We may be ignoring one of the most important fields of computing, spatial computing, even though many of the IoT, AI, digital twin, and AR/VR applications explore pieces of the field. You can argue that recent progress in the convergence of IT and business processes is really related to spatial computing, so it makes sense to think about recent developments in terms of this notion. That means we need to look at in a closer, more organized, way.
Many tech terms, including “spatial computing”, have had multiple definitions over time. The current definition of spatial computing seems to come around in the 1990s with the advent of virtual reality, including VR games. The goal with VR is to create applications that can visually represent credible real-world behaviors, and so many of my friends who worked on this characterize spatial computing as a kind of real-time, movement-accommodating, computer-aided design, where a moving collection of objects can be represented as “seen” from various perspectives.
VR obviously has been expanded to include augmented reality or AR, and this meant that in addition to a representational mission, spatial computing had to expand to include means for acquiring real-world state to augment. Thus, IoT and video analysis, inertial information, and other things to learn the properties of real-world elements to be included, was added in. This also provided the ability of spatial computing to create models of real-world systems to permit manipulation of the real world in some way (in the early 2000s).
The easiest way, and the most complete way, to feed real-world information to spatial computing is video, because video is a direct window to real-world properties and visualization. Thus, we could say that there are three “dimensions” to modern spatial computing—analytic, generative, and control—that correspond to the use of spatial computing to “view” video to understand what’s happening, to generate video from a model, and to use the model to interact with and assert some control over the real-word scene being captured.
The first, analytic, mission of spatial computing is already in limited operation with the examination of videos to identify scenes or activities prohibited by usage guidelines. These guidelines may be related to obscene, unsafe, or other prohibited, behaviors. With the advent of user-submitted video, this field has been expanding, and is likely to continue to expand. These are largely examinations of stored video, but there are examples of real-time analytic mission spatial computing in military and medical fields.
The second mission, generative, has applications today in gaming, where a model of a virtual scene is converted into a realistic visual representation. It may also be used in military and medical applications, and even for the general mission of creation of instructional videos.
The final dimension, control, is arguably the frontier of spatial computing, combining analytic and generative capabilities to create a real-world model that can be influenced in some way by the application, and through the application perhaps by the user(s). This is the specific dimension/mission of spatial computing that relates to real-world applications, including both the work-productivity-empowerment and more general consumer missions (autonomous vehicles, safety/security, etc.).
It’s been my view that all applications of this type are based around a model, a “digital twin”. This model then feeds any set of “visualization” applications needed, including AR/VR, and also any control applications needed. In both classes of application, my presumption is that the model has to signal (via a generated event) that some change has occurred, and signal the set of new conditions to be handled. The application could then change the visual field being shared, or generate appropriate control responses back to the model. Here’s an article on an example of this, a framework I’ve called the “metaverse of things”.
There are four levels of application work needed here. First, we need to be able to create the digital twin models. Second, we need IoT/sensor and video analysis applications to populate the models with real world conditions. Third, we need to be able to create both the visualizer and controller application sets that process the model state as needed, and finally we need applications that integrate that processing with real people. Depending on the relationship between the models and missions, the last two levels may be coupled, but in business processes at least, we could assume that some reformulation of current processes would be needed for integration.
The barrier, from what enterprises tell me about the creation of new applications relating to business process automation (AI or otherwise) is the sheer level of interdependence these applications expose. We have human and IT processes in place that have to accommodate what we do. We have different real-world activities to capture as models and to control. We have different sensor, controller/effector, and network requirements, and we have incumbent and prospective vendors for all of this. There are multiple ways a given new BP application might be done, each with its own required technology tools and organizational impacts. Where do you start?
This is the same barrier that impacts most vendors, because the breadth of the required solution exceeds that of their product lines and monetization options. Why launch a partial solution, hoping that others will do their own part in completing it without stealing your part? Why even think about it?
I think the answer to those questions may finally be emerging. What is it? Because if you don’t, someone else will. We’re, with spatial computing, introducing the key to unlocking application integration with the real world. NVIDIA clearly sees this, and has responded by productizing part of it and creating an alliance for another part. Can others, like AMD or Broadcom, fail to respond? Can players like Dell and HPE ignore the opportunities that this would create, and by doing so cede what might be the largest market for tech to ever exist to another server player? Not to mention Google, Microsoft, Amazon, Meta, IBM, and Oracle.
I think 2025 may be, just may be, the critical year for tech, and you all know how not-prone to optimism I am. We’ll see.