Part 1: Sources for AI Time Perception
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/
· Note: Quotes in the text are from the materials sent by our responders, which will remain anonymous, and therefore are not individually cited.
The Global AI Ethics Institute did a small survey about this topic in order to start an interesting conversation. We called on our members and general audience to contribute to this opinion piece, and we collected several answers that are summarized here.
As mentioned in the short survey’s introduction, the history of time is deeply rooted in the tight relationship between living creatures and their natural environment. So far, it seems that these questions, and many others surrounding the relationship between time and AI, have gone totally unexplored.
Before we introduce the questions we used, it is essential to note that this topic itself has a number of base assumptions that could be explored, and thus that respondents to our questions might need to grapple with: First, how likely is it that AI would attain a high and flexible enough intelligence that it could become sentient? Also, we attach the perception of time to the living, at the least — and usually to sentient beings; can we implicitly assume that an extremely advanced AI system would have its moment of awakening, and maybe a moment where it “dies”? Or could future sentient AGIs be immortal? Could that span of time also be a factor that influences the perception of time?
So, given these caveats, here are the questions we asked our contributors in order to spark their creativity and to get them to look at the topic at hand from a few different angles. Those questions were:
(1) What do you think will be the basis for the AI experience (and measurement) of time?
(2) What could we base it on, or what could be independently developed later?
(3) Since humans will frame the basis for AI time perception in the beginning, upon what should we model AI time perception, and for what purpose?
(4) Is subjectivity in the experience of the passage of time something that we could expect from AI systems?
Potential Sources for AI Time Perception
In answer to the first of our questions, expert respondents said that assuming that an Artificial General Intelligence (AGI) could be made strong enough to attain sentience and operate on its own, the basis that it might use for its own time framework might have at least a few sources and considerations:
1. Circadian rhythms — similar to organic biological processes that happen over a 24-hour time span. This could be something that might become a natural AI timing “device” emerging from its smaller components, just as a human’s biological clock has roots in the molecular level of the body and in its genes. This process in AI could work synchronously with some learning or update cycles.
2. Information processing/processing power — These elements of how AI and our brains work may serve as limits to how time is processed. Again, we humans have biological boundaries to our processing power — how much information our neurons can send and receive. “Each new human builds [their] knowledge and experience from [a] fresh start, having at [their]disposal all the written/recorded knowledge of people before them — but the ability to gain that knowledge is limited by the ratio of the amount [of] information to time. AI is limited by processing power per unit of time — but it can be improved — while humans have a limited capacity to change [the] amount of information per unit of time.”
3. Celestial movements (Micro/Macro cosmos) — could be used as some kind of reference for AI systems that have the power of sensing the position of celestial objects. We can speculate whether the sense of time inherent in this system would be the moon, sun, or planets in our solar system or references to some distant but observable galaxies. This system could be one of the most tangible references of time — in the long term — in connection with the need to observe the cycles of agriculture, which has always been the engine of human civilization, and which we humans take as a point of reference. Or AI could take as a basic measure something on the level of subatomic particles and their movement.
4. “State of hardware” — “An ‘AI experience of time’ as it relates to planning might be intimately connected to the degradation of circuits or hardcoded time increments that align it with our measures of seconds. Another consideration might be how an AI’s experience of memory is encoded and the effect of that memory structure on future planning. ‘Longer’ memory would have the effect of ‘lengthening’ time relative to an observer’s experience. That would be similar to and interconnected with the first sources, and it would be something like the decay of human bodies, but obviously on a different time scale. It could be an artifact of maintenance cycles, for example.
Even if we provide simplified answers as those above, we should consider many other philosophical questions: One of our responders points out for example that, “Humankind needs to decide whether the time is considered an objective [phenomenon,] as Kant maintains, or a subjective phenomenon. Does time exist without us or did we make it up with measurements like seconds, minutes, hours, etc.?” We have units of time measurements responding to our experience, what might be developed by future AI systems? Will they have much shorter basic units (some parts of milliseconds, based on electron movements) or much longer [ones], considering [their] potentially significantly longer lifespan, perhaps based on the beginning and the end of the universe?
Some respondents disagreed with the subtle implication that time can be objective and that we all therefore experience time identically. One person argued that time is always subjective: “It is observer relative. Aristotle constructed our experience of time as an experience of change in our environment and [the physicist and philosopher] Ludwig Boltzmann interpreted our experience of time as a ‘blur’ of our perception of change below a particular granularity. On that basis, ‘time’ would be approximated as a relationship of granularity/processing as experienced by an observer. For example, a machine that processes higher granularity than a human and fast enough to ‘not miss anything’ that we can see, could be said to have a slower experience of time. But we might be more interested in time as an indication of planning horizons. For example, a human believes that when certain changes are observed (i.e. aging and rotations of the sun), they will die. Consequently, every decision is considered relative to the changes that are expected to take place. To this extent, the concept of ‘during’ is a parallel simulation of changes like aging against other actions.” This line of thought is coincident with the “state of hardware” consideration described above.