Machine Learning – Andrei Drăgănescu

This is the reason Elon Musk has rockets and you don’t

Elon Musk pretends to know shit about rockets.

Listen, your brain is very good at two things:

making structure out of unstructured data
finding patterns in structured data

The brain is great at figuring out how to organize streams of unknown data, and we call this process understanding.

The brain is great at recognizing patterns in ready made structures, and we call this process grasping.

However,

the brain is awful at understanding structured data, unless it created the structure for that data.

For these reasons, whatever you learn:

start in the middle
jump right in, a.k.a. act as if you know
expose yourself to data
use your senses
always look for real world reference

1. Start in the middle

The joke is that this is the reason why Star Wars succeeded, because they started with episode four. We don’t know that, nevertheless, try not to start with the basic stuff. Go for the middle of the subject matter.

Try to figure it out by referencing the beginning. Don’t go too advanced either, because you’ll loose track of explanations on top of explanations. But, in general, the interesting factor of a subject is at medium difficulty, and the amount of “interesting” is what makes the learning reinforcement feedback fire.

2. Jump right in, aka act as if you know

This is the golden rule of learning, in my book. I think this is what keeps most people at bay when subjects hit their life and they need to learn:

Figuring out every single detail. That is the recipe to quit.

The rule? Skip it and plow ahead.

Plow ahead, expect bruises and blunders, but act as if you understand. Don’t let yourself be pinned at some level because you don’t get it. Don’t be disappointed if it doesn’t make sense, read on, it will.

We have background processing. All learning is exposing our brain to tons of structure made by others. Finding the patterns inside manuals of theory is hard work, but while you stream data in today, the yesterday’s stuff is still being worked on.

In nature we don’t need to know what the bark of trees or the fur of cats is made of. We know that somehow they are similar, even if very different. We, and many big brain animals, skip and plow ahead with trying to grasp more and more general patterns. And there is a point where it suddenly makes sense, and you think retrospectively and that past reading makes sense.

This is also the rule of least resistance. That fog of the mind is when we’re keen on grasping things way too early. Let it go. Let it gooo. Let it goooooo. Le … Damn you Disney.

3. Expose yourself to data

Learn a language? Sing their songs. Learn physics? Read on near phenomena like rain or lightning, or why water makes drops, or why dew forms.

Whatever you learn, use data as soon as it reaches you to connect to things you already know. There is nothing more memory forming than this, since it is so close to how direct experience happens.

4. Use your senses

Brain’s organization of data streams is not based on a priori systems. It is basic embedded biologic systems at work doing the same thing over and over, until data split in thin vertical silos begins to show connecting features between them.

Our brain’s organization of data is an internal process developed for our senses and the data they keep streaming in. It is actually optimized for that kind of raw data, and less so for structured reading of theory or formulas.

Therefore, whatever you learn, do your best to both see and hear and as much as you can touch or smell the subject. Math can be touched, make paper models of concepts, math can be smelled, all perfumes are algorithms. So if math can be enjoyed by all five senses, anything can.

5. Always look for real world reference

Take geometry. That is the best example, geometry is everywhere, it is easy to make real life stories to illustrate concepts. But, take trigonometry, the classical example of “too much too early”, where concepts are disconnected from reality until later on.

This means:

always make up a story.

If the textbook doesn’t help, well, do research! See what the hell is this thing you’re boggling your mind with used for anyway.

Overall, remember, this thing we call “learning” is not natural.

Learning as a self disciplined action is unnatural. You force yourself to be vulnerable, isolated, and exposed to things you don’t understand, repressing fear with your frontal lobe censorship of emotion.

What nature intended as learning is what you require very little effort for: figuring out reality. You are doing it.

You have a brain that is the best real time three dimensional image processing system yet, you have processes such as facial recognition and social cue filtering that work around the clock, you dream, you are a perfect motion machine, you remember vast amounts of data since you were born, so don’t ever, ever, ever feel intimidated by books, especially text books.

Textbooks are best attempts to put on paper the way other people figured out stuff. This is hard and most textbooks suck.

We only understand what we experience and we only grasp what we’re taught.

P.S. If you are an author of learning material, try and do this: build patterns that grow from the center to outside and explain them upfront as the key to the knowledge ahead.

Also, click the heart, even though I am unsure of the effect of it anyway.

Conscious machine

So there is this intelligent body and this consciousness that must somehow work together. The body doesn’t care much about its eventual expiration, it cares a lot about present integrity. The consciousness cares less about present integrity but is scared shitless of the incoming end.

Intelligence is evolution at work. But consciousness is a side effect that rose up from the predictive analysis which intelligence enabled. Your body knows what it does. It knows it so well it doesn’t care much about you, as a person. No cell asks you if you’re happy with its DNA reading. You are built on a scripted execution environment.

At the level of the body prediction is minimal. There is possibly some kind of intelligence at the level of matter and some primal form at cellular level, but there are no unscripted decisions in the formation, maturing and living processes of a being. If it were you’d start growing a skin pocket for that cell phone by now. The homeostasis solution is known and at this level, in this environment, awareness is not required. Levels, this word repeated in this paragraph is essential to grasp: we’re layered things.

Our internal bodily processes happen without any supervision from what we call “us”. Isn’t that weird? It is amazing that most of the known diseases would be cured much easier if we could “tell” the body to stop doing something. The freaking flu happens because of too much immune response, wouldn’t it be nice to tell your body to stop overreacting?

Why don’t we have a dashboard with status parameters? Because what we call “I” or “me” is merely a small side effect of a tool the body thought would be nice to use: consciousness. We have this illusion of free will because the body does not oppose anything in particular, unless you direct your self at you. Suicide and self harm are the exception not the rule.

We’re a thin film on top of everything and the everything we’re on top of doesn’t seem to share our goals and aspirations. It is important to realize how external to your body you are. That out of body thing reported so much, that is the constant thing happening. The senses are really immersive yet what you call you is out there already. This fluctuating electrochemical equilibrium that defines you is not made of the same thing the supporting layers are made of.

You are an intelligent machinery, with billion years old scripts executing non stop in perfect order and some error correction with highly accurate short term prediction abilities, stuck to this consciousness which is a small thing busy with being itself while depending on the intelligent machinery for its existence.

We have a dream, a waking dream actually, of cohesion, as if we’re the same, integral, contiguous, smooth, inside and outside, at all times. The intelligent machine is. The consciousness is. By themselves. Together the cohesiveness breaks.

Freedom. I find it funny when I think of some cell locked into its place and purpose, controlled by the greater good’s aim of resilience. We see this interference of the machine with the conscious aware being in the way we model what we call the world: always coming back to units with easy to understand designated functions, always striving to define and box freedom into degrees that are painless to stretch into, continuously finding solutions that work on old maladjusted systems, resistant to environment but frail on the inside.

Higher conscious levels don’t have the access required to devise a clear system. Most of what we call “I” and “me” is so busy being separate that we rely in huge proportion on insight offered by unconscious and subconscious processes.

In a world made by fully aware conscious superior beings there is only spontaneous structure, the memory is collective, knowledge freely shared, goals vary and the greater good is the shared understanding of the uncertainty ahead. But there is no such thing so far.

We’re passengers on a vessel with a known destination, except that we don’t want to get there. Some are the noble guests on the deck of this vessel. some clandestine in the hideouts of the machine’s belly. Above and below though we share the impotence of changing course.

You know, Fabiana I wrote this after going through the constellation based G universal database and trust network by Heather Marsh. Unsure of the connection :D, anyway …

Ergo sum. That’s all we know.

Unfortunately we don’t have direct knowledge on the first part, that is to say, I don’t have a direct knowledge of my own thinking because I am trapped in it. The moment my first thought appeared, it started to weave the layers of layers of personality, from the inside to the outside.

I don’t know if I think, or I feel, or I grasp, or I sense, or I act or or or. We can understand that we’re doing, whatever it is we’re doing, but direct knowledge assumes examination from an outside reference, and we cannot be outside our thought processes.

Meditation elevates one to upper layers, but not outside, because outside the thinking loop you cease to be you, therefore whatever is that which is looking will not pass back the knowledge of what it saw.

Limiting our basic philosophical perspective to ergo sum is a healthy choice. It is akin to:

“Eat food. Not too much. Mostly plants.”, M. Pollan

because right now that is all that our direct knowledge of how the complex relations between the billion moving parts of our metabolism work.

That’s all we know.

Perhaps, in the creation of artificial life and of the corresponding super artificial intelligence accompanying it, we’ll get that outside reference on the process of thinking that will validate or question the foundation laid by mr. Descartes.

The return of the mainframe?

I wonder will PCs revert to being dumb terminals with a twist.

I imagine buying this hardware that can’t do much, like a chrome book, that connects to everything and learns all there is online about me and it “knows” what to do next.

I connect to Facebook and suddenly my computer knows my funny bone.

I connect to Twitter and suddenly my computer knows my interests.

Just type in your Google account and we’ll teach your machine about who you are.

I imagine this future and I feel it coming at light speed. With Intel and Nvidia cranking out chips tuned for ML, with Google and Apple making farms on top of farms of computers built only for ML, the basic software, the operating system has to be just a pack of protocols and a browser which sports ML specific APIs that can access ML specific hardware, including ML specific storage.

I don’t believe in computers that won’t work without the internet. No matter how advanced we get in this area, I mean even if Facebook Free Basics will eventually break through and informationally enslave, or colonize, half of the planet, the idea of moving data through air or cables is prone to out of service events. Unless we begin to teleport electrons into machines, we’ll need offline capabilities.

But I do believe in computers that use the internet to become better. This is already the case for a decade. But ML and in particular industrial, global, privacy shredding ML will make computers actually become better online, not simply being more useful which is the case for now. Software will be about the same thing: learning. What do we want the computer for? Do we not want a wizard, an Oracle, that knows in advance what we need, instantly what we want and always what we mean? We don’t need productivity. We need tools that make what is required of us, in our place.

Machines need not be “creative”. We got all the creativity the world requires. Because it is our world. When machines will get creative, it will be their world. You know why, don’t you? Yes, it is because that is what creativity does: building worlds inside the information space. Creative machines will start making their own worlds.

But, this is reality not the far future at the end of A.I., the movie, with those jelly robots who couldn’t reverse that Ice Age. I hope you’ve seen the movie only if for that final act.

In reality, my future computer will be some kind of interactive machine which will support any kind of input and make sense out of it, not because it “thinks” like a human, but because it “learns” like one. We are heading in the future where your service providers will know you better than you know yourself. They already do, but we’re blessed with bad practices of interoperability that still keep data in silos. For now.

I amuse myself with the idea that the keyboard for future generations will be an “accessibility” option. Think about a 3D space where a ML tuned system predicts with 90% accuracy the next word in your sentence and if a checkbox is checked, predicts the entire thought line in less than 300 milliseconds, building on your chat history, which it has learned from since you were six and your parents gave you your first iDevice (or mDevice, who knows). Why type? Why speak? We already have an inner representation of a linguistic 3D space in our brains. That is why we like to move our hands and our bodies, to redraw the imaginary route between how concepts are connected inside.

If ML turns out to work consistently and reliably on small form factors every giant software company will be able to crank out their own chips because it will be worth it, because each consumer will want various ML enabled devices for all kinds of things. The home hub everyone tries to invent might be GPUs stuffed in an internet connected box, and the brand to make it will be the next Apple.

The problem with the entire internet becoming a huge virtual assistant is who will pamper our soft inside?

This diagram means nothing

If this is the best way to visually explain ML we’re fucked.

How to be your best self

Control yourself but never
Deny yourself.

Loose yourself but
Protect yourself.

Apply yourself don’t
Do it yourself.

Touch yourself but
Keep it to yourself.

Grab yourself and
Save yourself.

Be yourself but
Suit yourself.

Work with yourself to
Express yourself.

Contain yourself but don’t
Kick yourself.

Top yourself but don’t
Fuck yourself.

Validate yourself but don’t
Praise yourself.

Push yourself then
Treat yourself.

Help yourself to
Love yourself.

Step into the loop

I am made of things I know, about the things I know.
If you strip away the things I know, about the things I know,
I am left only with the things I know —
which keep me alive,
but I will not be me anymore,
I will just be It.

Many times I see the It.
Many times I see
the me on top of the it which makes me I.

I see some around me and I call them them.
The It calls them the others.
The others become them to me only when
I see the me in them.
I see the me in them,
because of the things I know, about the things I know, about the others.

And there are those who are among them,
about whom I even know things,
about the things I know, about the things I know, about them,
and I call them we.

And there is the other me in we,
Confusing my it,
who doesn’t get how an other can be me,
that other me out there in we,
which makes my it let go of me,
because of you.

I am now split in a me and a you,
and I call them us.

Could the fabric of the universe be made of more than space and time?

Consider for a moment that life is an embedded property of the universe, just like time and space are.

Life is not a phenomenon. Biology is a phenomenon caused by life. Just like time is a property of the time-space continuum, life is a property of a larger continuum space-time-life.

Time and life are two opposing fundamental forces of the universe. Time is the global entropy increase and life is the local entropy decrease. I don’t know if they’re symmetric, but they are closely so. It is a fractal who grows the same over and over again, no surprises here.

We could call this anything else instead of “life”, but I chose to call it “‘life” because the local entropy decrease is the root cause for biological phenomenons and evolutionary processes, which together we empirically associate with the concept of life.

Both awareness and consciousness are information constructs rising in the electrochemical equilibrium in the brain, equilibrium reached in the brain’s effort for more accurate predictions. It is neither consciousness nor awareness that causes the weird measurement problem, but the fact that when we measure we are observing time-space-life instead of simply time-space which we think we’ll find. Because we may be nothing more than self sustaining processes manifesting as biological phenomenons in the physical reality, by measurement our contained hyperlocal space infused with the effects of life — the space-life, transforms time-space into time-space-life.

Life is a hyperlocal property of the universe which makes extremely low entropy in the form of functional systems, the pinnacle of which are self sustaining processes.

Because of its nature, life develops resilience just as time, because of its nature, develops decay.

The first result of low enough entropy is the appearance of structure. The easiest example of structure are crystals. Structure is the first power of resilience.

A process has far lower entropy than a structure. Because a process has steps, a start and an end, the constraints to build a process are huge, much bigger than the constraints to build a structure. A process is the natural result of lowering entropy below the threshold of a structure. Homeostasis is a process resulting from lowering the entropy of a biological structure.

A self sustaining process has even lower entropy than a process. All natural self sustaining processes feed upon themselves. Any outside feed and the entropy would increase. In general, time injects outside feeds into self sustaining processes created by life, this way increasing their entropy and hence lowering their resilience.

Life, as an embedded property of the universe, generates structures. You might wonder then what is the difference between a crystal and a virus? Protein. It is very likely that viruses simply appear randomly as part of the process of life in the local time-space-life continuum.

These two properties of the universe we call life and time are “perceived” by us in a certain way: the animus and the tenses. They both are however way beyond the scope of our perception. Time is so global it encompasses our infinite universe, life is so local that it is broken into an infinite of pieces. Both time and life act upon space.

Because life is the inverse function of time, we may say it builds on the future to create the past, while time builds on the past to create the future. From life’s perspective any high entropy space is in the future and from time’s perspective any low entropy space is in the past. Both time and life are intertwined with space.

The phenomenon of biology is the manifestation in the physical reality of life’s process creation. Evolution is a form of creating resilience by compounding on previous states and it is a process that results spontaneously given enough detail is present in the structures and processes that result in life-space.

As life acts in reverse of time and lowers entropy, it follows that its peculiarity and special requirements to grow complex self sustaining processes is quite normal. By lowering entropy levels, processes become more and more complex and this complexity is what holds advanced biologic phenomenons to be more common in the observable universe. But the catch is that the rarity of advanced biologic phenomenons is not because of the lack of water and oxygen, it is because of process complexity. There may very well be biologic phenomenons based on silicon and nitrogen but there has to be a whole other array of compatible things around for the self sustaining processes that appear in life-space to keep compounding on themselves.

The great filter is us

Minimizing entropy means lowering side effects. Inside a process, to lower disorder you guard for side effects. That is why every advanced evolutionary process is three things: self sustaining, self referential and selfish.

The great filter is the maximization of the three properties of the biological phenomenons: the self sustaining feeding, the self referential information chains and the selfish bias. If there is a great filter it is only within ourselves and it may be beyond our capacity to avoid it. If there is a great filter and other civilizations have reached it, it is very likely they have destroyed themselves. Self destruction is the last name of the great filter.

Human self destruction is caused by one or more of the following local maximums: missing sustainability by feeding too much into the process itself (e.g. ecosystems break down), misinterpretation of perception by relying exclusively on interpretation (e.g. humanity is on a pedestal and the other animals are on a lower step) and protecting your kin by limiting outside feeding into the particular facet of the process we’re into (war!).

At present we could be in a race between developing true AI (or artificial super intelligence, or artificial life as I call it), which is the next lower entropy step for a super intelligent evolutionary process, and reaching our self destruction great filter. Best of luck.

What does being “alive” mean?

There has been a great deal of discussion about artificial intelligence and the potential dangers it poses to humans. Then there have been countless iterations of interpreting via stories moral lock downs such as robot rights, slave robots, artificial will and so on. But the main question regarding A.I. is

when exactly does A.I. become more than just a fancy tool?

My main hypothesis is that artificial intelligence will not be more than a query answering robot unless we factor in it unknowns and inevitable, time based, cycle endings. In particular we need a system who fights for its existence, a system that runs in a scarce resource, competitive environment, even more, a system that cannot access all the answers.

Artificial life is not only intelligent, it has the same goal as life in general: to defeat time because it is threatened by it. And, when we’ll create it, the artificial part will be lost, because, in the end, we’re meant to do it as our next evolutionary action towards time resilience. (more)

In that sense, we could say something is alive if:

the base goal is to preserve low entropy
all actions are determined by a predestined time resilience
it is exposed directly to the effects of time, in particular decay

Even if a being is immortal, all three of above can still apply, and the being is alive.

The goal of any living being is resilience as a form of preserving its status.

where R is resilience, S is status and P is preservation. To ease calculation R, S and P can be integers, but in reality R is a time delta, S is a matrix and P is an algorithm.

Status is a complex notion that sums up genetic, biologic, cultural, societal and other states. Preservation is an algorithm that employs innovation, mutation and other methods of predictive feedback initiation. Both status and preservation change in time based on environment updates.

At birth a living being has P = 0, therefore R = 1.
At the time of death, a living being has S = 0, therefore R = 0.

If we create an A.I. system and aim to make it work above the limitations of a fully determined program, we require the incorporation of the formula of “life” in the main loop. The stability of the main loop must decay due to direct interaction with a physical hardware clock.

A decaying main loop is created by updating a composite variable on every tick of the hardware clock. Say we have a variable made of many parts such as A..n then there is a homeostasis function H that produces a result of a composite variable A..n.

A..n = H(A..n, S, E), where S is state and E is environment

Each part of the composite variable is then injected in the loop as values for various internal parameters. The H function applies data from state and environment to A..n. State is the current execution state and environment is the input of the current execution state from sensors and detectors.

Thus resilience is not embedded in a system, it rises naturally from decay based on external action. The better the preservation the bigger, exponentially, the resilience. The better the status (both complexity and connections, as status is a sum of states) the better the resilience.

If we build an A.I. that is generating questions, not answers, using answers simply to point to new questions awareness should arise by itself, but only in the presence of decay and resilience, otherwise we’ll never know if the system is aware as it has no reason to expose it.

Know thyself the Facebook way

Here is a cool exercise for you. Go on Facebook. Find all kinds of pages and people and follow and like them. Then spend one hour a day interacting with your feed. The resulting feed will reflect yourself more than you know 🙂

Find 10 people that like whatever you hate.

Find 10 pages about whatever you find hugely outrageous.

Find 10 people or pages who post chain stories such as money for likes, fake charity posts and the kind.

Find 10 snake oil pages and like them. Doesn’t matter how skeptic you are do it.

Find 10 pages or people who post about aliens ruling the planet, reptiles impersonating Obama, etc, you know it when you’ll see it. Doesn’t matter you know its hoax.

Find 10 pages or people who post about faith. Try to find the ones who don’t post about snake oil and who don’t post about spiritual shamanism or miracle healing, but those how post about God, Buddha, Allah, Catholic faith, Orthodox faith, Jewish faith whatever. No matter how atheistic you are, do that.

Search for top notch scientific figures and places and follow them. Search for Neil DeGrasse, search for pages of NASA and CERN. See what Facebook suggests follow that, no more than 10 overall.

Find 10 major skeptics/atheists and follow them or like their pages. Start with Richard Dawkins Facebook will suggest the rest of the roster.

Follow or like contemporary musicians from mainstream music, about 10 of them. Follow or like 10 pages or people who post about classical music.

Look for people or pages posting about art. Look for Art, Ctrl, Del, for example then follow suggestions.

Please follow and like cats. There are so many cat resources. And dogs. And pandas. If you find animals with their own accounts, follow them.

Ok, you get the drill. On Facebook you can dive into whatever weirdness or outrageous thing you can think of. You can do it by simply clicking a button.

You can direct towards yourself the most diverse kind of information in the history of mankind. Do that.

Than browse around and see you feed evolve. Wait some months out and you’ll see that no matter how hard you cheat that it will end up showing less of some things and more of some other things. The more you use it and the more interactive you are the faster it will adapt.

There is no faster way to get and know thyself. Sure there is always travel and friends and adventure and love. But Facebook is free, simple, easy and always around. And it has the best, state of the art, people analyzing machines ever built. I don’t think even the NSA have such good profiling. Think about it, do you think spies and terrorists give away how they fancy nice dancing koalas, naaah.

We think that our Facebook profiles represent us. But no, it is the feed that tells the most about us. All those neural nets shouting: right back atcha!