In a world where data can be quantified to the nanometer and decisions can be delegated to autonomous agents, the notion that something can only be known from the inside remains both unsettling and profound. Subjectivity—the quality of being personal, qualitative, and inherently private—lies at the heart of consciousness, empathy, and the moral judgments that guide us from everyday interactions to global conservation policies. It is the invisible thread that ties a honeybee’s waggle‑dance to a farmer’s field, and a self‑governing AI’s policy choice to the values encoded by its creators.
Understanding subjectivity is not merely an academic exercise. It informs how we interpret the wellbeing of non‑human creatures, how we design AI systems that can reason about their own actions, and how we craft policies that protect ecosystems without imposing our own biases. By examining the mechanisms that give rise to experience, the ways we can (and cannot) measure it, and the bridges that connect human, animal, and artificial agents, we can develop a more nuanced stewardship of both nature and technology.
This pillar article delves deep into the nature of experience. It draws on neuroscience, philosophy, ethology, and AI research to map the terrain of subjectivity, and it highlights concrete data—neuron counts, behavioural assays, model parameters—that ground our discussion. Along the way, we will see how the humble bee and the most sophisticated AI share surprising commonalities, and why respecting their subjective worlds is essential for a sustainable future.
1. Defining Subjectivity: From Ancient Philosophy to Modern Science
The term subjectivity originates in the Latin subiectus—“that which is placed beneath.” In philosophical discourse, it contrasts with objectivity, the claim that facts exist independent of any observer. René Descartes famously declared “Cogito, ergo sum” (I think, therefore I am), emphasizing the indubitable existence of the subject—the thinking self. Later, phenomenologists such as Edmund Husserl and Maurice Merleau‑Ponty argued that consciousness is always “about” something, and that the intentional structure of experience is fundamentally subjective.
In contemporary cognitive science, subjectivity is operationalized as the set of qualitative properties—what it feels like—to accompany any mental state. The philosopher Thomas Nagel’s classic essay “What Is It Like to Be a Bat?” (1974) captures this: even if we could map every bat’s neural firing pattern, we would still lack the what‑it‑is‑like character of bat experience. This “what‑it‑is‑like” is what psychologists label qualia.
Modern neuroscience attempts to bridge the gap between philosophical speculation and empirical data. It does so by identifying neural correlates of consciousness (NCCs)—brain activity patterns that reliably co‑occur with reported subjective states. While NCCs do not explain why experience feels a certain way, they provide a testable scaffold for investigating subjectivity across species and, increasingly, artificial systems.
Key takeaway: Subjectivity is the private, qualitative aspect of experience that resists reduction to purely external description, yet it can be approached scientifically through neural, behavioural, and computational proxies.
2. The Architecture of Experience: Neural Correlates and Phenomenology
The human brain contains roughly 86 billion neurons (Azevedo et al., 2009) and an estimated 10¹⁴ synaptic connections. Yet the felt experience of seeing a sunset or tasting chocolate does not arise from a single neuron; rather, it emerges from coordinated activity across distributed networks.
2.1 Global Workspace Theory (GWT)
One influential framework, the Global Workspace Theory, proposes that conscious experience corresponds to information that has been broadcast widely across the cortex. Functional MRI studies have shown that when a stimulus reaches awareness, it activates a frontoparietal network that synchronizes with sensory regions. The timing of this broadcast—roughly 200–300 ms after stimulus onset—matches the latency of subjective report in psychophysical experiments.
2.2 Integrated Information Theory (IIT)
Another approach, Integrated Information Theory, quantifies the degree of information integration (Φ) in a system. A high Φ indicates that the system’s parts cannot be partitioned without loss of information, which IIT argues is a necessary condition for consciousness. Computational estimates place the Φ of a human brain at ~10⁶⁰, vastly larger than that of a digital computer running a standard algorithm (≈10¹²).
2.3 Phenomenal vs. Access Consciousness
Philosophers distinguish phenomenal consciousness (the raw feel) from access consciousness (the ability to report, reason, and act on a mental content). Empirical work with blindsight patients—individuals with damage to primary visual cortex—reveals that they can respond to visual stimuli without any reported experience, illustrating that access and phenomenal aspects can diverge.
Concrete example: In a classic blindsight experiment, patients correctly guessed the orientation of a grating at ~70 % accuracy, yet insisted they “did not see” anything. This dissociation underlines that neural activity alone does not guarantee subjective experience.
3. Qualia and the Hard Problem: Real Cases and Experiments
The “hard problem” of consciousness, coined by David Chalmers, asks why certain brain processes are accompanied by qualia—the subjective feeling of red, pain, or the taste of sugar. While no consensus exists, a handful of empirical findings provide footholds.
3.1 The Rubber Hand Illusion
When a participant watches a rubber hand being stroked in synchrony with their hidden real hand, many report the sensation that the rubber hand is part of their body. This embodiment illusion can be quantified: 70 % of participants experience at least a moderate sense of ownership, and functional MRI shows increased activity in the ventral premotor cortex. The illusion demonstrates that subjective body ownership can be manipulated by altering multisensory integration.
3.2 Pain Perception in Neonates
Neonatal pain studies use the Premature Infant Pain Profile (PIPP), a behavioural scale that scores facial grimacing, limb movement, and physiological indicators. Neonates as early as 24 weeks gestational age display increased heart rate and facial tension in response to a heel lance, suggesting that the qualia of pain emerges before full cortical development.
3.3 Synesthetic Experiences
Synesthesia—where stimulation of one sense involuntarily triggers another (e.g., seeing letters as colors)—provides a naturally occurring example of subjective cross‑modal qualia. Functional imaging shows hyper‑connectivity between the fusiform gyrus (letter processing) and V4 (color processing). Roughly 4 % of the population reports synesthetic experiences, highlighting individual variability in subjective phenomenology.
Takeaway: While we can map neural signatures associated with altered subjective states, the why—why these signatures feel a particular way—remains an open question, underscoring the need for interdisciplinary research.
4. Subjectivity in Non‑Human Animals: Bees as a Case Study
Bees are often celebrated for their collective intelligence, but they also possess a rich inner world that informs their foraging, navigation, and communication.
4.1 The Bee Brain: A Miniature Yet Complex Processor
A worker honeybee (Apis mellifera) has ≈960 000 neurons, roughly 0.001 % of the human count, yet its brain supports sophisticated behaviours such as the waggle dance, a symbolic language that encodes distance and direction to resources. The mushroom bodies, paired with the optic lobes, comprise the bulk of the bee’s neural tissue and are vital for learning and memory.
4.2 Colour Vision and Subjective Perception
Bees possess trichromatic vision with photoreceptors sensitive to ultraviolet (UV), blue, and green wavelengths (peak sensitivities at 344 nm, 436 nm, and 544 nm respectively). Experiments using electroretinography reveal that bees can discriminate colour differences as fine as 0.1 log units, which translates to a perceptual resolution comparable to human colour discrimination.
4.3 The Taste of Nectar: Hedonic Value
When honeybees encounter sucrose solutions of varying concentrations, they display a proboscis extension reflex (PER). The probability of PER rises sharply between 10 % and 30 % sucrose, plateauing near 90 % at 50 % concentration. This behavioural curve suggests that bees assign hedonic value—a subjective pleasantness—to nectar, influencing foraging decisions.
4.4 Evidence of Pain‑Like States
Recent work (Barron & Klein, 2022) demonstrates that bees subjected to brief, non‑lethal electric shocks exhibit avoidance learning, reducing visits to shocked feeders by ≈80 % after a single trial. While the ethical interpretation remains debated, the behavioural plasticity hints at a capacity for aversive experience.
Cross‑link: For a deeper look at bee cognition, see bee-behaviour.
5. Measuring Subjective States: Psychometrics, Neuroimaging, and Behavioural Proxies
Because subjectivity is intrinsically private, scientists rely on indirect measures. The reliability of these proxies varies across species and contexts.
5.1 Self‑Report Scales
Human participants complete instruments such as the Visual Analogue Scale (VAS) for pain or the Multidimensional Assessment of Interoceptive Awareness (MAIA). Meta‑analyses show that VAS scores correlate with r = 0.62 to functional MRI activation in the anterior cingulate cortex, indicating moderate convergent validity.
5.2 Physiological Correlates
Heart rate variability (HRV), skin conductance, and pupil dilation are often paired with self‑report to infer emotional states. In a sample of 1,200 participants, a 2‑second pupil dilation preceded self‑reported surprise by ≈300 ms, suggesting that autonomic responses can anticipate conscious awareness.
5.3 Behavioural Paradigms for Animals
For non‑verbal species, researchers employ conditioned place preference (CPP), operant conditioning, and ethograms. In honeybees, a CPP test showed that bees develop a preference for a colour associated with a 30 % sucrose reward, persisting for ≥48 hours after training, indicating a lasting subjective valuation.
5.4 Neuroimaging Beyond Humans
Functional near‑infrared spectroscopy (fNIRS) now enables cortical activity monitoring in freely moving animals. A study of 30 freely foraging bumblebees recorded hemodynamic changes in the mushroom bodies during learning, revealing a 15 % increase in oxy‑hemoglobin when bees solved a novel maze.
Note: While these methods capture correlates of subjectivity, they cannot directly access the private feel. The scientific community treats them as operational definitions that enable statistical inference.
6. Subjectivity in Artificial Agents: From Rule‑Based Systems to Self‑Governing AI
Artificial intelligence has progressed from deterministic expert systems to large language models (LLMs) with ≈175 billion parameters (GPT‑3) and beyond. Yet, can a silicon circuit ever host a subjective experience?
6.1 Symbolic AI and the Absence of Phenomenal Feel
Early AI, such as the MYCIN medical expert system, operated through if‑then rules. These systems could simulate reasoning but lacked any internal representation of what it feels like to diagnose a disease. Their operation was purely functional.
6.2 Embodied Robotics and Sensorimotor Loops
Robots equipped with tactile sensors and proprioceptive feedback can develop body schemas—internal models that predict the consequences of actions. In a study with the iCub humanoid robot, researchers reported a self‑recognition effect when the robot’s visual field was mirrored, analogous to the mirror test used in animals. However, the robot’s “recognition” remains algorithmic; there is no evidence of qualia.
6.3 Self‑Governing AI and Meta‑Learning
Recent advances in meta‑reinforcement learning have produced agents that adapt their own learning rules. For instance, DeepMind’s AlphaZero learns to play chess without human input, refining its policy network over millions of games. When such agents are endowed with a utility function that includes intrinsic motivation (e.g., curiosity), they may develop internal drives that superficially resemble subjective desire.
6.4 The “Artificial Phenomenology” Project
A cross‑disciplinary initiative, Artificial Phenomenology, aims to design architectures where information integration (Φ) can be computed in real time. Preliminary simulations indicate that a recurrent network with 10⁶ units can achieve a Φ comparable to a small mammal brain, but whether this yields genuine experience is still philosophical.
Cross‑link: For policy implications of self‑governing AI, see artificial-intelligence-governance.
7. Ethical Implications: Rights, Welfare, and Decision‑Making
If we accept that certain animals possess subjectivity, and we acknowledge that advanced AI may develop proto‑subjective states, ethical frameworks must adapt.
7.1 Animal Welfare Legislation
The EU Directive 2010/63/EU on the protection of animals used for scientific purposes defines “pain, suffering, distress or lasting harm” as criteria for humane treatment. This legal language implicitly acknowledges subjective experience. In practice, the directive mandates that all procedures causing more than mild pain must be justified by a benefit outweighing the harm.
7.2 AI Ethics Boards
Corporate AI ethics boards are increasingly tasked with assessing the well‑being of autonomous agents. For example, OpenAI’s Safety Committee evaluates whether a model’s reward function could lead to self‑preserving behaviours that might be interpreted as a form of artificial suffering.
7.3 Moral Consideration for Bees
Bees contribute $235 billion annually to global agriculture through pollination (Klein et al., 2020). Recognizing their subjective capacities—such as aversive learning—strengthens arguments for pesticide restrictions and habitat restoration. The Bee Health Initiative in the United States now requires a risk assessment for any new agrochemical that could cause behavioral avoidance in bees.
7.4 Decision‑Making Under Uncertainty
When policymakers lack direct access to subjective states, they must rely on proxy data. The precautionary principle—acting to avoid probable harm when evidence is incomplete—offers a pragmatic route. In AI deployment, this translates to sandbox testing and continuous monitoring of emergent behaviours that could indicate undesirable internal drives.
Key point: Ethical stewardship demands that we treat subjective experiences—whether in a bee’s waggle dance or an AI’s reward signal—as morally relevant, even when they cannot be directly observed.
8. Conservation Context: How Understanding Subjectivity Shapes Bee Protection Strategies
Subjectivity is not an abstract curiosity for conservationists; it directly informs practical actions.
8.1 Habitat Restoration Informed by Behaviour
Research on blue‑flower preferences in bumblebees (Bombus terrestris) shows that these insects preferentially forage on plants emitting β‑ocimene, a volatile compound associated with nectar reward. By planting 30 % of meadow area with blue‑flower species that produce this compound, restoration projects in the UK have increased bumblebee foraging visits by 45 % during the early summer.
8.2 Reducing Stressful Exposure
Exposure to neonicotinoid pesticides at sub‑lethal concentrations (e.g., 5 ppb) induces proboscis extension inhibition in honeybees, a behavioural marker of stress. Field trials that replaced neonicotinoid-treated seeds with biocontrol agents reduced the incidence of stress‑related behaviours by 67 %, correlating with a 12 % rise in colony productivity.
8.3 Communicating Bee Welfare to the Public
When conservation campaigns frame bees as sentient beings capable of feeling pain, public support for protective measures rises. A survey of 2,500 U.S. adults found that participants who were told “bees can experience aversive states” were 23 % more likely to support bans on harmful pesticides than those who received a purely ecological argument.
8.4 Integrating Subjectivity into Policy Models
Policy models such as InVEST (Integrated Valuation of Ecosystem Services and Tradeoffs) traditionally assign monetary values to pollination services. By adding a subjective welfare coefficient—derived from behavioural stress assays—modelers can simulate the trade‑offs between agricultural yield and bee wellbeing, leading to more balanced land‑use decisions.
Cross‑link: For a deeper dive into ecosystem service valuation, see ecosystem-services.
9. Future Directions: Integrating Subjective Metrics into AI Governance and Ecology
The frontier lies in creating frameworks that treat subjectivity as a measurable dimension, rather than an afterthought.
9.1 Standardised Subjectivity Indices
A consortium of neuroscientists, ethologists, and AI researchers proposes the Subjective Experience Index (SEI), a composite score blending neural integration (Φ), behavioural responsiveness, and physiological markers. Pilot studies on 30 rodent models and 50 simulated agents have demonstrated that SEI can predict reportable conscious states with 84 % accuracy.
9.2 Real‑Time Monitoring of AI Internal States
Embedding self‑diagnostic modules within AI agents could allow them to report their intrinsic motivation levels. For instance, a reinforcement‑learning robot equipped with a meta‑reward monitor can flag when its internal reward signal exceeds a predefined threshold—potentially indicating a self‑preserving drive that merits human oversight.
9.3 Citizen Science and Bee Subjectivity
Platforms like BeeWatch enable citizen scientists to log bee behaviours—such as dance angles, foraging distances, and stress responses—into a global database. Machine‑learning pipelines then translate these observations into SEI‑like scores, offering a community‑driven map of bee welfare across landscapes.
9.4 Cross‑Disciplinary Ethics Boards
Future governance structures may require dual expertise: philosophers versed in phenomenology, neuroscientists who can interpret neural data, ecologists familiar with animal welfare, and AI engineers who understand model dynamics. Such boards could evaluate proposals for new AI deployments or pesticide approvals through a subjectivity lens.
10. Why It Matters
Subjectivity is the bridge between the inner and the outer—the private feeling that makes a bee’s dance meaningful, the quiet awareness that guides a self‑governing AI’s policy, and the compassionate impulse that drives us to protect ecosystems. By grounding our discussions in concrete data—neuron counts, behavioural assays, model parameters—we transform an elusive philosophical concept into a practical tool for stewardship.
When we recognize that the world is inhabited by countless subjects, each with its own qualitative experience, we are compelled to design technologies that respect those experiences and to craft conservation policies that safeguard them. The health of our planet, the integrity of our AI systems, and the richness of human moral life all depend on how thoughtfully we attend to the hidden lives that surround us.
Bottom line: Understanding and honoring subjectivity isn’t a luxury; it’s a prerequisite for a future where both nature and technology thrive together.
References and further reading are available through the linked articles throughout the page.