I recently learned about Peter Gärdenfors and his idea on conceptual spaces using Voronoi tesselations. I have been reading through some of his papers and came across this paper that he co-authored with Chris Sinha in 2014:
Time, space, and events in language and cognition: a comparative view
I have been studying time as well and its a short read and decided to read it.
The main idea of the paper is that human understanding of time is based on events. They propose that time in humans can be completely represented as event based and that a separate metric time system using clocks and durations is not universal and is instead a cultural and historically system that continues to help us today.
They show 2 types of time systems human use: use 2 different time systems: D-time (deictically based) and S-time (sequentially based). These map to McTagger’s famous paper “The Unreality of Time”, where he calls them A-series and B-series.
D-time/A-series can be described as being egocentric where any future event must “pass” to the past. This is also called “passage time”. D-time gives as words like yesterday, tomorrow, and next Christmas.
S-time/B-series can be described as sequential where events are described in relation to each other: before, after, earlier, later, first, last, etc. This has also been called “positional time”. S-time can also be cyclical. Calendars and clocks are based off S-time.
They state that humans believe objects are the core category of the physical world that humans track, but instead should be tracking changes in the world and learning to anticipate changes, events, etc. They reference Gärdenfors’s previous research that shows event structure and its subparts form the fundamental building blocks of meaning and grammar.
They talk about detach representations of events and how humans can apply events from one memory to other situations. This sounds like the same idea of disentangled representations that talks about how sensory data and the higher level concepts that may be activated are separate and so the higher level concepts must be stored in some other representation.
The describe a cognitive model of events that accounts for causality. I think this model could be implemented in computers. An event can be described as the combination of an agent, action, patient, and result. The agent is able to act, which in their framework means to exert force. Actions are modeled as force vectors or sequences of force vectors like walking. The result of an even it modeled as a change vector that represents the changes from before and after the event. The patient is the object being acted upon. In terms of causality, the force vector is the cause and the result vector is the effect. Even though time is not explicitly represented, it is there implicitly because actions and events happen over time. With this model, they argue that the order of changes (S-time) is enough to represent causality instead of some external metric time interval or durations.
They quote the mathematician Hermann Weyl: “The objective world simply is, it does now happen. Only to the gaze of my consciousness, drawling upward along the world line of my body, does a section of the world come to life as a fleeting image in space which continuously changes in time.” showing that the subjective experience of time bears no resemblance to ultimate reality.
I love that they propose a casual model. I wonder if it has been tested in machine learning. I have also been spending a lot of time thinking about time and how it should be represented in computers so its good to see that this simpler comparative model of before and after could be enough to model time in computers.