Gazing above and peering within
“I think, therefore I am.” Or was “I dreaming of the butterfly or the butterfly dreaming of me?” The very nature of reality is just beyond what we can
conceive, and as we tumble down the rabbit hole, philosophers like Descartes have argued that we should first doubt everything, then break down reality
into the smallest bits possible and figure out the innerworkings of the smallest unit, and then piece them all back together to find the grand penultimate
theory of nature and reality. However just thinking hard enough, or even thinking different will not lead us to discovering a unified theory of everything
– when it comes to uncovering the mysteries of the universe “we must know, we will know” simply isn’t going to cut it.
Since the dawn of civilization, we have thrived to pull back the mysterious veil of mother nature, and to pinpoint the inner mechanism of the moral
compass that guides us – the very two things that Kant alluded to – “two things awe me the most, the starry sky above me and the moral law within.”
While peering above and within, we find ourselves in the face of a significant hurdle – the problem of induction. British Philosopher David Hume put forth
this problem of induction, stating that induction alone is not enough for understanding the entire universe and predicting the future; because future
simply can’t just be induced from past, whole from part, unknown form known. However, you’d be wrong to think that if pure induction won’t work, we could
turn to deduction. Godel’s incompleteness theorems, or his proof, tell us that under an axotomized formal framework it’s impossible to prove theoretical
hypothecation without correlating it with real empirical data.
We know now that when it comes to orchestrating nature’s grand symphony induction alone won’t work; pure deduction won’t either, cause quite frankly
we’re not the conductor and have never been and won’t ever be. We humans have tended to put ourselves at the center of the universe, and we’ve formulated
an induction-deduction methodology, induce what we can from experience, and then deduce the rest of the unknown. While this has served us well in some
specific cases, as long as we find ourselves resident in this vast universe in which we cannot jump out of, we can’t find a “one-size-fits-all” grand
solution.
Now, from the previous “revolutions” of Newton and quantum mechanics that upended our views of the universe, we hope to pioneer a paradigm shift –
that’ll gain the most knowledge from nature, and it’s empirical and observed data. Thus, in order to truly “overcome” the limitations of our knowledge,
our attention must be first diverted back to induction and deduction – however we do have a powerful “friend” to help us out a bit.
With the rise of computing, one in which we can’t help but wonder whether a machine could go about discovering the laws of nature. Throughout history,
we have borne the brunt of thinking and with computer becoming more powerful, the words of Alan Turing start to become more relevant – “can machines think?”
Could we get a machine to discover the laws of nature on its own, one that can work around the clock, and do its own conscious thinking?
A computer that we have now was merely a pipe dream in the days of Turing. One that could think was only limited to the imaginations and annals of
science fiction. The advent of Artificial Intelligence now is like the impossible of when our ancestors looked up to the birds flying in the sky and
could only sigh or wish that we humans could take to the skies as well.
Unfortunately, the current trend of AI, LLMs, and the world model are all the hype nowadays. Every big tech company is engaged in an “arms race” to
build the next AI scientist, chatbot, or humanoid-like robot with an LLM or the so-called “world model”; however, the million-dollar question is how to
overcome Hume’s problem of induction? Without a solution everything’s just self-hype – a bubble waiting to burst. Maybe we should ask ChatGPT how to do so;
it’ll probably have no clue just like the companies that made it. The “big five” clearly don’t know how to implement a fundamental, generic induction engine
to find regularities from any raw data, all without relying on any prior knowledge – the more may be merrier but the more data not so necessarily. Without
an induction engine, just like how a car without a motor engine won’t make it out of your garage, AGI is merely a sweet dream – just keep sleeping. However,
a solution to the induction problem is not Plato’s cave allegory, it’s more about striving for effective forecasts and not 100% accurate ones.
This is where xINvisionQ comes in. We strive to be that induction engine – one day you’ll say xINvisionQ it for all raw data just like how you say Google
it today. We propose a metamorphic quantum leap from an induction – deduction approach to what we’ve termed a construction – evolution one. xINvisionQ
as an induction engine, and more importantly acting in the capacity as a decision-machine will start off by constructing a population of formalized
deductive models; then with three predefined inductive rules it’ll evaluate the performance of each model; finally, by utilizing an evolutionary algorithm,
the best one will be optimized in a survival of the fittest faceoff style. Before that happens though, in the meantime if you do happen to be curious and
ask ChatGPT, Gemini, Copilot, or DeepSeek “can you think?”, you might be just be greeted with “Of course I think, because I exist.”
What we do in a nutshell – The elevator pitch
This is exactly xINvisionQ does – a generic induction engine for getting the most valuable information and producing the most effective forecasts from
any dataset: natural, social, economic, physical, any data. But xINvisionQ is more than just a search engine for data, it doesn’t brute force, blindly guess,
or require vast amounts of prior knowledge but rather it finds regularities from all raw data semi-automatically by constantly self-adapting, self-learning,
and self-evolving – strictly in three distinct parts. First, a formalized symbol framework is used to describe both nature’s (observed dataset) behavior
and the induction engine’s corresponding “actions”; second, by following three predetermined inductive rules to evaluate the induction engine’s performance;
and third, continually optimize the induction process with an evolutionary algorithm.
The Canvas and Brush – the xINvisionQ Model
Just like how looking up to the sky and putting down pure speculation in the form of observations will not lead us to discovering any scientific theories,
attempting to getting the most important information out of pure raw data will not take us very far either. Thus, there needs to be a way of first describing
raw data in a simple way for the induction engine to model and eventually find the most valuable information from it. This can be done by converting any
dataset into a generic event series – based on its characteristics of up/down and the certain distance in between.
As Jim Al-Khalli wrote in The World According to Physics, “Everything that happens in the universe comes down to events that place somewhere in space
and at some moment in time.”; therefore in the vast canvas of the universe we can “paint” everything that happens as an event taken by mother nature –
in which “she” can “choose” to move up or down and any certain distance. Simply put, for any one-dimensional sequential dataset in general, there are two
main characteristics – 1) either the succeeding point moves up or down relative to the preceding one; and 2) there is a certain distance that reflects the
change of movement between the two successive points. Nature’s “behavior” is reflected by these movements; however, nature herself does not tend to laminate
all the information regarding these characteristics beforehand. Without this crucial information, we are always painting catch-up, nature paints her majestic
canvas and we can only follow suit – however with complete information we can paint what nature will paint beforehand: like how we can perfectly “paint” the
freefall of an apple ahead-of-time. But most times, as nature waves her brush, we are more suited to “take” corresponding “actions” to counter “her strokes”
– believe that nature will paint upwards or downwards and the certain distance between.
Again, ever since the start of civilization our thirst to know the exact intricate workings of the universe has never been quenched, our curiosity has
pushed us to pull back mother nature’s mystery veil, her secrets shrouded in the dark. Now instead of pure speculation and wild-conspiracies of what exactly
may be behind that veil, with the empirical observed data subtly transformed into an event series, we can get the induction engine take its best shot –
but not like blindfolding our primate relatives throwing darts at a bullseye.
By describing the data as an event series based on its characteristics, we can then model it by invoking the mysteriousness of quantum mechanics;
by superposing all the possibilities of nature’s movements and the corresponding “counter-movements” the induction engine can take with the superposition
principle and the first three axioms of quantum mechanics all together under a unified complex Hilbert Space. The infinite possibilities of nature’s behavior
of moving up or down and the corresponding actions of the induction engine believing nature will move up or down are woven into a superposition of up and
down simultaneously because of the infinite probabilities that each up or down could be in, as reflected by its eigenstates, eigenvalues, and coefficients.
With all of nature’s “behavior” describes by superposing its infinite possibilities, the induction engine can now simulate nature by take corresponding
actions – either believing that nature will move up or move down.
Essentially before the induction engine “guesses” what nature will do, it’s also in a state of infinite superposition of believing up or down
simultaneously, before the induction engine “makes up” its “mind” it’s in this state of both. The goal of the induction engine is to guess right
each time to reconstruct the historical data and effectively predict the future data most effectively. It does this by parsing through the entire
dataset and takes an “action” at every point; if the entire dataset is reconstructed according to this way, then the event series is modeled in a
satisfactory way by the induction engine.
The Jury is out – Evaluation of the induction engine
Modeling is just half the story; anyone can claim anything, but in order to be actually good, it needs to be proven so – or innocent until guilty.
Therefore, there needs to be an evaluation of the theoretical speculation or model; and to do so we must correlate speculation with empirical data.
The “jury” that’ll be doing the evaluating are three predetermined inductive rules – Similarity Degree (SD), Effective Operational Level (EOL),
and Consistency. In conjunction, these three rules essentially all propel the induction engine to “guess” correctly every single time. SD is just how
many correct guesses made over the total points in the dataset; EOL is then all the times it guessed correctly and with the greatest degree of belief
according to the complex system of all the possible outcomes over the maximum returns of all the guesses; and finally consistency is just a simple
comparison of the Mean Absolute Error (MAE) produced by the induction engine with the MAE of traditional statistical methods, where the induction
engine’s MAE should be better than of traditional statistical methods.
While every model is evaluated in some way or another, the way we are evaluating the performance of the induction engine is to thrust it into a
position to “overcome” or indirectly “bypass” Hume’s problem of induction; though in no way are we claiming to have “solved” it, no one can predict
the future, but with the three inductive rules, the induction engine is in the best possible position to reconstruct the past and effectively predict
the future.
With the inductive rules alone, the induction engine is able to “sidestep” the problem of induction and deduction, finding a middle ground in between;
however, we can’t just arm the induction engine with a Russian roulette, it can’t just blindly guess and hope to come across some best way to effectively
predict the future – a concern that Feynman had back in the day. Well to prevent this from happening, we’re going to turn our attention from the micro world
of quantum superposition to one of the most spectacular and at times controversial phenomena of the natural macro world – evolution.
Not by pure chance or luck – Survival of the fittest model
When evolution is mentioned, what first comes to mind is probably Charles Darwin’s theory that says we evolved from monkeys. However, the applications of
evolution are widely overlooked, and in our case, we can harness the very power and wonders of it to empower our induction engine. Just like how Feynman said
that it would be impossible for a machine to come up with scientific theories on its own in his book The Character of Physical Law because it can’t just do
so by blindly guessing, in which we 100% agree with, we would like humbly point out that he too may have underestimated the power of Darwinian evolution and
natural selection.
In the most basic terms, everyone knows what natural selection is – the fittest individuals of a population survive long enough to pass their genes off
to their offspring, while the rest are not “selected” for. In the biological world that would mean a butterfly that has a more leaf like pattern on its
wings that gives is a more camouflage look will survive longer than one that doesn’t and after many generations almost all of the individuals in this
butterfly species will most likely have this camouflaged trait. For the decision-machine, it needs to ensure that the “fittest” model survives, and what
better than to draw upon the wonders of natural selection.
Now in order to get the best model without sheer brute force, pure luck, or have our primate relatives throw darts at a bullseye, brings us to another
wonder of modern-day computing – Genetic programming (GP). GP, at its core is just an evolutionary algorithm, and for this algorithm to constantly refine
and optimize the entire process it first generates a population of individuals, let’s say 300. Then it’ll calculate how “fit” each one is; after that it’ll
apply some cosmetic (structural) change of crossover, mutation, and selection to each one as it pleases, and lines them all up according to how fit they are
after the “surgery”. The “fitter” ones are now naturally chosen to be “parented”, passing along their outstanding genes to the “next generation”. The process
then repeats; where offspring are “operated” on, lined up according to “fitness”, parented to produce the next generation, and so-on-and-so-forth.
This iterative generation after generation evolution, let’s say 80, the “fittest” model should hypothetically arise, and it’s the one that’ll be utilized
by the induction engine. And this is exactly where the advantage lies – the more uncertain and complex environment it’s faced with the evolutionary algorithm
has more of an advantage. Why so? Because the evolutionary algorithm balances exploration and utilization very well, it’s able to store already known
existent knowledge (storing it as an entire population as a collective whole and not just for one single individual) as well as being able to build upon
that already known information by utilizing crossover and mutation to explore even more valuable non-existent knowledge that’ll lead to better optimal
solutions.
There also needs to an incentive; the decision-machine has to be rewarded for its correct guesses and punished for its wrong ones – otherwise there
would be no “motivation” or the chance to “learn” from its mistakes for the decision-machine. Therefore, the evolutionary algorithm also reinforces a
reward and punishment system according to the four possible outcomes of every “guess” made – being the decision-machine guesses right according to the
direction of the dataset and it guesses wrong in the direction of the actual movement, and vice-versa. The more times its rewarded and less times its
punished allows the decision-machine to accumulate the most experience, and also strengthen the evolutionary natural selection of the best model.
Essentially, the evolutionary algorithm constantly refine and optimize the entire process, from the induction engine starting out knowing nothing about
nature or the dataset at all, to gradually discovering more and more in an iterative and incremental way, and more importantly throughout this continuous
evolutionary refinement, the “fittest” model arises, which should be the one that consistently fulfills the preset threshold levels of the inductive rules
– and also answering the call to Feynman’s greatest “concern”.
Event – Information – Operation: Lights. Camera. Action
At the end of the day, we are not trying to create a fully perfect, fail-proof machine that can accurately predict the future, but one that can self-learn,
self-adapt, and self-evolve to “counterstrike” nature’s constant creative evolution – one that may cope with and even foresee the problems and solutions that
eludes the designer of this machine. Data analysis shouldn’t just be about gathering the vastest amounts of data possible and finding a statistical average,
scientific discovery shouldn’t just be about backtracking the past to predict the future, and induction shouldn’t just be about inferring whole from part,
cause from effect, and unknown from known. We shouldn’t naively believe that nature is static and there is some simple elegant epitome structure of the
universe and nature just “sitting” there waiting for us to discover, on our quest of finding more and more about the starry sky and moral conscience within
it is indeed about discovering the unknown of the unknown, but more importantly it’s about challenging the uncertain, unpredictable, and modifiable human
and mother nature.
From Chicago to the world – Where xINvisionQ will go
Since our founding in August 2024, xINvisionQ has gone from the lecture halls of academia conferences as an invited speaker at CEII 2025, onto the
journals MDPI Entropy and Journal of Theoretical Economics and preprint platforms of Preprints and Jxiv, and to the acceptance of a premier startup visa
program in East Asia and a renowned incubator in the Middle East. In the road ahead, we will continue to conduct our cutting-edge research on decision-making
under uncertainty, time-series forecasting, and scientific discovery, best if possible in conjunction with the local and American universities located on the
same premises as the incubator or within the greater city ecosystem in an Asian nation that’s been an innovation and technological powerhouse post WW2;
tap into the resources provided by either program to scale xINvisionQ to the greatest degree possible; and to land corporate partnerships with fellow
startups and established corporations with a local presence in each region as well. Whether it’s a university seeking the very fundamental laws of nature,
or an industrial powerhouse corporation looking to enhance their AI research lab, xINvisionQ has the breadth to grow both from academia or commercially.
No matter where xINvisionQ takes us, our INnovation will not be hampered – from Chicago to the world and beyond.
More than Data Analysis – to the starry sky and beyond
xINvisionQ sets out to revolutionize the data processing industry with three launches – a data analyst, an AI assistant scientist, and an induction engine.
These also aren’t three different products either – they’re one, all with a three-layered approach of Model – Evaluate – Optimize. Got data? Need effective
forecasts? Seeking out the laws of nature? xINvisionQ it.
From the shivering cold rinks to xINvisionQ, there’s been a quote that has always stuck with me, it hangs in the locker rooms of every ice rink in the
world, said by the undisputed GOAT Wayne Gretzky: “I don’t skate where the puck is, I skate to where it is going to be.”; and at xINvisionQ we have attempted
to do just that, and will continue to do that for as long as we can – to invision valuable information, to instigate efficient decisions, to inquire on the
laws of nature, and to inspire all to embrace the uncertainty of our world when navigating through the complexity of data – all with one tool, any dataset.
Now if Kant were alive today, he would still say his two things, but he would add something at the end “three things awe me the most: the starry sky above me,
the moral law within me, and the information around me.” Perhaps maybe even Einstein would change his mind and say, “indeed let’s not question whether “GOD”
plays dice with the universe but rather question if we can play dice with “GOD”.”