icon picker
POC: Spatial Intelligence (GPS 0.0)

Robust and efficient multigenerational spatial intelligence

Where do we find ourselves

Knowing where things are is important. Especially when coordinating activities amongst multiple agents. Landmarks, invariant features of the spatial environment, are indispensable. However, recognizing invariance is a problem in itself. The longer a feature stays invariant, the better it might serve as a landmark. E.g. I might use a fancy red sports car car next to which I park my regular grey vehicle as a landmark, but might do better remembering the storefront opposite it, or its position with respect to a prominent street intersection, or via a pin in my GPS, or ... (you get the idea).
Remembering directions, distances has been essential to the survival of most creatures. Towards this end, landmarks are important to remember and share. Humans can use language to improve on this capacity across generations. Not everyone discovers North on their own – most of us are born into a culture that helps us learn what it means, and then use it to situate ourselves and coordinate with others.

Lost without Siri

Many of us in the twenty first century have noticed a loss in our sense of direction and our ability to situate ourselves in a broader geographic context. This seems to be correlated with the advent of portable GPS systems, particularly those in our smart phones like Google Maps. We struggle to quickly know which way is North. Why? because we may use our always on communication device to show us where we are relative to where we might want to go. We have lost the habit of remembering where we are.

This has not always been the case

Lera Boroditsky reminds us that illiterate kids with the right language practices can have a better sense of direction than PhDs. The question is can the PhDs acquire that capacity when needed?
Excerpt from one of her
… I'll start with an example from an Aboriginal community in Australia that I had the chance to work with. These are the Kuuk Thaayorre people. They live in Pormpuraaw at the very west edge of Cape York. What's cool about Kuuk Thaayorre is, in Kuuk Thaayorre, they don't use words like "left" and "right," and instead, everything is in cardinal directions: north, south, east and west.
And when I say everything, I really mean everything. You would say something like, "Oh, there's an ant on your southwest leg." Or, "Move your cup to the north-northeast a little bit." In fact, the way that you say "hello" in Kuuk Thaayorre is you say, "Which way are you going?" And the answer should be, "North-northeast in the far distance. How about you?"
So imagine as you're walking around your day, every person you greet, you have to report your heading direction.
But that would actually get you oriented pretty fast, right? Because you literally couldn't get past "hello," if you didn't know which way you were going. In fact, people who speak languages like this stay oriented really well.
They stay oriented better than we used to think humans could. We used to think that humans were worse than other creatures because of some biological excuse: "Oh, we don't have magnets in our beaks or in our scales." No; if your language and your culture trains you to do it, actually, you can do it. There are humans around the world who stay oriented really well.
And just to get us in agreement about how different this is from the way we do it, I want you all to close your eyes for a second and point southeast.
Keep your eyes closed. Point. OK, so you can open your eyes. I see you guys pointing there, there, there, there, there ... I don't know which way it is myself …

Can we try and measurably improve this capability?

Yes. We propose to develop a mixed reality system of Spatial Intelligence learning, that makes it more efficient for humans and robots to communicate and coordinate with each other about a given world. Using principles from Roy Rappaport's theory of language and ritual, and Karl Friston's principles of Active Inference.
For humans it will take the form of cognitive transformation via augmented reality enabled interventions that include pointed language use.
For robots it will take the form of components that include natural language processing and generative model development by means of a programing by example protocol. See more .

CONTEXT: Adapting existing Open AI Gym Frameworks

By way of background, we have some experience with the OpenAI API and Gym framework. This might come in handy for ABM simulations.
For example, the OpenAI "Simple World Comm" , depicted below, might serve as a skeleton game environment for Spatial Intelligence exploration. To model agents are foraging for food in a 2D spatial environment, with adversaries capable of rudimentary, one-way, symbolic communication.
This link can't be embedded.
Setup details:
There is food (small static blue) that the good agents (green moving) are rewarded for being near, there are ‘forests’ (large green static) that hide agents inside from being seen, and there is a ‘leader adversary’ (dark red) that can see the agents at all times and can communicate with the other adversaries (light red) to help coordinate the chase. [GIF below: 2 good agents, 3 adversaries, 1 obstacle (large black), 2 foods, and 2 forests] ... the good agents reward, is -5 for every collision with an adversary, -2 x bound, +2 for every collision with a food, and -0.05 x minimum distance to any food. The adversarial agents are rewarded +5 for collisions and -0.1 x minimum distance to a good agent.
I think we need to make the setup a bit richer for our purposes, (i) incorporating multiple generations of (ii) active inference agents with and without (iii) ritualization of communication for an emergent absolute coordinate system (N-S-E-W). Goal would be to try and quantify inter-generational and social learning, as well as performance gains from richer symbolic communication, with and without "error control" via ritualization.
The intuition is to take an agent-capability lens to model the combined effects of ritual, language and active inference. We would like to extend this to include important social and/or intelligence capabilities in a multiagent setting — using data from anthropology as well as via ABM simulations. For example, the capabilities could include: Memory, Prediction, Coordination, Communication, Construction.
Construction, in addition to physical building, could include abstraction, counterfactuals and pure fabrication/lies (but lies that are surprisingly healthy, “poetically honest”, and useful in practice — e.g. names of a thing Vs the thing). Furthermore, one way of constructing a shared world is to construct a quantized coordinate system (e.g. north, south, east and west). The starting thoughts of the POC below around Spatial Intelligence sketch the notion in a multiagent context.
One other quick connection of spatial intelligence to the capability of memory. Mapping the task of remembering to images and space has a long and well proven tradition. Having tried to memorize the order of a deck of cards myself I can attest to the power of this approach.
We hope to be able to extend this rudimentary spatial intelligence framework for more demanding tasks like building institutional facts (e.g. country borders) and abstractions (e.g. multilayered cosmologies with heavens, hells, and nonhuman beings of various sorts).

CONTEXT: Navigational technologies are important and valuable

GPS technology has taken a few centuries of development

... in 1714, the British government passed the Longitude Act, which led to the formation of the Board of Longitude. The board offered a large monetary reward for an invention that would solve the problem of determining longitude ... John Harrison, an inventive London clockmaker, developed a friction-free clock that proved a superior method ... Harrison’s clock passed a series of tests and on a nine-week Caribbean journey, lost only 5 seconds, equating to 1.25 minutes of longitude.
Fast-forward to 1959 when a joint effort between DARPA and the Johns Hopkins Applied Physics Laboratory began to fine-tune the early explorers’ discoveries. TRANSIT, sponsored by the Navy and developed under the leadership of Dr. Richard Kirschner at Johns Hopkins, was the first satellite positioning system ... The TRANSIT system, which used six satellites (three for positioning and three as spares), was based on the finding that the Doppler shift could establish the location of the satellite in relation to the receiver station.
In 1973, the Department of Defense (
) called for the creation of a joint program office to develop a unified navigation system. The resulting system, named the NAVSTAR Global Positioning System, successfully launched from Cape Canaveral in 1989. Twenty-three more satellites were later launched to complete the configuration.
There are three components necessary for a GPS to work: ground stations that control the system, a configuration of satellites fitted with atomic clocks, and receivers carried by users. With all three of these components in place and operational, it was time for DARPA to focus on improving the receivers soldiers were using. One of the early receivers was the PSN-8 Manpack. Between 1988 and 1993, 1,400 of these units were produced and utilized by military personnel. However, these units proved to be cumbersome and impractical for military use. The units were large and weighed close to 50 pounds.

And it remains important for efficient coordination to this day


CONTEXT: How does this connect to JTF?

We may be thinking too literally on this one, but we see religious cosmology and the accompanying world views as points of linkage to the spatial intelligence POC. Let me try and run my argument by you in outline form:
"World Building" is spatial, among the many other things that it also is, as Rappaport indicates above
Cognizing space is fundamental in our cosmologies: Many of us feel the need to look at problems at a 60,000 foot level to identify the main issues, Popper philosophized Three Worlds, Shamans travel to the world of the spirit masters for new perspectives on challenging problems, etc.
Folks like Latour (see ) and Rappaport ( seem to be reaching for an ecological embodied and networked cosmology -- for a postmodern synthesis of science and religion. Yet, I think we are left disoriented without a concrete world view (where are the spheres? what is above/below, far/near, inside/outside? where are the stars? what is space made of?)
Using spatial intelligence as an entry point to understand and model cosmological primitives, I think we are naturalizing the role of ritual for humanity’s distributed “world building”. And in a particularly concrete way -- by exploring the cognizable spatial structure of various built worlds.
Religions, including post-modern ones, imply a cosmic order and a world view. By naturalizing how this is done cognitively via ritual, language and active inference we will be following through on Sir John’s expectation that “scientific revelations may be a gold mine for revitalizing religion in the 21st century”

Want to print your doc?
This is not the way.
Try clicking the ⋯ next to your doc name or using a keyboard shortcut (
) instead.