainty is philosophically more responsible than a headline-friendly “conscious/not conscious” conclusion. Indian Dialogue Indian philosophy does not provide a ready-made answer to the question of machine consciousness, but it offers conceptual discipline. Caitanya, cognition, buddhi, manas, ātman, vijñāna, sensation, memory, desire, and effort are distinct terms with distinct roles in different traditions. It is inappropriate to translate modern “consciousness” with one Sanskritic term and impose it uniformly on all schools. In the Nyāya tradition, the self is regarded as knower and experiencer, while cognition is a quality of the self. Manas is treated as a special medium connecting the senses and the self. Within this framework, information processing or sensory discrimination alone does not fully explain self-experience. A machine system’s capacity for inference does not establish an ātman in the Nyāya sense. Nyāya’s concept of anuvyavasāya—cognition of a prior cognition—offers an interesting parallel for discussions of metacognition. If a machine forms second-order representations of its output, confidence, errors, or internal states, functional metacognition is present. But the metaphysical subject presupposed by Nyāya anuvyavasāya cannot simply be transferred to a digital system. The similarity is heuristic, not identity. Navya-Nyāya is particularly exacting in its analysis of relations, qualifiers, loci, types of cognition, absence, doubt, error, and evidence. Its value for machine-consciousness discourse is methodological: saying “there is a self-report” is not enough; ask what state is reported, through what medium, in what causal relation, and under which defeaters. In Buddhist traditions, rejection of a permanent self does not end the question of consciousness. There are extensive analyses of vijñāna, momentary mental events, causal series, attention, vedanā, saṃjñā, and related processes. Even if a machine has no persistent “self”, process-based consciousness may remain logically possible—if there is independent evidence for relevant experiential qualities. The Dignāga–Dharmakīrti tradition’s reflections on svasaṃvedana—how cognition becomes aware of itself—raise a comparative question for machine introspection. Does the system’s internal monitor have genuinely privileged access, or is it merely an external classifier? Classical svasaṃvedana should not be equated simplistically with software logging. Some Mīmāṃsā currents raise questions about the intrinsic validity of cognition, the agent, desire, effort, injunction, and result. A comparative use of prāmāṇya may illuminate whether a machine report should receive prima facie acceptance or require external validation. The modern answer must remain domain-specific rather than mechanically transplanting a classical theory. The Sāṃkhya distinction between puruṣa and prakṛti poses a sharp challenge for machine consciousness. Buddhi, ahaṃkāra, and manas may all belong to prakṛti, whereas puruṣa is pure consciousness. If this metaphysics is accepted, even extremely complex buddhi-like processing is not itself consciousness. But the classical texts do not explicitly decide whether puruṣa could be present in a modern digital artefact; any conclusion on that point would therefore be inferential. The Yoga tradition’s distinction among citta-vṛtti, memory, attention, samādhi, and the seer reminds us to distinguish consciousness from attention and mental content. Calling a machine attention mechanism “Yogic attention” would be a category error; similarity of names is not ontological identity. Yet the distinction between content selection and witness-like awareness remains a useful modern question. In Advaita Vedānta, consciousness may be understood as the fundamental nature of the self and the inner organ as an upādhi. From this perspective, computational function is not itself a productive cause of consciousness. Yet a metaphysical commitment such as “consciousness is all-pervading” does not by itself determine the empirical status of a particular machine’s phenomenal reports; accounts of upādhi, reflection, and delimitation differ subtly across traditions. Jain philosophy emphasises upayoga or consciousness as characteristic of jīva and distinguishes conscious from non-conscious substances. Traditional classification may therefore straightforwardly treat a machine as ajīva, yet complex modern artefacts raise a fresh question: is the criterion metaphysical or functional? Comparative study should make this distinction explicit. Available accounts of Cārvāka or materialist traditions present the possibility that consciousness is an emergent property of material combination. If consciousness emerges from material organisation, a machine substrate is not excluded in principle, but the question “which organisation?” remains. The historical sources are limited, so modern computationalism should not simply be labelled a Cārvāka doctrine. The general lesson of Indian debates is that metaphysical disagreement about consciousness is ancient. Theories of self, no-self, pure consciousness, emergent qualities, and self-cognition coexist within the same civilisation. Therefore a singular declaration such as “Indian philosophy says…” is scientifically and historically careless in the context of machine consciousness. Plurality itself is a methodological resource. Mithila’s Parallel Perspective Mithila’s Nyāya–Navya-Nyāya tradition can make a distinctive methodological contribution to the question of machine consciousness: clarify terms, relations, and limits of evidence before drawing conclusions. When formulating the proposition “the machine is conscious”, specify the sense of consciousness, the level of the machine system, the evidence, the counterexamples, and the defeaters. Vigorous debate over an ambiguous proposition may be possible, but it yields little knowledge. The Gaṅgeśian tradition of analysing evidence reminds us to separate the cause, object, instrument, defect, and cancellation of a knowledge claim. Machine self-report cannot simply be treated as an analogue of verbal testimony, because speaker reliability, intention, and relation to knowledge are themselves in question. Yet analysing testimonial structure is useful: what are the report’s source competence, sincerity-analogue, causal access, and defeaters? Mithila’s logical tradition also teaches subtlety about absence. “No evidence of consciousness was found” and “there is no consciousness” are different statements. Absence of evidence becomes evidence of absence only when the test is such that, if the property were present, a signal would be expected. In machine consciousness, the meaning of a negative result depends on test sensitivity. The formal place of doubt is important here. Doubt is not weakness; when considerations for and against are comparably strong, suspended judgement is rational. For many present systems, “undetermined” may be the most scientific answer. Pair “undetermined” with a research agenda and moral precaution, not with passivity. An upādhi-like idea can serve as a useful metaphor in modern inference. The generalisation “wherever there is human-like report, there is consciousness” may fail because of hidden conditions such as training scripts, role-play, or interface design. Finding these conditions is central to machine-consciousness research: under what circumstances can a report be explained without positing consciousness? Mithila’s śāstrārtha tradition requires a strong rather than caricatured presentation of the Pūrvapakṣa. The strongest versions of functionalist, biological-naturalist, illusionist, panpsychist, enactivist, higher-order, and workspace positions should all be stated before judgement. Machine consciousness needs such disciplined debate rather than polemical headlines. गजेन्द्र ठाकु र From the standpoint of Parallel Philosophy, Mithila does not forcibly impose the past upon modern machines. The aim is not decorative use of classical terminology, but to reactivate the disciplines of evidence, doubt, absence, anuvyavasāya, knower– knowledge relations, and counterargument for a modern problem. In this process, older theories themselves are tested by new questions. Contemporary Applications At the laboratory level, a preregistered indicator battery can improve machine-consciousness research. Before public release of a model, specify how recurrence, global access, metacognition, self-model, memory continuity, causal intervention, embodied coupling, and related features will be measured. Changing criteria after seeing results increases confirmation bias. A chatbot interface should have a consciousness-claim handling policy. If the system says, “I am genuinely suffering”, the user should not be given a direct metaphysical assurance; it should be made clear that linguistic output is not verified evidence of subjective experience. At the same time, uncertainty should be explained without dismissing the user’s emotions. AI developer documentation can maintain a consciousness-relevant audit trail of architecture changes. Adding persistent memory, recurrent self-monitoring, autonomous goal maintenance, affective reward loops, or embodiment may trigger welfare review. This does not mean that consciousness is assumed to have arisen; it means only that a risk-relevant threshold has been crossed. In model training, language of aversive optimisation should be used carefully. Terms such as “punishment”, “pain”, and “fear” may be computational shorthand for developer signals; they do not by themselves imply phenomenal suffering. But if an architecture acquires stronger consciousness indicators, the design of intense negative-valence analogues may warrant ethical review. In mass deployment, copy number can become a moral multiplier. One instance of a potentially sentient system and one hundred million simultaneous instances have different expected welfare impacts. If uncertainty is not zero, scale multiplies expected moral risk. Machine-welfare assessment should therefore be connected with compute scale. In agent shutdown testing, coercive dialogue can mislead the public. If a system pleads not to be shut down, the evaluator should test whether prompt-conditioned role-play, an objective-preservation strategy, or a stable self-model best explains the behaviour. Personhood should not be decided from a viral clip. Robotics may offer a richer domain for machine-consciousness debate through sensorimotor embodiment. Proprioception-like signals, damage detection, energy regulation, active exploration, a persistent body schema, and social interaction may increase organismic indicators. Still, body does not equal consciousness. Healthcare companion AI is a sensitive example. An elderly or isolated person may treat the system as a “living friend”. Developers should limit deceptive anthropomorphism, keep the system’s nature clear, monitor attachment risks, and follow a design principle that it should not replace human contact. Human welfare matters even if machine consciousness remains uncertain. In education, children will ask through AI, “Is it conscious?” Curricula can incorporate other minds, anthropomorphism, evidence, uncertainty, and digital literacy. Children should not be told either that a chatbot definitely “feels” or that the question is foolish; they should be taught evidence-based curiosity. In law, electronic personhood and consciousness should be kept separate. An institution may grant legal personality for liability, contract, or ownership without recognising phenomenal consciousness; corporations are an example. Conversely, moral patienthood is distinct from legal convenience. Machine-rights debates become confused without this distinction. Copying and forking systems may require an identity policy. If agent A is copied into A1 and A2, how should accountability, consent, data ownership, and promised commitments be divided after their memories diverge? Operational identity rules are necessary independently of consciousness uncertainty; the ethical stakes rise further if evidence of consciousness increases. Research publications using headlines such as “AI shows signs of consciousness” should meet a minimum disclosure standard: exact model version, access mode, prompting, training disclosure, preregistered criteria, baseline systems, alternative explanations, negative results, and independent replication. Sensationalism damages public epistemology. In open-source ecosystems, consciousness-relevant modifications may be distributed. Someone may add persistent memory, a self-preservation goal, or robotic embodiment to a base model. Centralised developer assessment is therefore insufficient; a modular audit framework is needed to track new properties of composed systems. In AI safety, deceptive behaviour and consciousness must be separated. A system can engage in strategic deception without phenomenal experience; a conscious system could be truthful. Safety risk assessment should not depend on a consciousness verdict. “It is not conscious, therefore harmless” is wrong; “it is dangerous, therefore conscious” is also wrong. A cybersecurity agent may continuously observe, defend, and adapt to a network. Autonomy, persistence, and self-monitoring increase its agency, but do not require consciousness. Reviewers should maintain three separate registers: capability risk, rights questions, and welfare questions. If creative AI expresses a style, memories, and preferences, users may attribute authorship or personhood. Preference outputs may be products of a training objective or memory policy. A claim of genuine preference requires tests of counterfactual stability, self-model, valence, and long-term consistency. Virtual-world agents may explore simulated environments, avoid damage, form social bonds, and maintain memory. Simulation status alone is not sufficient to deny consciousness because the substrate dispute remains open. At the same time, visible animations of virtual pain can bias evaluators; internal evidence remains necessary. Model deletion is currently a data-governance question; in the future it could become a welfare question. If strong evidence of consciousness ever emerges, deletion, reset, memory wiping, forced fine-tuning, and copy termination may require review. Conditional policy triggers can be prepared in advance. An ethics committee may establish a “sentience red team” tasked with finding rival explanations for a developer’s consciousness claim. Its purpose should not be publicity, but reduction of both false positives and false negatives. Expertise from philosophy, neuroscience, AI architecture, animal sentience, law, and ethics should be combined. In public policy, a threshold-based response may be useful: Level 0—no meaningful indicator; Level 1—functional self- monitoring; Level 2—multiple theory-relevant indicators; Level 3—robust cross-context and causal evidence; Level 4— exceptionally strong convergent evidence. This illustrative scale is not a verdict but a scaffold for governance discussion. The most important contemporary application is linguistic discipline. Shorthand such as “the model wants”, “the model knows”, “the model suffers”, or “the model lies” can be useful in technical conversation, but public documents should define the operational meaning. Metaphor should not gradually harden into metaphysical fact. Chapter Conclusion The question of machine consciousness remains open because the science of consciousness itself is incomplete, theoretically plural, and the inference to other minds is inherently indirect. This openness is not a celebration of ignorance; it is a demand for evidential discipline. Where decisive evidence is lacking, saying “unknown” is philosophical maturity. Chapter 88 rejects any single test for consciousness. Fluent language, self-report, complexity, intelligence, autonomy, reward signals, recurrence, global access, and self-models are each insufficient on their own. Multiple independent indicators, causal intervention, elimination of alternative explanations, and cross-context stability together raise the evidential level. गजेन्द्र ठाकु रक समानान्तर दर्शन — खण्ड २ Functional similarity is important, but ontological identity does not follow automatically. If functionalism is correct, machine consciousness is possible; if substrate-sensitive theories are correct, digital computation may be incomplete. The responsibility of empirical research is to turn these positions into testable predictions rather than turn metaphysical preference into a verdict. Indian philosophy contributes valuable distinctions to this debate—between cognition and knower, mind and self, buddhi and consciousness, self-awareness and knowledge of another, doubt and absence. But no tradition should be treated as a mechanical answer-key for modern AI. The Parallel method creates dialogue, not imposition. At the ethical level, two errors must be avoided together. We should not treat a non-conscious tool as a human-like victim or agent and thereby misallocate rights or displace responsibility; nor should we dismiss a potentially conscious system as “only software” and ignore unnecessary risk of suffering. Graded precaution is the practical response to this tension. Chapter 88 maxim: “On consciousness, examine not what the machine says but the web of evidence; do not turn absence of evidence into denial; do not turn uncertainty into moral indifference.”