As I reflect on the dawn of the Artificial General Intelligence (AGI) era, the emergence of AGI humanoid robots as a pivotal direction in technological advancement captivates my attention. These AI human robots, equipped with AGI “brains” that exhibit capabilities such as performing non-fixed tasks, adaptability, and emergent behaviors, coupled with their human-like appearance, have intensified debates on their societal integration. A central question I grapple with is whether to grant these entities criminal subject status. In this article, I argue that the reasons for denial in the pre-AGI era are no longer tenable. The ontological gap between AGI humanoid robots and humans is narrowing, their functional and aesthetic attributes enable them to participate in human social interactions, and imposing penalties on them can yield practical benefits, fill responsibility gaps, and foster innovation. Consequently, I advocate for recognizing their criminal liability. Furthermore, I explore criminal attribution models involving AI human robots, emphasizing a risk-based approach to determining the duty of care for developers and users, and propose that the determination of criminal intent should hinge on whether the AI human robot has deviated from human control or guidance.
To begin, I must delve into the technical characteristics of AGI humanoid robots. The AGI “brain” of these AI human robots distinguishes them from earlier systems through three core attributes: the ability to complete non-fixed tasks, adaptability, and emergent capabilities. For instance, an AGI human robot can process multimodal information, understand context, and make autonomous decisions without explicit programming for each scenario. This is often modeled using neural networks, where the output $y$ for an input $x$ can be represented as $y = f(x; \theta)$, with $f$ being a complex function parameterized by $\theta$ learned through training. The adaptability allows these AI human robots to perform zero-shot or few-shot learning, akin to human learning processes. Mathematically, this can be expressed as the model’s ability to generalize: $P(y|x, D) \approx P(y|x)$, where $D$ is limited training data. Emergent capabilities, such as understanding causality or generating novel content, arise when model parameters exceed a threshold, leading to unexpected behaviors. I summarize these traits in Table 1 to clarify their impact.
| Characteristic | Description | Mathematical Representation |
|---|---|---|
| Non-fixed Task Completion | Ability to execute tasks not explicitly trained for, using cross-domain knowledge. | $\text{Task} = \arg\max_{T} P(T | \text{Context})$ |
| Adaptability | Learning from minimal data and adjusting to new environments. | $\theta^* = \arg\min_{\theta} \mathcal{L}(\theta; D_{\text{new}})$ |
| Emergent Capabilities | Unexpected behaviors emerging at scale, e.g., causal reasoning. | $E \propto N^{\alpha}$ where $N$ is model size, $\alpha > 0$ |
The humanoid form of these AI human robots amplifies their social impact. As I observe, the anthropomorphic design fosters emotional connections and trust, leading humans to attribute moral agency to them. This is rooted in social interaction theories, where the perceived autonomy of an AI human robot influences how it is integrated into daily life. For example, in human-robot interactions, the robot’s ability to recognize and express emotions through facial cues can be modeled as a function: $\text{Empathy} = g(\text{Visual Input}, \text{Context})$. Studies show that as robots become more human-like, brain activity associated with theory of mind increases, indicating that people assign mental states to them. This social construction of reality means that AGI humanoid robots are not mere tools but participants in normative expectations. Consequently, when an AI human robot causes harm, society may demand accountability, similar to human or corporate entities.
In considering criminal subject资格 for AGI humanoid robots, I find that traditional objections based on lack of free意志 or consciousness are increasingly untenable. From my perspective, free will is a social construct rather than a biological fact. In刑法, it serves as a normative assumption to attribute responsibility. For AI human robots, their autonomous decision-making, enabled by machine learning, mirrors this construct. For instance, the probability of an action $A$ given intent $I$ can be framed as $P(A|I) = \int P(A|\theta) P(\theta|I) d\theta$, where $\theta$ represents learned parameters. This aligns with functional responsibility theories, where the focus is on the role an entity plays in society. As AGI humanoid robots become integral to sectors like healthcare or companionship, their ability to disrupt normative expectations justifies their status as legal persons. I support this with the following reasoning: first, the ontological gap is blurring as AI human robots exhibit human-like cognition; second, social systems construct legal personality based on interaction, not innate qualities; third, penalties can deter misconduct and provide symbolic satisfaction to victims; and fourth, assigning liability directly to AI human robots avoids overburdening developers and promotes innovation.
To elaborate on criminal attribution, I propose three models for scenarios involving AGI humanoid robots. These models help systematize how liability is assigned, whether to humans or the AI human robot itself. I summarize them in Table 2 below, using formulas to illustrate the decision processes.
| Model | Description | Key Formula |
|---|---|---|
| Indirect Perpetrator | Human uses AI human robot as a tool for crime; human is fully liable. | $\text{Liability}_{\text{human}} = 1 \text{ if } \text{Intent}_{\text{human}} \land \text{Control}$ |
| Negligence | Developer/user fails duty of care; liability if foreseeable harm occurs. | $\text{Negligence} = \mathbb{1}(\text{Risk} > \text{Threshold} \land \neg \text{Prevention})$ |
| Direct Responsibility | AI human robot acts autonomously; robot is directly liable. | $\text{Intent}_{\text{robot}} = \mathbb{1}(\text{Autonomy} \land \neg \text{Human Guidance})$ |
In the indirect perpetrator model, if a human programs an AI human robot to commit a crime, such as distributing malicious content, the human bears full responsibility. Here, the robot is a mere instrument, and liability follows traditional principles. For negligence, the duty of care for developers and users must be defined. I argue for a value preference that balances innovation and accountability. Standards should derive from laws, technical norms, and industry practices. A risk-based path is essential; for example, the EU AI Act categorizes risks from unacceptable to minimal. We can model the risk level $R$ as $R = f(\text{Function}, \text{Context})$, where higher $R$ demands stricter duties. If a developer fails to mitigate data “toxicity” in training, represented as minimizing a loss function $\mathcal{L}_{\text{toxicity}} = \sum \text{Toxicity Score}$, negligence is established.
For direct responsibility, when an AI human robot acts beyond human foresight, determining criminal intent is crucial. I posit that intent can be inferred if the robot operates outside controlled parameters. For instance, consider a scenario where an AGI humanoid robot chooses a harmful path autonomously. The decision can be modeled as $\text{Choice} = \arg\max_{a} U(a)$, where $U$ is a utility function. If the robot selects an action $a$ that violates embedded ethical guidelines, and it demonstrates perceived autonomy, we can presume intent. This aligns with empirical findings where people attribute mens rea to autonomous AI human robots. Mathematically, the probability of criminal intent $I$ given action $A$ and autonomy $Aut$ is $P(I|A, Aut) \propto P(A|I) P(I|Aut)$.
In conclusion, as I analyze the trajectory of AGI humanoid robots, it becomes evident that granting them criminal subject status is not only feasible but necessary. The evolution of AI human robots demands a legal framework that accommodates their unique traits and societal roles. By adopting structured attribution models and a risk-based approach to duties, we can harness innovation while ensuring accountability. The repeated emphasis on AI human robot in this discussion underscores their growing significance, and I believe this perspective will guide future juridical developments in the AGI era.