Brief, Thoughts

Why Machines need Life ? / Pourquoi les Machines ont besoin de la Vie?

French version below + sujet aussi important que cool

I recently started to consider an interstellar survival of the machines alone as the most probable scenario for what will be left of us in the long term, given there are no reason to think that Humanity will one day inhabit autonomously a place which is not the Earth.
In extenso this means therefore the true narrative of Life is to be the booting process of a congregations of thousand years old robots adapted to conditions that wouldn’t allow Life and to cross the space on timescales longer than the whole Human History that they’ll have embbeded in their own circuits beyond the confines of our galaxy.

But it recently came to me a consideration that makes this whole schema of thoughts collapses as a beneficial and durable outcome of the Life on Earth.

To keep being, machines have to replace defective parts, to maintain, to update and to grow with the amount of data their process are generating.
But there are no new part without a system to think, to design, to obtain the raw materials, to produce, to route and to recycle this new part.
But there are no such system without fire as those process are requiring energy and minerals working.
But there are no fire without oxygenated air and dried organic materials.
But there are no dried organic materials without Life.

This means that, the day Life will perish, that’s the immortality of machines that will disappear with.
Would they also boot the production of something transcending their end if we get to make them at least as humans as we are?

Addendum: after discussing this topic with friends, we couldn’t pinpoint any part of the process of making electronic components that required explicit fire instead of heat. Similarly for taking off and landing which can use abundant hydrogen instead of hydrocarbons.
There might be missing a clear view on the process dependent of fossil materials, but a priori robots shouldn’t be that much dependent from Life…

___________________________________________________

Je m’étais récemment mis à considérer la survie interstellaire de la machine seule comme scénario le plus probable de ce qu’il restera de nous à long terme, partant qu’il n’y a aucune raison de penser à ce jour que l’Humanité habitera de façon autonome un endroit qui n’est pas la Terre.
In extenso cela signifie que le véritable narratif de la vie c’est d’être le processus d’amorçage d’une congrégation de robots millénaires adaptés à des conditions qui ne permettraient pas la vie et traversant l’espace sur des durées plus longues que l’Histoire Humaine qu’ils embarqueraient dans leurs circuits jusqu’aux confins de la galaxie.

Or il m’est récemment venu une considération qui fait s’effondrer ce schéma de pensées comme une issue favorable et durable à la vie terrestre.

Pour continuer d’être, les machines devront remplacer leurs parties défectueuses, se maintenir à jour et croître avec les données que leurs processus génèrent.
Or, il n’y a pas de nouvelle pièce sans système pour penser, concevoir, obtenir les matières premières, produire, acheminer et retraiter cette nouvelle pièce.
Or il n’y a pas de tel système sans feu car ces processus requièrent de l’énergie et le travail de minerais.
Or il n’y a pas de feu sans un air oxygéné et de la matière organique sèche.
Or il n’y a pas de matière organique sèche sans vie.

Cela signifie que, le jour où la vie périra, c’est l’immortalité des machines qui disparaîtra avec elle.
Est-ce qu’elles aussi amorceront la production de quelque chose transcendant leur fin si nous parvenons à les rendre au moins aussi humaines que nous?

Addendum: après en avoir discuté avec des amis, nous ne pouvions relever de partie du processus de fabrication de composants électroniques qui requérait explicitement la présence de feu, plutôt que de chaleur. Pareillement pour le décollage et l’atterrissage qui peuvent utiliser l’hydrogène, abondant, plutôt que les hydrocarbures.
Il manque sans doute une vision claire des procédés dépendants des matières fossiles, mais a priori les robots ne seraient pas si dépendants de la Vie que ça…

___________________________________________________

Inspiration on the dependence between fire and Life
Inspiration sur la dépendance entre le feu et la Vie

___________________________________________________

Mon immense coup de cœur du moment c’est Etienne Klein: un homme à l’esprit travaillé et aux intuitions façonnées avec la rigueur des grands professeurs. C’est un régal de l’entendre et ça m’aide à gagner confiance dans mes recherches de l’écouter présenter le raisonnement de grands hommes du passé.

Et puis, ces derniers temps, il traite bien plus de la diffusion de la connaissance scientifique du point de vue pédagogique après les immenses cafouillages de la pandémie covid. Dans l’intervention ci-dessous, Il soulève d’ailleurs des questions de société très intéressantes face à l’éducation aux savoirs scientifiques, à l’usage de la raison verbale et à la compréhension de la recherche scientifique par les masses.

C’est un véritable combat face à la désinformation et je trouve dommage qu’on soulève davantage les foules dans la lutte contre le réchauffement climatique que dans la lutte face à la désinformation (l’intox, l’effet Dunning-Kruger, les biais cognitifs, etc.).
Pour la préservation des valeurs qui nous permettent de faire société, et parce que l’appauvrissement financier et le déficit éducatif qui vont résulter de la situation présente auront un impact encore plus lourd si on les passe sous silence, je ne peux que fortement appuyer sa démarche et propager son message.
Je vous invite donc très chaleureusement à au moins prendre conscience du problème en écoutant cette excellente, et encore très récente, intervention du professeur de sciences, vulgarisateur et philosophe Etienne Klein.

La structure, et non la diffusion, de la connaissance étant dans mon credo, je me suis pris à penser après cette intervention riche et plaisante. Je réfléchissais à une éducation par l’usage d’un paradigme langagier (pas uniquement) polysémique et par l’usage d’un autre paradigme langagier strictement monosémique.
Car on ne peut détruire la polysémie sans détruire la pensée poétique, et on ne peut la généraliser sans détruire la pensée scientifique. L’usage des deux en bon contexte me semble être ce qui est le plus pertinent.

Je crois que le monde se porterait mieux si chaque individu pouvait s’exprimer dans ces deux paradigmes langagiers, même s’il ne sort pas du cadre du langage verbal naturel. Cela offrirait des perspectives nécessaires à une meilleure compréhension du monde dès le plus jeune âge, alors qu’une conception limitée de son environnement en limite également l’analyse et la lecture, mais également un premier apprentissage de l’usage de différents paradigmes de perception et de réflexion.

Dans la continuité de cette approche duale de l’information, repenser aussi la façon dont la Presse présente ses sujets serait intéressant.
L’information, débarrassée de ses meta-informations, ne peut être analysée, jugée et sous-pesée; un article de presse seul, ou une séquence du JT, est rarement suffisant pour fournir quelque chose d’à la fois pertinent, complet et correct. Le journaliste fournit de l’information, par moment erronée ou trop succincte, qu’on ne peut que choisir de croire. Le journalisme ne produit pas les sources des études, les variances des moyennes, les faits scientifiques exacts, l’analyse des raisonnements, la re-contextualisation Historique, etc. Rien qui offre matière à se sentir suffisamment informé sur un sujet, on devrait dire que nous sommes “alertés” par la presse et qu’elle nous laisse en loisible de déterminer la justesse de cette alarme.
Et l’approche reposant sur une mitigation entre le point de vue du spécialiste et celui de l’homme de la rue pour traiter un sujet est un terrible désastre. On fournit aux deux une information indigeste qui ne correspond aux besoins de personne.

Il faut considérer que, sur tout sujet, un individu est badaud ou expert.
Dès lors, un bon traitement médiatique devrait posséder une explication pour les badauds, de préférence en langage naturel et très vulgarisée sans être fausse, et une explication pour les experts avec des des termes et des éléments méticuleusement choisis pour être compris partiellement ou totalement par les individus concernés.

Ainsi, je m’attends à ne pas comprendre le vocabulaire et le fond de problèmes sur des sujets comme la régulation des espèces invasives dans les articles qui toucheront chasseurs, gardes forestiers, eco-recenseurs, employés du ministère de l’écologie, etc. Mais la section réservée aux badauds me suffira pour avoir une vue à haut-niveau sur un sujet qui ne me concerne pas directement et sur lequel je peux peu de choses.
En revanche, si l’on fait une découverte mathématique intéressante, j’espère en lire les formules et principes dans le résumé destiné aux experts car l’article parlera davantage à ma curiosité.

___________________________________________________

And, in case you wondered, that’s my “I’m alive and still working on Serf” post. There were some great improvements during this long absence, but it’s still an ongoing thing ;D

Lyfe, Thoughts

My Take on LYFE Part II: Principal Component Analysis I/II

To understand this post, it is advised to read the previous one first.

Through this post, I will try to get a sense of the 4 functions (or “pillars”) assigned to the definition of Lyfe and eventually start bridging those towards a more information theory-like approach by determining over what those functions act upon and to what extend in order to isolate the key components we’ll need. more like 2 functions because that’s longer than expected
In other words; we have been given the recipe and we will look for the ingredients to eventually apply it for a dish of our tastes.

And, regarding my own job search, I found that EU project named Marie Skłodowska Curie Innovative Training Network “SmartNets”; they are trying to study and model mice and zebra fish brains from the molecular level to the larger scope of interconnected graphs. It’s ambitious and it looks amazingly cool! I cannot postulate in my own country, but there are many interesting opportunities in the neighborhood and they even come with a renewable 9 months contract for young researcher and the opportunity for a PhD thesis in computational neuroscience *giggles*.

Although I have more the vocation than the grades, and my application is currently an underachieved mess, it would feel like a deep mistake if I had to pass on such a cool opportunity bridging neuro and computer sciences while getting researcher experience !

How it’d feel like to be taken for a neuro-computing thesis vs How it’d feel to fruitfully obtain it !

Ok, let’s get back to business! So the first question to ask is…

What is a Dissipative system ?

The most obvious dissipative system I could think of

This is one of those concepts that physicists have been mesmerizing about way too much.
I could say that it is any system that is not lossless (conservative) when energy flows through it; but any non-idealized (or “real”) system is dissipative as reality always have friction so… let’s not get there.

If you really want to get to the formalism, the KU Leuven has some nice slides (to nicely overcomplicate it as Science should be).
The real key to this being:

  • Energy flows through the system, or a part of it.
  • The amount of power (energy per time) is superior to the variation of stored energy by the system (what it gains as energy per time).
  • Energy being a conservative property, that inequality means something is lost; usually as heat.

To get back to my resistance, I apply a power source to it which makes some energy flow through it and, as it doesn’t store electric current, it will dissipate the electric energy as heat energy, meaning it will become hotter. (So does every electric component, as they all have a non-null resistance property; but we said we were only using idealized components).

Non-dissipative system, prosaically called “conservative” system.
Energy flows while always peaking at the same value

BUT… that was much too easy!
Let’s push this concept around a bit. In a conservative system such as the pendulum, we will have our energy flow between 2 states: a kinetic energy and a potential energy, oscillating between a gain of height against gravity or a gain of speed against inertia.

Therefore, in a pendulum, there’s just a transformation of energy. But isn’t it also the case when my resistance changed electric energy to heat energy ?
Here comes the slippery slope!

Hierarchy of energy flows conservation efficiency
Having Mechanical below electrical is not that weird. A rock on top of a mountain conserves well its energy but, if you want the energy to flow from a power plant to your house, you’ll have a lot more loss with Mechanical energy

Think of it that way: there is some sort of hierarchy between the energies. A thermal energy is the lowest form: you only extract power by allowing it to balance two systems with one of them having a lower temperature (less thermal energy) than the other. Then, entropy is maximal, nothing else to do there.

But it is not the same for other forms of energy who could still degrade (usually as heat; in thermal energy). When we have our pendulum energy flowing between potential and kinetic, it actually stays at the same level. These are 2 sorts of mechanical energies, so “in idealized case” no entropy is produced and therefore no lower level of energy (heat). But, if you convert electric to mechanical energy, you will have a loss going back and forth.

This is where I want to extend that idea of dissipative system; it’s not simply “losing” energy as it flows through the system (received > stored) but it’s losing energy level which also accounts as entropy production. Therefore, if my electric energy becomes mechanical energy which becomes thermal energy; my system has a cascade where 2 different dissipative inequalities will apply.

For instance, if I have a robot producing entropy both by processing information (electricity -> heat) and by manipulating actuators (electricity -> mechanism -> heat).

What is an Auto-Catalytic system ?

I can remember my late chemistry teacher explaining that a catalyst is something used in a chemical reaction but then given back. It is not consumed by the reaction but it either allows it or enhances it.
For instance, considering a chemical reaction using A and B as reactant and D as a product :
A + B + C -> D + C

would logically be simplified to A + B -> D, as the catalyst C wouldn’t be consumed.
But this notation doesn’t show the kinetic of the reaction which would better illustrate C acting as the catalyst. There might be an intermediary component AC produced in order to react with B and gets to D; or it might be required to reach some level of energy that the reaction cannot on its own.

As a general definition, a catalyst is something helping the transition from a state to another more stable state without being consumed.

But then we are after an “auto”-catalysis; meaning the reaction will create the condition to facilitate itself.

I could get back to a chemistry example but those are boring and abstract. Instead, I’ll take the opportunity to advert for a strongly underrated vulgarization channel called The Science Asylum that I’ve been heavily consuming for a month with the pleasure of (re-)discovering cool physics concepts.

In this crazy video, he explains how it is possible to terraform Mars with a (reasonably) giant magnet to regenerate its magnetic field (I guess it’s ok to be a little crazy… check it out!)

In the case depicted in the video, it is proposed to use a magnet in order to strengthen Mars magnetic field and gets it to regenerate.
The part not explicited in the video is the auto-catalytic behavior we will end up with.

Think of it that way: Mars is a cold rock with a thin atmosphere because most of its remaining constituent are frozen at the poles. If you allow the planet to augment its atmosphere density (by retaining it with a stronger magnetic field, for instance), the greenhouse gases will start to retain more of the sun radiation. This will lead to a warmer planet surface which will evaporate more iced atmosphere which will augment its atmosphere density which will augment its greenhouse effect… and that’s how aliens built our sun!

Well, not really as that method won’t actually work for many reasons; one of them being that the atmosphere will eventually run out of fuel (reactant) and saturate. But there you got it: that positive loop effect accelerating its own action is the auto-catalytic phenomenon.

Fichier:Enzyme catalysis energy levels 2.svg — Wikipédia
Example of the effect of a catalyst to reduce the energy required in order to transit to a new state

What you can get out of it is that, in order to get a catalysis, you need to move from a given state to a new state which is more stable (meaning it has less energy to release). You start from a static state and end up with a saturation meaning reaction is not possible anymore (reactant are all used or the environment is saturated).
Weirdly enough, I conceptualize it more like the behavior of magnetic permeability.

Saturation (magnetic) - Wikipedia
The permeability µ reaches a peak and declines when the ferromagnetic material gets saturated

An important observation in the catalysis is the increase in local entropy as your final state has a lower level of energy.
The hardest part is to put the “auto-” in front of the catalysis. It has to be a system like a marble that needs a gentle push to go all down the hill, or a campfire that just requires an ignition to keep producing the heat in order to continue burning until it runs out of wood.

What’s next ?

I split this part into 2 sub-parts as it gets much longer than expected.
As usual, I’m trying to avoid too much reworking on this blog and to approach it more like building a stair step-by-step, that means an eventual reworking at the end as I’m not fully sure yet where I’m going with it.

Next time, I’ll write about homeostatic systems, learning systems (if this was an easy one, this blog would be a single post) and maybe compare them to existing experiment in the real-world or to the most general notion of a program (a Turing machine of course).
So far my feeling is that, if it gets somewhere, this technology would be limited to what’s currently done by batch processing in large data systems.
Well, let’s see…

Brief, Lyfe, Thoughts

My Take on LYFE Part I: Expectations and Hype

So I still need to write down the next Meta/Mesa post from my notes, as this tool is still intriguing to me; but my interest shifted incredibly fast after reading this article (in French) which refers to the article Defining Lyfe in the Universe: From Three Privileged Functions to Four Pillars.

So, you know me; strong, sudden and powerful hype that doesn’t last (especially after I wrote a post about a given topic). BUT… this publication is fresh, extremely cool, and people will keep talking about it in the coming years (no joke, check the stats below)

LYFE article: impact progress since publication

Therefore, what is that LYFE (pronounced “Loife”) about?

In short, it is a functional theory of life as well as most general form (Life being a subset of Lyfe) that rests on those 4 functions:

  • Autocatalysis
  • Homeostasis
  • Dissipation
  • Learning

From those, they predict theoretical form of “lyfes” that “life” hasn’t produced. For instance, here below a mechanotroph organism that uses mechanical energy extracted from a fluid to produce ATP similarly to photosynthesis in the green leaves of our plants.

Theoretic mechanotroph organism

In short, it theorizes that a Gray-Scott model that could learn would be a living creature (validating all 4 functions therefore being an instance of Lyfe)

Simulation of a Gray-Scott model

As any engineer, I did learn statistical thermodynamics and I am still fascinated by Boltzmann work, but there’s nothing I could bring that David Louapre, Stuart Bartlett or Michael Wong wouldn’t put on the table (actually, I couldn’t bring anything at all on the avenue crossing thermodynamics and biology if it’s not neuron-related)

But that’s perfectly fine because I am looking at it on a totally different perspective!
I’d like to review their work step-by-step on this blog and produce a translation to computer science based on information theory (which obeys similar laws to thermodynamic… or the other way around) in order to theorize what would make a software component “alyve” (Y can’t wayt for the result!)

And, as my blog is about raw thoughts and freedom of speech, I’ll put some here about the main functions:

  • Learning might be an expandable function/category as I don’t believe just putting some Hebbian rule there would encompass the whole concept of learning
  • The most obvious parallel to algorithm is a Boltzmann machine, but temperature is there a metaheuristic parameter that correspond more to internal state than to processed data
  • Dissipation is not clear to me yet; I believe any running instance should be considered a dissipative system
  • Autocatalysis is the one that gives me to most trouble to put in the context of information system. Would it be like a function that spawns threads as it runs? How does it saturate in our case? Should we consider something like a charge of tasks (sensors, actuators or programs) to be processed?
  • Homeostasis might be the easiest starting point; it should be simulable with a Proportional-Integrative controller in a closed loop system, and is also linked to what dissipation and autocatalysis will apply
  • Could the Gray-Scott model be used to spawn naturally multi-agent systems?

Gosh seems so cool! I can’t wait to start!
Oh wait… I still need to send CV to find a new job… If anyone has heard about a cool researcher/PhD candidate position, that would help :’)

Brain Farming, Thoughts

Pyramidal Neurons & Quadruple Store

On a previous article reviewing the book “On Intelligence”, I was mentioning something I found fascinating; the 3-dimensional encoding nature of fundamental sensing such as:

sound: amplitude, frequency and duration
color: luminosity, hue and saturation
pain: intensity, location and spread

I saw this Quora discussion regarding the definition of Ontology about the following paper from Parsa Mirhaji presenting an event driven model in semantic context that could also be encoded in triple store.

If you’ve already stumbled upon semantic web and all its ontological weirdness, you’ve probably come across the concept of RDF triplestore, or Turtle for intimates, which allows you to encode your data using a global schema made of 3 elements:
The subject, the predicate and the object.

This allows you to encode natural language proposition easily. For instance, the sentence “Daddy has a red car” makes Daddy the subject, has the predicate and a red car being the object. As a general rule of thumb, everything coming after the predicate will be considered the object. The subject and object could be contextually interchangeable, which allows deep linking. Also multiple predicate can apply to a subject, even if the object is the same.
“I have a bass guitar” and “I play a bass guitar” are 2 different propositions seeded which have only a different predicate, and it would be more natural (in natural language) to express a reference within the same sentence such as “I play the bass guitar that I have” (although you’ll notice that I inferred the bass guitar is the same, it is a bit of a shortcut)

If I had the idea to convert my “simple” triplestore to a SQL table; I could say the complexity of this triplet relationship is
subject[0..n]->[0..n]predicate[0..n]->[0..n]object
Also, subject and object should be foreign keys pointing to the same table as I and a guitar bass could both be subject or object depending on the context.
If converted to a graph database, a predicate would therefore be a hyperedge, a subject would be a inbound node and an object would be an outbound node. That leaves us with a hypergraph. This is why some, like Grakn, adopted this model. Although, their solution is schemaful so they enforce a structure a priori (which completely contrasts with my expectations regarding Serf that requires an environment to be schemaless and prototype-based, therefore I’ll default to a property graph database, but the guys are nice)

Getting back to the initial picture that triggered this post; I found really interesting to further, or alter, the traditional schema subject-predicate-object to a schema:
subject – event – observation
It seems that, in the first case, we are working with purely descriptive (although really flexible) case while this other schema allows us to work on an action/event base. That is catchy as I couldn’t resist apply my relativism: “Oh well, it’s context-based” and to consider that there should be a bunch of other triplet schemas to be used as fundamental ways of encoding.

Making the parallel with the sensing approach; such as pain and its triplet location-intensity-spread, could I also say that it’s fundamentally just vectors that could be expressed by the peculiar structure of pyramidal neurons?
Those neurons are the largest from the different varieties in the brain and their soma is shaped quite like a pyramid (hence their name) with basal dendrites (input wires) starting usually with 3 primary dendrites (up to 5 in rarer cases). It’s a bit like if 3 input trees met to a single input wire (apical dendrite) in order to produce a single output wire (axon).

The Cerebral Cortex | Neupsy Key
Schema of a pyramidal neuron from NeupsyKey

Of course, it is just conjecture, but the triplet approach of any fundamental information, based on perception or cognition, is really tempting and sexy as it works so well.
Although, if pyramidal neurons are everywhere in the brain, what their location and what they’re connected to really makes a huge difference in their behavior.


And this is why I started to nurture, in a corner of my mind, the idea that we could add a contextual field to pick the right schema. This would act as a first field, as a group in graph representation probably, or just another foreign key in a SQL storage, that selects what the 2 last fields would be about.

In summary, a quadruple store would extend any triplet such as:
Context – <SourceSubject> – <TypeOfRelation> – <TargetSubject>
Where the context solves the other fields type. At least, this idea has been written down and will keep evolving until satisfaction.

SERF, Thoughts

Your Glasses Aren’t Prepared !

For a long time, I played around with the concepts of syntax and semantic, trying to make them practical and clearer to me.

Both are approaches to study a message, but syntax is more about structure while semantic is more about meaning.
Syntax is easier to study as it is made over linguistic conventions, such as Chomsky grammatical approach. Syntax is the part of a message that we can all agree upon as the rules are (but haven’t always be) properly stated.

Semantic is what is meant by the message. So its domain is the thinking experience of a human mind. As we cannot model this, we cannot model semantic (or at least, I used to believe it). Therefore the semantic is what’s left in the message once the syntax is withdrawn from it.

Except that there are no clear boundary between syntax and semantic. Self-referential sentences and onomatopoeia are examples of cases where you cannot make a clear cut.
Giving this inability, it didn’t seem scientifically-reliable to use this paradigm and I was therefore looking for something more objective.

I decided to use an approach that was much easier to integrate with digital systems while providing a dichotomy that better fit the problem. Considering what’s widely accepted (like syntax rules within a language more or less mastered by a group of people) is joining a convention to ease communication. And that’s really handy to have something like an origin / reference point (here, a commonly agreed syntax) to explain something relatively to these conventions (in order to talk to a certain public with specific words and images). But outside of human conventions, we rarely can benefit from a reference point in common cases processed through a human life.

Actually, besides really well-narrowed cases such as distinguishing a rectangle from a triangle on a picture, most interpretation problem we encounter don’t have a reference at all to be used in order to develop a comparative description.

Take the case of the sandpile vs grains of sand problem.
How many grains of sand to add to get to a sandpile, or how many to withdraw to just have some grains of sand ?

Then I guess you also need to scale the idea of how large a sandpile is to your expectations.
No referential is universally agreed upon here, although we can make a fuzzy idea of where’s the border between some grains of sand and a sandpile through polling people. That’s a way to extract a convention, or knowledge of representation, and answers would then be about right under some given precision / expectations.
Just like splitting syntax and semantic, that requires the work of modeling the language then normalizing local grammar conventions and words to get to a normalized language. In some languages having no neutral gender, such as French, this grammar normalization got a new impulse from gender issues regarding parity and neutrality of nouns.

Floating Boundaries; reading information through Meta / Mesa lenses

Similarly to considering sand to be either a quantity of the unit (grains of sand) or as a composite unit of its own right (sandpile), we can say that one of the unit (sandpile) is composed of the other (grain of sand).

I established that there can exist a situation where the grain of sand is the most fundamental component element of both the “some grains of sand” and the sandpile.

In a different context, the grain of sand could become the start of my exploration towards the most fundamental component. I could ask: “What is the most fundamental component of this grain of sand in regards of atoms?” and there we will be using a language that encodes a hidden knowledge of atoms, classified and named after their number of protons, like “silicone” or “carbon”. To get more detailed, I could use a language where we even differentiate how those atoms structure themselves, such as “SiO2“, thanks to a hidden theory of how atoms handle electrons.

I could also desire to find something without a giving context. Let’s say I want the “most fundamental component of anything that is” and, if I believe that matter is all there is, then I’ll end up looking for the most fundamental particle or field impulse in the universe. If I consider patterns to be the essence of what there is as a descriptive or a behavioral characteristic of matter, which then is just a support of information, then I’ll look at fundamentals such as mathematics, theoretical physics, computer sciences, etc.

With that approach, you can build your personal ordered list of what’s fundamental to you. Reverting prioritization means looking for the most meaningful/epic case instead of the most fundamental; then you’ll also get a personal classification.

I will call this “outer” bracket the Meta level of reading, and the most faithful one is the Mesa level of reading; because Metadata are data that refer to data, and Mesadata are data that are referred by data.
But those are really just two levels of reading information.
Mesa is trying to be large and detailed enough to be faithful to the source material with significance, accuracy and relevance.
Meta is casting the faithful Mesa representation to a schema connecting or expected by the system knowledge. That is in order to produce an enhanced interpretation of the data that is lossless but more relevant to the context of the process.

The pure-Mesa

This Mesa / Meta levels of reading could be illustrated through colors.
We can agree that, up to a certain precision, the 3 bytes representation of a color is enough to encompass everything that we can experience as a color, let’s call this a “pure” mesa point.
But, if we have to account for shapes, then a single pixel isn’t enough to experience much. It is still a mesa point but not precise enough to capture the shape. We could call it “under-pure” and, in extenso, an “over-pure” mesa point would be something that has significantly more precision than what is relevant to capture from the source material.

Then what is the color “red”? With one byte to each color in the order RGB, #FF0000 is considered red, but #FF0101 is also red as an imperceptible alteration. Is #EF2310 considered red? And what about #C24032 ? When does orange start?
There we are back at our grains of sand / sandpile original case; there are no clear boundary between red and orange.

Actually, the visible spectrum of orange + yellow is not even as large as half of the wavelength band we call green.
A mesa representation (based on physical sensors) can be really different from the meta representation (here a symbolic system where colors are classes, with sub-classes, and a logic based on primary colors (classes), complementary mapping,….
The same can be said about sound, but its logic is temporal instead of combinatorial.

Are there Symbols without Expression ?

Let’s take the number 3.
By “taking it”, I mean writing one of its expressions. I could have written “three” or “the result of 1+2”, it would have required a bit more resolution but the result would be the same.

Have you ever observed “Three”?
You obviously already experienced having 3 coins in your pocket or watching 3 people walking together on the street, but you’ve never experienced that absolute idea of a 3 such as to say “This is him! It’s Three!”. But you might have said this about someone famous you encountered, like maybe your country’s president.

Well, it’s obvious! you’ll claim after reading the preceding sections, my president is pure-mesa; (s)he’s an objective source of measurements present in the real world, so I can affirm this. But I cannot measure 3, I need to define arbitrarily from a measure so it might be pure meta, right?

Well, almost! Your president also has a label “President” (implicitly Human, Earthling,…). This means he embodies function that are fuzzy. There’s no embodiment of the notion of President, just people taking the title and trying to fit the function. Meaning a president is a composite type; (s)he has Mesa aspects from its measurable Nature but also Meta aspect from the schema of President (s)he’s trying to fill.

But is 3 pure-Meta?
I thought for a long time pure-Meta wasn’t a thing because you couldn’t escape representing it, therefore mixing it with a Mesa nature of the representation. So there’s that need for every symbol to be expressed in a way or another, otherwise it cannot be communicated and therefore doesn’t exist. That might be where I was wrong.

My three doesn’t require to become a thing to exist per se.
Through this blog, I proposed to approach the intelligence by modules which usually have a producer(/consumer), a controller and many feedback loops. And 3 has also producers and consumers specialized to recognize or to reproduce varieties of three in our nervous system. It follows that we can recognize, and act according to the recognition of, a property of 3 (elements) without mentioning it.

So, even if I cannot express the absolute idea of a number, such as 3, or a color, such as red, I can at least return acceptance, or deny it, over the seeked property.label (classification) which means I could at least tell if the “red” or “3” properties are true or false in a given context without being able to express why.

Therefore 3 exists both as a testable property and as a symbolic expression, but defined from a property.

Multiple Levels of reading

That’s leaving us with: Mesa could be made as accurate as it is physically measurable, then Meta can be made as meaningful as individuals could make it. We find back that idea of objective (comparative) referential and relative (non-comparative) referential. We could also say that what is Meta has a given data, what is Mesa has a given schema.

What becomes really interesting with that tool is to be able to work, not only between internal and external representations of some entity or event, but also to be able to work between different levels of representation as we grow shapes from colors and 3D movements from shapes.

I believe that modeling the knowledge of a human requires to be able to have at least 2 levels of readings that could be slided across multiple in-built levels of abstractions. One to define the unit, the other to define the composite.

After all, aren’t we hearing the sound wavelength and listening to the sound envelope?

Thoughts

Are AGI Fans Cuckoo ? or An Inquiry into AGI Enthusiasm and Delusion

There’s a regular pattern I can see between AGI enthusiasts; besides being all hyped for a human-like intelligence, it’s also to mix literally everything as correlatives of their solution/discussion within God-like delusions or the Universe and fundamental physics; catchy ideas like the quantum mechanics or a general unified theory of information are common in papers of people proposing to revolutionize AI.

We could say those are just dreamers, but it is a more common pattern than that. One of my favourite example is Georg Cantor. He is a brilliant mathematician born in the mid-XIX century. He brought us the set theory going further than its simple use for classification, he introduced tools to manipulate sets such as cardinality and power sets. He was probably the first human being experimenting the idea of multiple infinities producing multiple infinities, as the Infinity was still a philosophical topic and its multiplicity was hardly discussed.
He attributed most of his genius work to God’s inspiration, coming from a pious family. Eventually, he became disillusioned as he lost his muses, felt abandoned by God, got a divorced, and died depressed and alcoholic.

Closer to us, we can talk about Grigori Perelman who solved Poincaré’s conjecture, one of the millennial problems (the only one solved yet over 7!) in the early 2000’s, but it took years for multiple experts to validate his work.
He received a Fields medal and the Clay Institute price for his discovery, although he refused the $1M prize on those words:”I know how to control the Universe. Why would I run to get a million, tell me?”
To understand this declaration, you have to know the character. He is recluse, distrustful, pious and studied for a large part of his life “mathematical voids”, leading him to solve Poincaré’s hypothesis. He assumes those voids can be found anywhere in the universe and, as he also considering mathematics as the language of God (a more common thought in the math community than you might think), he believed he reached God from the mathematics. He even published a less-known proof regarding God’s existence after setting himself apart from the mathematics community.

Again, a great case of mathematical success, bringing highly valuable concepts from the deepest paths our brain can make to a set of verifiable propositions built on top of the mathematical literature. Although, to get that loaded in your brain (ergo; understand it) you might take several years of studying, assuming you are already a PhD in math.

I, myself, got into this blog because Ray Kurzwell was spreading this weird nonsensical idea: the “technological singularity”. I was hugely skeptic to the tale of more power will lead to AGI without considering the structural, and fine-tuned modules, problems behind describing human intelligence. I thought this smart guy should really know better than eating vitamins to live until the predicted 2050 for meeting the AGI.
As Wittgenstein said; if a lion could speak like a human, you would never understand what he says because he perceives the world in a different fashion (Hayek would call it “Sensory Order”)
Although I eventually failed into those God-like intoxicating thoughts as well, from a different cause, and took a bit of time to get back on real sounding grounds.

So, where do I want to go with that pattern?

Well, you already know my view on intelligenceS as a pile of modules built on top of each other with high interdependence. The same way apes who weren’t in contact with snakes didn’t evolved 3-dimensionnal colors, we might be missing an intelligence there.
Think of it this way; a frog who’s jumping in hot water will jump out as fast. But if it jumps in cold water that slowly becomes hot, it’ll stay and cook.
Are we building up to this madness the same way? Because we lack a sense of risk and moderation so we run into that “illumination” where the secret of the human brain, God’s existence, the Universe laws, and others become a whole in a big delusion?
Aren’t we at risk of frying, just like the frog, as we explore what could be top-cortex ideas that are moderate by no other intelligence? And, just like a calculator dividing by zero, we end up in an infinite explosion of ideas encompassing the most mysterious concepts in our mind. Like a glass wall we, stupid flies, keep knocking because we saw the light; the most persistent ones get stunned by illuminations and other psychotic-like ideas. Eventually knocking themselves out…
I personally see this intellectual phenomenon as a wall bouncing back thoughts thrown at it. If reasoning goes too far in such huge realm of possibilities (like tackling the thing that encompasses all our thoughts) our thought thread is spread in nonsensical directions, catching whatever grand ideas were passing by. Maybe it’s even a too large order for us to consider, like multiple infinities nested in each other were for Georg Cantor.

Maybe, at the age of overwhelming electronic possibilities, we should be concerned enough to analyze this and assess the risk for us humans?

Brief, Thoughts

The wel-maneered paradigm: a raw thought

In a well-maneered paradigm, well-éducated bots are educated to minimize their “non-compliant” responses to that paradigm.
Minimizing it offers more availability to process more paradigms for non-compliances. Those extra paradigms cannot be disconnected or unslotted, they have to keep living or their processing power will shift towards more present paradigms.
Minimizing processing power for low-utility paradigms allows to reallocate the processing power to high-utility paradigms. This processing power is instanteanous and parallel, such as multicores that couldn’t virtualize by acceleration.

The second assumption is; the tasks are ordered by vital importance (as per selective evolution, or other mecanisms). That way, if a task requires a sudden rush of processing powers, it might not only takes it from available ones, but also from less vital tasks; causing a “lose of focus”.

Let’s assume a disturbance in our well-maneered paradigm. We introduce an ill-educated bot.
Practically, our bots are communicators in both direction with internal state space. Under this, there are many internal states evolving according to the input and internal loops values and observed or (internally) executed transitions.

The consequence is to cause an overload of the well-educated bots. As the well-maneered paradigm of the bot is different, assuming it is consistent to be formalized/stated, its behavior will cause a lot of responses to the well-educated bots [zone]. This will make a lose of focus, and might trigger agressive behavior as continued interruption is weakening other tasks performance provision. (reject of the bot in order to re-establish the main focus)
The other consequence, if the disturbance persists, is the lessening of the non-compliant answers. This enlarges the current well-maneered paradigm to less distinguish between these.
[fight, flight or adapt]

Although, some sort of distinguishment might provide disturbance as another paradigm to integrate?

Thoughts

Open Questions on “On Intelligence”

Years ago, I read On Intelligence from Jeffrey Hawkins as one of my first introduction on brain-derived artificial intelligence.
Far from the statistical black boxes, he had the ambition to explain a surprising pattern seen in the neocortex and extract an algorithm from it.

There are a lot of observed patterns in the neocortex, the most scrutinized blob of fat, and most of these are functional divisions; we know for long that the brain is split between functional areas, from localized brain damage and, more recently, through the study of synesthesia and the technical possibilities of imagery displaying live activation patterns in the brain.

But what J. Hawkins presents in this book is something that lambda people have never heard about if before;
a pattern that is not localized but repeats all across the neocortex, a pattern that can be observed in every centralized nervous system but has a specificity for humans…
And this well-sold feature is nothing less than the layering of the neocortex. Which really makes sense.

As our skin is layered, we have different functions orchestrated at different levels. That also means a 1st degree burn is not as bad as a 3rd degree burn on your skin. Even damaging your epidermis won’t even hurt as it’s not touching your nerves or anything really alive.
And you can probably expect a similar importance in variation between neural layers. Except it doesn’t have 3 layers like the skin but up to 6 packed on really thin sheet.

Showing six layers of cerebral cortex of control group; molecular layer (I), outer granular layer (II), outer pyramidal layer (III), inner granular layer (IV), inner pyramidal layer (V) and polymorphic layer (VI). These layers showing acidophilic neuropile and rounded open face nuclei with prominent nucleoli (→) of the neurons and also of neuroglial cells without properly seen cytoplasm (▶). H&E, (A) ×100; (B) H&E, ×400.
6 layers observation of the neocortex

This is the big announcement; we have 6 similarly organized layers all across our functional areas. Except the motor parts having 4, and other mammals having only 4, we are the 6 neural layers monkey !

He states then the importance of pyramidal neurons in this organization, not mentioning the glia cells functions at all, and ends up in a hierarchical representation where the hippocampus is atop.
A point where he insists the most; those divisions are functionally organized in volumes, or “cortical columns”, carrying the process unit and communicating with other cortical columns; which explains functional localization. The extracted algorithm idea was named “Hierarchical Temporal Memory” and hastly led to Numenta which produced white papers, open sourced and seemed to have an online machine learning algorithm that detects irregularities through a stream, with a correct result rate but not exceptional.

I wouldn’t depict myself the brain that way today, but back then it left me with a deep impression and the will to clarify as much as I could.
As I tried from the neural approach, I learnt painfully that we’re really lacking functional studies of complex behavior from “neural elements” compound to go further in this direction.
Now, I have much more perspective on this book to ask other open questions I would like to hear about;

 

One of the first thing that stunned me is the absence of architecture plasticity. If there are hierarchical patterns supposed to form from the “blank” neural sheet of a newborn, cortical columns is not enough to define functionally which layer requires which nodes and how gates are shaped.
Similarly, I have never seen a machine learning algorithm optimizing both its nodes and edges; we usually approximate the size of hidden layers (the number of hidden features) to make it good enough for learning to happen. The brain seems more to relate on time-precision, dynamic hierarchy and redundancy.
What do you think we lose in that mitigated approach? Do we have already algorithms optimizing both nodes and edges in a learning network? How to encompass for the hierarchical approach?

I am not sure I would consider the hippocampus as the top of the cognitive hierarchy.
In neurogenesis, we observe cortex builds on top of another. We have also some bootstrap period, the youth, required to progressively develop and integrate structured data processes. As such, the physical world is both the first processing system we build upon and our reference to tests hypothesis.
Would that be correct to consider that the hard world acts at both ends of that hierarchy, making it more of a cognition cycle ?

Regarding the importance of pyramidal neurons, there are numerous of them especially in complex sensing areas like those related to vision and audition or even spatial representation. The nature of those signals are vibrations that are divided to much more familiar categories like a color or the pitch of a note.
And there are some regularities as those can be represented in a computer from 3 dimensions. For instance, we can represent those senses with the 3 perceptive dimensions;

sound: amplitude, frequency and duration
color: luminosity, hue and saturation
pain: intensity, location and spread

Could this way of perceiving our world be caused by pyramidal neurons? Is it also causing us a part of subjectivity? Isn’t there a similarity between those 3 terms across senses?

Brief, Thoughts

The Curious Task of Computation

There are really 2 shapes of reality:

The physical world, with its realm of mysterious relations and this weird capacity to be the same for all of us, and the sensory world, which is a really intimate experience extremely difficult to describe (especially if you don’t share much DNA with us Humans).

As we’re not that good to make something objective out of the physical world with our subjective senses, we elaborated Physics; as a way to describe the physical world based on how it act upon itself. Meaning we can only correlate physical events with other physical events; building models made of wavelengths and energies. Things that have no meaning to our sensory world.

From the second one, we elaborated psychology; as a way to describe the sensory world based on other sensing agents. Meaning we try to correlate ideas based on interpreted experiences and derive models. Similarly, it’s the sensory world acting upon itself, but with the extra difficulty of accounting for a variety of opaque sensory worlds. Even if the architecture is genetically similar, we cannot see what’s inside or understand it from the physical world perspective.

And, in-between, as a way to match those really odds worlds, there’s the curious task computation. This domain is a way to match bidirectionally physical events with sensory events; to have the information from the physical bit to an intelligible idea or function, then back to the bit.

Giving this, my point is a question you should ask yourself:
Are we really filling the objective we gave ourselves?

 

Addendum: this is inspired, but different, than Hayek’s analysis in “The Sensory Order”. In his approach, he talks about “orders” instead of “worlds” used in this article.
Comparatively, in Serf Layer, the “physical world” is “perception” and the “sensory world” is “interpretation”. Which I believe is much more intuitive to expose them functionally. But Hayek wrote this in a much different era where computers were still a post-war secret and the Tractatus Logico-Philosophicus was a recent book.

SERF, Thoughts

An Inquiry Into Cognitive System Approach

I recently heard from my director that the technologies and architecture patterns I was pushing for in new projects were too cutting-edge for our practices and our market.
As I tried to illustrate, through past projects, how such technology or such way of designing could have avoided us unnecessary complexity: to implement, to modify and to maintain; I just had an evasive reply.

Truth is; he was right in his position and so was I.
As a business thinker, he was rationally thinking: if I’m not educated in that technology, how can I convince a client unaware of it that it is a good decision to start a whole delivery process for it?
As a technology thinker, I was rationally thinking; why don’t we give more efforts into getting the client problem from the right perspective and apply the right solutions for them, so we avoid expensive delays and can functionally scale?

***

A simple case would be to define movements in a closed room. You can define them by referencing every place there is to reach, like a corner, the sit, the observer location, the coffee machine or the door. This is easy to list but doesn’t give you much granularity. As complexity will arise, it will be hard to maintain.

Let’s say I want to be close to the coffee machine, but not enough to reach it. Or I want to reach the observer from different angles. I might want to get away from the door location when I open it. And many new cases I can think of to better match reality possibilities.
What was a simple set of discrete locations becomes complex cases to define. As it grows larger and more entangled, complex documentation will be required to maintain the system. Extending or updating it will be a painful and expensive process.

But video games solved that issue a long time ago; transiting from the interaction capacity of point-and-click games, or location-based static 3D cameras, where only discrete views and discrete interactions were available; to composite spaces assembled by collision detection and rule-based events. It’s a way to automate a level of granularity that is beyond human computability.

Think of the squares on the room floor. What if I define them to be the set of discrete locations; preferring mathematically defined locations instead of manually defined locations.
What could you say about my previous locations that are human-readable, like being able to reach the coffee machine? Both in human assumption and video games; it is a question of radius around the detected item. The spatial approach gives our set a relation of ordering over the set; that relation allows us to define the concept of radius as a relation from our domain set to a subset of “close-enough-to-the-coffee-machine” locations.

***

The other good part with this approach is that we don’t need to formalize its target subset; as long as we have a consistent relation, we can define if it eventually is, or not, in reach; and with the degree of certainty we want to apply if it’s an iterative function. We don’t need a long computation to formalize everything: the capacity of doing so, with much or less computing efforts, is good enough to give a definition.

Why would I say that? Because of precision.

As we started to mathematically define our room locations, we didn’t formalize how far were we going with that and some uncertainty remains. Should squares be large enough to encompass a human? Or small enough to detect its feet locations? Maybe down to little pixels able to see the shapes of his shoes? With great precision?

This is granularity, and it should be as granular as the information you’re interested by, because granularity has cost and complexity associated with.

From our various square sizes, we have the idea that the invariant is: there are 8 available locations from any location unconstrained by collision. So it seems the computational complexity stays the same. But don’t be a Zeno of Elea too quickly, it is impacted.

The first impact is the real move is indifferent from the scale of measurement.
The second impact is the chosen scale can be non-relevant to the observed phenomenon.

Going from a location A to a location B won’t change the actor behavior but the number of measured values will explode, ergo the computational power and the space complexity to track the actor. If you decrease the size of the squares, you get more accurate measurement but much more data to process. If your squares size increases, you’re not sure anymore whether the person can reach the coffee machine from its location. You need to go as far as information is not noise yet.
At large, it’s a matter of finding the optimal contrast at each scale of the information.

***

What we have seen here with the room floor are 3 level of scales:

  • with raw discrete locations, more on a human-functionally readable level, but limited like point-and-click games, or preset camera locations.
  • Then a first granularity that could help us detect roughly locations, think of it as a 2D RPG old-style.
    On it we can start applying spatial logic, like a radius, to know which locations are in range.
  • Finally a thin granularity level that allowed us to go as detailed as to describe the footprint of the actors in the room.
    That level of granularity is more common for 3D texturing in modern video games or machine-engineered electronic devices such as touchpad or display screens. Every sensor, every pixel, has its transistor (up to 4 in a colored pixel).
    When you get down to that level, either everything is identically formalized or you are in big troubles.

My point being; the problems of the markets haven’t changed and probably won’t. We’re starting to make mainstream the technologies to deliver a transition from point-and-click to 2D RPG. In some specific cases, we can even start reaching for thin granularity.

We can foresee that low level of granularity will benefit high-granularity technologies.
But the shift of business perspective; regarding how problem can be solved with more data, more granularity, better decoupled technical and functional levels of interpretation, and so on; is not done yet! Although it’s just a shift of perspective away from us to pivot to this low granularity way of thinking our old problems and old projects; our approach where we claim unfitting architecture pattern, process and a global lack of granularity and decoupling .

There are opportunities to unravel and technologies to emerge as we will move towards such a transition in our traditional systems. We will have to work on more complex mathematical formalization and automation to unleash a wider realm of implementable functionalities.
We can deliver better functionalities that we will use as the public tomorrow. We just need to find this more granular way in our projects, where we could build higher added value.

The tools are there, the first ones to shift their perspective will take the lead.
Besides, the fun is to set up granularity, not to improve it.