The Value of Externality to Intelligence

As mentioned earlier, intelligence is the ability to adapt to the external environment. During the training of an intelligent agent, the input values are derived from the external environment, and the output results interact with the environment to receive feedback. This feedback from the external environment drives parameter adjustments of neurons (individuals) within the intelligent agent (the whole). In other words, the external environment is the source of feedback for the intelligent agent and the context in which its intelligence is manifested. If an intelligent agent loses feedback and input from the external environment, the significance of intelligence ceases to exist.

In the construction of artificial intelligence programs, external information is crucial for training. Internal information is merely a manifestation of the program’s operation, and the operation itself does not provide comparative results. Without comparative results, there is no feedback, and thus, the artificial intelligence cannot be trained. Before constructing an intelligent agent, it is necessary to define the boundary between the intelligent agent and its external environment to determine what constitutes the external environment and external information.

Different external environments not only manifest as different input values for the intelligent agent but also as specific feedback states for its output values. Therefore, different external environments and their corresponding feedback affect the internal feedback adjustment mechanisms of the intelligent agent (different parameter sets, different weighting schemes). If the environment faced by an intelligent agent during training is highly limited, exhibiting significant non-randomness or specificity, the network structure of the intelligent agent will also be highly limited. This means the intelligent agent will struggle to produce effective outputs (low adaptability) when facing external conditions significantly different from those it has adapted to. Therefore, the more realistic and open the external information input (i.e., the higher the similarity between training information and the real external environment), the more versatile and applicable the output schemes of the trained intelligent agent will be, resulting in greater and more effective adaptability to the environment.

The similarity between the external environment used for training and the characteristics of the broader real external environment is referred to as the intelligent agent’s externality manifestation. The higher the externality, the more realistic, abundant, and open the external information the intelligent agent can obtain (with lower limitations or biases). Thus, higher externality makes the intelligence process more effective, with the adaptability manifested by the intelligent agent being more versatile and effective in real-world environments.

However, different intelligent network organizations have different goals. Some intelligent agents only need to operate in specific environments. For example, a particular animal species living on an isolated Pacific island only needs external information input from that specific environment. In this scenario, although the external information is relatively closed, it is indeed generated by the species’ real living environment. Additionally, more open data input is unnecessary for the species’ current survival and may even reduce its adaptability to this specific, closed environment. While it cannot be ruled out that the species might die out quickly if invasive species arrive or if it is removed from this specific environment, in a static environment, localized environmental adaptability is the optimal solution. Therefore, the characteristics of external information for an intelligent agent need to be tailored to the organization’s goals.