AI And Human Centered Perception of Time
Authors of this Survey Analysis: Aco Momcilovic — President of Global AI Ethics Institute; Kevin LaGrandeur — Research and Academic Affairs Director of Global AI Ethics Institute
The full document can be downloaded here: https://globalethics.ai/artificial-intelligence-and-time-measurement-time-experience/
If AI attains a state where it is conscious of time, and if it even sees it as relevant, it might depend on us and our coding as a basis for developing its own timekeeping methodology. So, to begin, it might require “some kind of metrics, points that define positions in time.” There are several possibilities for how this might take place: “One is assigning data to [a] certain position in calendar time, the second is assigning data to persons,” which data then confers age and thus a sense of time; and “third is [establishing a] relation between information and time according to [human] temporal ability” to process information.
According to some of those surveyed, that means an AGI at its early stages might need three levels of understanding time, in relation to humans and their communication — (1) calendar time to differentiate between old and new, archaic, and fresh ideas and concepts, and relation between ideas throughout time (how one idea stems from the other). (2) Time in terms of serving people throughout their seasons, respecting age-appropriate discourses, and (3) Human capacity for processing a certain amount of information per unit of time to properly measure information in terms of human time.
AI and Human-Centered Time Perception
How long should/will AI systems be human-centered and dependent (or allopoietic in systems theory terms), and will they at some point become self-centered and self-propagating (or autopoietic)? That will have consequences for the decision to mind the gap between the ability of humans to process information and an AI’s own ability; this self-or other orientation will also make a difference in the AI’s ability to develop a sense of human aging to connect with a human-like sense of time (if it uses that technique) — or to completely abandon those concepts and expel humans out of the loop and creation of its time framework.
Some propose that “we should not model the AI time perception outside the current time perception frames. We should adjust AI to our needs, not adjust ourselves to AI.”
“In that case,” another expert proposes “time could be framed as a choice of ‘process further’/’defer’ or ‘take an action’ for ‘real time’ choices. The idea is that at very fast timescales it would occasionally be advantageous to consume more processing resources (‘longer time’) than to take an action faster. This way we can construct the ‘experience of time’ as a kind of optimization problem: to act or to deliberate.”
As for the matter of potential subjectivity of time perception, some believe, “It is too complicated to answer this question. There are too many open questions. What is subjectivity?”; while others are more optimistic: The AI, because of its different methods of time sense and mode of existence “will be able to experience multidimensional time at the same instance”; moreover, says another, “all time is subjective to the observer’s frame of reference”. The idea of the observer and their capabilities is a very interesting one, and connected with the notion of sentience (GAIEI addressed this topic in one of its papers ). And based on these assumed capabilities the notion of perception of multidimensional time would bring us to an AI that might resemble one we now find in science fiction.
Some survey respondents warned that the notion of an AI “experiencing” time anthropomorphizes AIs too much. Furthermore, there are opinions that “AI systems will not need to have any perception of time. AI systems are (currently) tools built for a purpose,” and even with an advanced AGI that can perform many tasks, as a human can, “If such tasks don’t require understanding the passing of time, there won’t be any reason to implement that.”
To conclude, as in many similar cases, it is very hard to provide clear answers. As we know, it is not completely clear what the experience of time is like for the human race. But could thinking of how time might exist for advanced AI and the possible process of creating and coding it for AIs help us, again, better understand ourselves and the ways our brains operate?