Industrial Multiverse
TLDR:
Introduction
The concept of the multiverse has been ideated for a long time in philosophy, physics and science fiction. In recent times, the industrial metaverse has grown in popularity as a major “channel” for extended / virtual / augmented reality applications, often on top of digital twins (as the connective layer) and including generative AI (mostly for sizzle). In this article, we explore an alternate vision of the industrial metaverse, as a bridge between the limited but highly valuable human and the rapidly growing machine capabilities.
Definitions
Multiverse
?
Metaverse
In short: multi means many, rail means rail, and meta doesn’t mean Facebook Meta, it is a Greek prefix that can mean “beyond”, “after” “changing” or other related(semi?) things. In modern usage, adding meta- to something makes it more meta.
Declaration: I’ve decided I really don’t like the word meta.
The metaverse
The metaverse as popularly described is an immersive experience where lots of people can do stuff in a virtual world where they can see each other, and the places they do this stuff can change. There can be many “places”. The easiest current example of an “early metaverse” is Fortnite or Roblox – there’s lots of games, and your character can enter any game, and the other people in that game can be interacted with. The visual identity of the user is also malleable; you can wear different skins.
There are several implicit assumptions that go along with this, that help us to anchor ourselves in the concept. For the designers of games, and indeed any software, the balance between innovation and familiarity is paramount. Users must be able to easily adopt new ways of doing familiar things (walking), in order to grasp unfamiliar concepts (low gravity). The fusion of familiar and unfamiliar concepts gives rise to low gravity parkour, 3D race tracks and “immersive meetings” where avatars sit around a virtual table talking about low gravity parkour games.
In this context, the “multiverse” means “many universes”. There are many games, many places to visit, and finally enough meeting rooms. At this point, the leap occurs to ideating the many things we can use these many universes for. Playing games, visiting places, and… meeting people.
I, personally, have many doubts about this version of the metaverse. I do not doubt the “forecast market size and growth“ posts that many marketing and analytics companies produce, however, I seriously doubt that this tame, linear, view of the metaverse will be what powers this growth and adoption. For games, yes, because it already works as is, and many “metaverse” games are popular not because they’re metaverses, but because they’re fun. For work, no.
The industrial metaverse
The industrial metaverse is the work version of the metaverse. Same descriptions of immersion and being able to see what is happening, and then a long, low gravity leap to business benefits. Improved predictive maintenance, improved safety, lower operating costs etc. In nearly every version of this play, the causal link between “immersion” (which settles out as really the only defining characteristic of most industrial metaverses) and a benefit such as “improved predictive maintenance” is either missing, or attributed to “AI”.
Immersion of the human + AI to do the work = better meetings.
There are exceptions of course, and the main exceptions are tech cos who are actually building and selling metaverse solutions, as opposed to consulting firms and large incumbent software companies who seek to market every new fad that comes out.
A great example of a real metaverse tech co is Minverso, who creates immersive training environments for heavy industry, e.g. mining. Specific use cases include improved preventative maintenance – the trainee can practice following the work instructions to perform the task virtually, in a simulation of the actual workshop they will work in on site.
Siemens recently hit it out of the park with their CES event, and introduced a fantastic use case around a factory project, utilising digital twins and simulation, and detailed owner/operator benefits.
Microsoft and Nvidia are also pushing very hard, albeit MSFT seems to have pivoted a bit to XR as their version of metaverse (as many others have) and Nvidia is pushing very hard to make Omniverse and USD industry standards in metaverse creation.
For the industrial metaverse, there are clear value props associated with immersion and simulation:
?
These are clear value and benefits to these use cases. They're also quite “narrow” – designed to do this one thing well, and they will no doubt search for adjacent use cases to deliver more value with the same or similar technology, but no one who ideates the metaverse would say that training is the only or even the primary use case.
This then is the state of the metaverse in 2023, and industrial metaverse in particular: “wide” descriptions of immersion that will “change the world”, and “narrow” use cases that deliver benefits but are not sufficiently “metaversy”.
Industrial Multiverse
Let’s now consider a different interpretation of the metaverse and unbox some of our metaverse assumptions.
Singular
A singular universe is one in which all entities are unique, and presupposes that there is a single version of events from future to present to past. In other words, there is one “version” of reality, and just because we can’t “see it all” or “see into the future”, it’s still occurring and will occur and we just need to improve our ability to predict those occurrences.
We are singular creatures. There are many people, but there’s only one “me” and only one “you”, and there can never be more than one “me”. Even a clone is a separate singular entity. There could be many clones, but they are still individual entities. Similarly for assets, and the world we seek to immerse ourselves in.
Games create new universes in order to create novel experiences. The industrial metaverse seeks to create a virtual experience that most closely resembles “reality”.?
In other words, the industrial metaverse environment seeks the opposite of the game metaverse environment. It’s no wonder we feel cognitive dissonance when using this word.
The change in perspective I propose is plurality: there is no “objective” reality to bind ourselves to. The world as we see it is a single version of the world, interpreted through the very specific rods and cones of our eyes (each of which is unique even in our own head). Our organic cameras only see a certain portion of the electromagnetic spectrum, and we interpret that portion differently to others who are looking at the same place. We cannot see heat, except in extreme temps into the near infrared, we cannot see vibration, except with very large amplitude, we are limited to line of sight (we cannot see through dust or buildings) and we are limited to a singular viewpoint. Even when we look at a bank of cameras on a screen, we look at each one in turn, and the faster we “scan” multiple viewpoints the less data we ingest and comprehend. Same same for hearing, a selected portion of the spectrum, and in many industrial settings we must wear hearing protection, further limiting the fidelity of data we ingest. Our smell is quite good, and sophisticated, and unfortunately will be absent from metaverse type experiences for a while. Touch also is very good, but with gloves on, and strict training in “not touching stoves or wires”, it’s also deliberately limited in scope for discovery of new information, especially in kitchens and industrial settings.
A singular immersive perspective can undoubtedly improve situation awareness, and this in turn can improve decision making. There are undeniable benefits associated with improved decision making, a topic I write about extensively. We can, however, go much further than maximising a singular human view of the universe.
?
Time bound
An extension of singular thinking, time bound thinking presupposes that the future is unknown, but only because we don’t have enough information. Once we have enough information on the present, we can predict the future. This view has been massively pumped up by recent advances in ML and AI, and the many and mostly underappreciated parkour jumps in predictive analytics. We can, for many simple things, create predictions of sufficiently high accuracy to place bets with vastly improved odds of success. I say place bets, because most people still do not understand that the predictive process is inherently probabilistic, and for every “prediction” we must discard or prevent many other “predictions” from seeing the light of day. This weighting and discarding of probabilities are often subject to judgement, and heavily influenced by bias in the data and the decisions of what data to collect and analyse.
Similarly for our past, the idea that the past is known is still prevalent. Anyone who has conducted an accident investigation, or performs root cause analysis for a living, knows that there are many versions of the past, told by different singular entities (inc machines that ingest and store a single data point), and that the “recreation” of past events is a process, and nearly always involves judgement and the discarding of lower probability possibilities. I encourage you to watch Air Crash Investigation if you want some very concrete/mountainous examples. This is also true for what resides in our head. One of the many fallacies of the mind is that we remember what happened. We encode what seems important to us as we ingest it, which is a fraction of the fraction of “reality” that we perceive, and we then store fragments of this across parts of our mind with many other similar parts. These fragments can get mixed up or altered (it’s not a hard drive, it’s more like a bag of lego), and we then retrieve parts of that memory and potentially other memories or even hallucinated fragments that fill in the blanks. For want of a better term, our memory sometimes is partly hallucinated and we have no idea which parts… hint hint, all those throwing rocks at Chat.
领英推荐
Linear
As linear creatures, we move from the past into the future, inhabiting only the present. We think we are moving into the future, but we are assembling a view of the past to explain what is happening now. We don’t know the future, and if you are of a quantum physics bent, you might refer to wave functions and collapsing when we speak of the present. The many universes may or may not collapse, or be piloted, by our measurement, but the “reality” of every universe is that the future is probabilistic, and the probabilities are determined by the possibilities that we analyse, and then there is one state that occurs, and that is the present. The many possible futures “collapse” to a single past.
As singular, linear creatures, we examine a fraction of the present, and we further bias our understanding of what is happening by biasing (remembering parts of) the immediate past. We hear the screech of tires behind us, and then we turn and see the car on its side, and then we see the tire marks, and the whole while we assemble our understanding of the present in real time, based on the fragments of information we parse and remember, which are impacted by emotion and attention (that screech made me jump!). We assemble the present from the past, and we project into the future to test our assembly against what we think might happen next. Based on what we ingest next, we firm up or discard the assembly and keep repeating. Past to present to future, working backwards from what’s inside our head. It works, because we live in a singular and time bound universe, and linear thinking helps make sense of the progression of events, and this is the universe that exerted the evolutionary pressure to create this way of thinking.
So what?
This is probably the right place to pour a second glass.
The so what is that with the advent of modern technology, we can ingest and store 1000x the data that a human can, far beyond the bounds of human perception, and we can utilise this data to examine and predict many futures, and to analyse and learn from both the past and our predictions. We can create awareness across as many versions of our universe as we want, and we can seek to maximise learning in a non-linear way.
We’re thinking way too small with the metaverse as an immersive place to insert the Mk1 eyeball. Let’s not be bound by the flesh we inhabit. It’s a metaverse, not a meatverse.
Thinking small = singular, time bound and linear = flesh.
Thinking big = pluralistic, timeless & non-linear = multiverse.
In a pluralists industrial metaverse, the objective is not to create the most “realistic” immersion for a human. The objective is to maximise organisational learning to make better decisions. That maximisation will include highly realistic immersion for human decision makers, but it should not be limited by the attention and perception of humans. The industrial metaverse must encompass at least some of the multiverse of possibilities that the organisation faces, and work backwards from there, finding the right place for human decision makers to have maximum impact and enabling machines to grow to do the rest.
?
Hurdles
The first hurdle is taking the singular human out of the centre of the picture. In the same way that we finally accepted the universe doesn’t revolve around the Earth, we similarly need to accept that the metaverse needn’t revolve around us. The industrial metaverse is defined by the industry and organisation we operate within. The boundaries of our industrial metaverse are set by the physical and organisational realities of our work. The metaverse must improve organisational outcomes or we should not do it at all, and therefore should be anchored by the reality we are managing. In other words, the situation and problems we face should define our metaverse, not the human eyeball. For my designer friends, I am not eschewing human centric design; rather I am saying that the problem space should not be human perception, it should be the organisational problems, and the technology solutions should be human centric.
The second hurdle is moving our thinking beyond what a human normally sees or does in the real world. Humans create software, and software reflects the human, in the same way that products reflect organisations. That which creates, creates in its own image. Unless it deliberately doesn’t place itself at the centre; if the situation and the problems are at the centre, then there is freedom to best solve for the problems within the context of the situation. Critically, the move from singular, time bound (“what is the right answer?”) view to the many possible futures based on situations and evolving models, requires the consideration of hundreds or thousands of possible futures. This is a far cry from the “give me 3 options” of today, and when we bring these possibilities into the present, considering they may appear and change in very short timeframes, and that our past will be littered with millions of unrealised possibilities that we can still mine, the limitations of humans as the sole decision makers are laid bare. To improve how we manage complex systems of systems, we need to move beyond gut feel and PowerBI. The caveat to this of course is that for many organisations, with simple systems and simple models, this may not be true. A metaverse of possibilities might be overkill; I tend to think these are the systems where AI as a supporter to humans will still make major inroads, but I struggle to see much value from near term metaverses in this situation at all (unless of course, they are very narrow products, ala immersive training, in which case I'd argue they are an immersive training product). If the system to be managed is sufficiently simple, keep it simple, sapien.
?
The third hurdle is moving machine decisions to the fore for problems that machines solve better than us. The incredible advances in AI, particularly agents, open the doors to considering how we can utilise the incredible amounts of data and capabilities that are being brought to bear in industrial settings. We know that most data we collect is not used, or at best is used far in arrears and that those using it are very often far removed from those who make operational decisions. We hope that long feedback loops and processes like continuous improvement can help us evolve fast enough to cope with change, but we limit our ideas of what is possible to small single digit numbers per annum.
We subconsciously limit our ability to conceive of potential change to the limit of humans to process change. The challenge of unlocking non-linear learning is almost certainly going to be rate limited by the ability of humans to process change, rather than the ability of machines to grow and learn.
To be clear, I am not proposing some maximal “machine future”. Humans are the current primary form of singular decision maker, with limited perception and cognitive bandwidth, and highly evolved comprehension. We operate in different modes, whether seeking Goals or seeking understanding of Situations. Enhancing Situation Awareness (SA) is a key user benefit of the metaverse, and the design of the solution to enable improved SA and faster regaining of SA is critical. There are ways to do this. We should do them. Immersion will absolutely play a critical role. However, we should not be thinking of humans as the only decision makers, nor spatial computing as the panacea of human decision making. Hybrid decision making, humans + machines, is without doubt the frontier that will occupy us for this century.
There are many near term possibilities for us to explore:
etc.
Soon the machines will play ever increasing roles of importance, and the metaverse may be the key link to help humans stay sufficiently in the loop of what the machines are suggesting and doing. If we can simulate, forecast and predict 1000 possible futures, we can and should use agents to observe and compare across all of them, to find the patterns and learnings that help us make decisions in the present, with our small but highly sophisticated mental models.
But what does it look like?
To ideate a multiversal metaverse, we can visualise some of the key elements.
The timeline of an industrial metaverse will look like hundreds, or thousands, of alternate timelines, some branching off and suddenly ending, some budding multiple children that run for many days or months before all but one end, and some that sit far away from the others because they’re so unlikely or decoupled. Visualising possibility and probability at many points in time is not really a thing, yet, but there’s many tantalising visions portrayed in science fiction. Similarly, our modern industrial processes support this. Think of feasibility studies or options presented in a recommendation or risk assessments during a project. Each of these potentials is an “alternate universe”. We are constantly assessing possible futures, but the vast majority of these assessments are closed and locked inside static documents. While text mining will provide invaluable learnings for models, the real learning we want to happen is by decision makers, and no human has time to even find, never mind review or learn from, all of these possibilities. Chatting them all together into a summary is good for keeping meetings on time, but less good for training your mental model.
Running through the many past and future alternate universes is the strong, bundled thread of “reality”. Some parts of the thread are unaccompanied by any future alternates; something happened which no model predicted, and no plan encompassed. It happened, it’s done now, but that doesn’t mean we can’t learn from it. These “unaccompanied” past sections are perhaps the most important learning opportunities, and in an industrial multiverse we would be able to access the analyses, simulations or investigations conducted in arrears to determine what we need to learn and change to ensure that in a similar situation, our models better predict this occurrence. Almost certainly, some previous assessment or prediction encompassed this situation, and being able to find and link that from anywhere in the metaverse is critical to discovering the invisible but causal connections between situations.
The recent past, the present and the near future are a constantly shifting array of possibilities, many of which can continue to co-exist “in the past”, because we simply don’t know which of the many universes collapsed to form “the past”. Think of it this way – when you read an ops report from the Shanghai office dated 3 days ago, that is based on data from last month, you are “collapsing” the many possible universes down to 1 – the state of play in Shanghai 1 month ago. You don’t know what is happening right now, but you assume that if anything “bad” was happening (read: emergency), that you’d be notified. In a metaverse, there is only one set of possible futures. All decision makers, human and machine, local or remote, can see what has happened, what is happening and what possible futures are being considered, within the bounds of information velocity (i.e. if you only report every month, it won't matter how fast your internet connection is...)
Across a complex organisation, with thousands of possible futures across many time horizons, the ability of an unaided human to make sense of this is limited. In fact, without artificial guidance, this bewildering array of possibilities may be worse than useless. Cognitive overload and high uncertainty reduce decision making effectiveness. The collection and aggregation of data will be useful for analytics, and the ability to visualise and create relationships in data across our lakes and warehouses may be a secondary benefit. But the primary purpose of a metaverse is to improve decision making, and that must include the front line managers and workers, and the many machine decisions makers spinning up every day. With many possible future universes, the role of human decision makers is to shape the creation of good possibilities, limit the probability of bad possibilities, and manage the reaction to the emerging “collapsed” universe.?
A complex organisational metaverse will be far larger than any human can parse; which is fine, because we should be limiting the “view” of workers to that which aids them In achieving their goals. Goal directed immersion, with freedom and guidance to explore. But unlike the retail metaverse, we don’t want all of our workers spending their time exploring and sitting in virtual meetings – we want them working towards our goals, and utilising insights into the future and learnings from the past to make better decisions than what is possible today.
Shaping possible universes to improve the probability of favourable realities is a worthy goal for humans in the machine age.
Many features of this metaverse emerge from such a vision. The ability and desire to create possibilistic and probabilistic views of the future, persistence of historic simulations and considered options, the ability to link actual data to simulations, the ability to visualise probability and time, the need for agents to guide and request precious human comprehension, the need for extended reality (XR such as AR or VR) to create immersion at the right place and time, amongst many others more prosaic but just as critical enabling features.
Summary
The objective of an industrial metaverse is to maximise organisational learning and optimise the decisions that the organisational elements, both human and machine, make. That optimisation will include highly realistic immersion for human decision makers, but we should not limit our machines by the attention and perception of humans, and it should not be limited to a singular, time bound view of the situation and problems. The industrial metaverse should encompass the multiverse of possibilities that the organisation faces, and decision making should be viewed as the process of shaping those possible futures to improve the probability of achieving organisation goals.
Takeaways:?
I will leave some examples of how such a multiversal view could change our ideas on metaverses and the role of humans another time. I can think of some very different views of smart cities, value chain optimisation, many sided marketplaces, wikipedia and teaching kids. And for those Greg Egan fans out there, this might also be how you create the Truth Mines.
#digitaltwin #industrialmetaverse #metaverse
Rob is the co-founder of Geminum, a digital twin business that serves heavy industries such as mining, power, construction, manufacturing and aviation.