Introducing the Hypermanifest - Mediating Time

Part 2 of my Hypermanifest Series

This is a follow up from my previous essay on the Hypermanifest, a construct that I have coined based on the interface between humans and AI. Drawing upon concepts of Willfrid Sellars’ Manifest image (the shorthand through which humans engage with the world) and Jean Baudrillard’s Hyperreal; a state at which the distinction between reality and simulation blurs to an extent where the simulated can feel more real than the original. In other words, the representation replaces the original. For a more in-depth discussion of the term, follow the link here.

This segment on outlining the Hypermanifest will focus on the concept of time; more specifically, the proposition of time zones that enable a synchronicity between humans and AI. This act of mediation, drawing upon the complex nature of internal and external temporal systems that humans currently use as well as the algorithmic processes of AI, could allow for the potential to differentiate between high pressure/quick turnaround and more long-term, careful planning that requires more time - similar to that of the dual-system or dual-process theory (popularised by Daniel Kahneman in “Thinking, Fast and Slow”).

I will discuss the ways in which humans perceive time and how AI could not only match algorithmic processes to create such a time zone, but how this can be implemented in daily life.

Comparison of Time Systems 

Humans and Time

Our human perception of time can be a mediation of both internal and external rhythms; from Neural sequences, Theta oscillations and Circadian rhythms to cultural constructs like clock time and language. Neural sequences (patterns of neural activity that represent temporal information in the brain, important for encoding and recalling sequential events) and Theta oscillations (rhythmic neural activity that occurs at frequencies between 4-8 Hz, associated with cognitive processes) are vital for maintaining and encoding temporal and spatial information (which also lends to the creation and storage of working memory), while Circadian rhythms allow for the regulation of sleep-wake patterns, hormone release and other daily functions.

This is in comparison to external constructs such as clock time - I will include clocks and calendars as well as global time zones in order to create structures for appointments and routines between people. Language builds upon this through providing structures for conceptualising and communicating time to others; tense, time markers and syntax allow for creating temporal sequences.

Time, therefore, is not just one external system but an interplay of internal and external, the latter of which is a construct that mediates structure for groups of people independent of their own internal time processes. Maybe a timezone that incorporates AI algorithms and human time structures would enable the same sort of mediated construction for alignment purposes and transparency, both of which are my aims for the Hypermanifest.

AI and Time

Although AI would not experience time in the same way as humans do, we can look to the structures of algorithms and running time to create a comparison to human timekeeping. This would take into account processing speed and parallel processing - where AI can handle multiple tasks simultaneously and at high speed without the same limits that multitasking can have on humans. Many AI algorithms function within polynomial time complexity, where processing time grows at a manageable rate even as input size increases - in other words, this gives it the flexibility to handle complex tasks swiftly. These elements of processing cycles, algorithmic runtime and model update frequency can create a kind of “cognitive rhythm” for AI systems. Their ability to operate across vastly different timescales can lend its unique flexibility in terms of temporality to match human speeds as well as processing and presenting long term trends when needed.  Taking all this into account, we can see how these properties could facilitate alignment with human time management needs.

In fact, one could say that humans and AI could both have a form of internal and external time. For humans, the interplay of Circadian rhythms, neural sequences and clock time can be similar to AI models’ connections between RNNs (Recurrent Neural Networks) and Transformers. RNNs, a class of artificial neural networks where the connections between nodes form a temporal sequence, are well suited for tasks involving sequential data and maintaining information from previous inputs (memory). Transformers, a type of neural network which runs off temporal embeddings and multi-head attention, could use the former to represent focuses of time systems, while its attention mechanism could identify relevant information between them (i.e. between long and short term planning).

Mediated Time Zones 

I propose the creation of mediated time zones for synchronisation between humans and AI. A hybrid system would integrate the benefits of both RNNs and Transformers - the former for continuous, real-time processing across each zone and the latter for in-depth analysis and long range planning. An attention mechanism could then be implemented to switch between or combine outputs as needed.

These could consist of 3 categories that can integrate both internal and external human time systems with AI’s algorithmic processes:

• Immediate Response Zone

AI can use real time data processing and attention mechanisms to respond instantly, similar to humans’ responses during peak alertness via circadian rhythms. This could be implemented for emergency services, live customer support and real-time monitoring.

• Short-Term Planning Zone

Just like humans’ working short term memory, this can be mediated with AI’s use of predictive analytics to assist with short-term planning. Examples of this would work for event and project planning for immediate needs; e.g. tasks that align with human work patterns and times of peak productivity.

• Long-Term Planning Zone

This would be reserved for strategic planning for time periods that span from months to years, which requires a sharp memory and complex decision making. AI, with its long term memory and pattern recognition, working with humans’ historical knowledge and external means of planning time (language, calendar years) could implement this zone for such things as economic systems and sustainable planning for cities.

In this way, the integration of AI’s algorithmic time with the internal and external rhythms of human time can form a new mediation of time through zones - allowing for more effective and harmonious interactions. This aspect of the Hypermanifest could provide a method that enhances alignment in increasingly advanced frontier models and algorithms across the board. Below is a flow diagram of how this could all link together, as devised by Claude:

Figure 1: Overview of the Hypermanifest

Implications to Consider

These are some future thoughts that I will expand upon as I continue to work on this particular area:

• Adaptation 

How would the interface be designed intuitively and accessible for humans? What cues would allow users to navigate the different temporal modes? Could the time zones adapt dynamically and flow between states - if so how would this allow for the particular task and the user’s cognitive state at the time? What would the run on effects be for how we perceive time (through external methods of language and clock time and internal rhythms)?

• Privacy Considerations

What information would people be willing to exchange with AI, and at what rate? Possible solutions could range from Helen Nissenbaum’s contextual integrity (the notion that privacy ensures the appropriate flow of information according to contextual norms), whereby each time zone could have its own sets of norms and expectations with information flow or certain sensitive information is only accessible within specific time zones or temporal states, to AI systems that adapt to individual user patterns with specific privacy techniques. Above all, transparency between the AI and user is key.

I hope to develop this section further by breaking down the elements of its interface, potential systems and implementations in specific case studies. I believe that this aspect of the Hypermanifest would allow the specific strengths of AI and humans to thrive and align in an increasingly uncertain world.