Inceptionism is a bizarre phenomenon that has recently popped up in the internet world.
Created by Google in 2014, it consists of dream-like hallucinogenic images developed
by a computer vision program called DeepDream.
The program uses an advanced AI system known as a convolutional neural network.
But that's not exactly a term we hear very often.
So what exactly is DeepDream and how does artificial intelligence come into play?
When you hear the word "dream", you'd be inclined to think about those vivid experiences
we manifest while we're asleep.
However DeepDream in this context, refers to a program designed to categorize images.
It's the way this machine learns.
A convolutional neural network is a specific type of AI made up of computing systems based
on the biological neural networks found in animal brains.
That is to say, they're modeled to learn in a way similar to how we learn.
The name "deep dream" comes from the concept of deep learning.
It's the process a neural network goes through to actively analyze things.
And though it has very little to do with how our own biological neurons work, we might
gain a better understanding by looking at how we learn.
Two educational psychologists by the names of Ference Marton and Roger Säljöl suggest
that we humans have two different approaches to learning: a surface approach and a deep approach.
In humans, the concept of surface learning is how we memorize parts of information that
we might be questioned about later.
In contrast, our process for deep learning involves actively searching for the meaning
of information.
In regards to machines, deep learning is more like a series of fine-tuning based on a set
of probabilities and conditions.
An algorithm.
And we see the practical uses of this sort of programming, throughout the internet.
It's how the internet detects content that might be pornographic.
Or how a YouTube video might be demonetized for using certain keywords or songs.
Another example: you might have noticed the presence of an artificial intelligence and
how it "learns" on Facebook, where AI automatically tags uploaded pictures with
the names of the people in them.
With regards to DeepDream, artificial neural networks are programmed to analyze specific
patterns within images.
They learn to recognize and identify similar patterns in other images.
And in cases where they're being trained to detect specific objects, they're exposed
to millions of images containing the desired object.
What DeepDream does is take this a step further, by repeatedly analyzing the details in an
image over and over again in multiple layers, drastically enhancing the patterns it detects
each time, until we get something like this:
So how does an AI make new visual connections to specific objects from basic patterns?
Well, you might've heard of certain songs that when played backwards or in slow motion,
have random noises or strange messages from obscure voices.
Perhaps you thought you saw Jesus in your morning toast.
Or perhaps you were looking at a forest one day and thought you saw a face lurking in
the patterns of light and shadow.
Pareidolia is the psychological phenomenon where we look at an image or we hear a sound,
and our minds interpret a familiar pattern when there isn't one.
Skeptics often attribute pareidolia for why some of us might see ghosts or even shadow people.
It's derived from the Greek words para, meaning "beside or instead of"
(in thiscase meaning something wrong), and eidolon, which is the noun for image, form and shape.
You could say that DeepDream uses algorithmic pareidolia to create these bizarre, dream-like images.
It's designed to detect faces and other patterns in images, and automatically classifies
those images within its algorithm.
And as you can see, this makes for some rather unsettling sights to behold- especially since
the AI often seems to detect target patterns within photos that aren't really there.
This all seems rather complicated though.
Why is it that these artificial visualizers tend to create objects in images
that weren't originally there?
According to a Google research blog posted by its own software engineers, when an image
is given to the neural network:
"We ask the network, "whatever you see there, I want more of it!"
This creates a feedback loop: if a cloud looks a little bit like a bird, the network will
make it look more like a bird.
This in turn will make the network recognize the bird even more strongly on the next pass
and so forth, until a highly detailed bird appears, seemingly out of nowhere."
This "over-interpretation" is similar to how we as children might watch the clouds
and interpret random shapes from them.
Since DeepDream was trained mostly on animal images, it naturally translates shapes into animals.
It's also why you might see it change a blank horizon into a city skyline, probably made of eyes.
The common presence of eyes in all animals might explain why the DeepDream seems to be
so fixated on seeing them in almost any pattern.
Neural networks seem to grant us a new channel for both creating and understanding abstract art.
When I look at the vibrant, alien-like compositions and landscapes of what these artificial minds
create- I'm reminded of the mysterious unknown prevalent in Lovecraftian lore.
The study into neural networks seems to have only just begun, but it does appear to grant
us insight into what our AI has learned about our world- as well as what that could mean
for AI advancing in the future.
And from this, we may even be able to draw parallels into understanding how our own organic
minds visualize the universe around us.
I would like to explore the powerful psychological effects these images might have on us, but
that's a video for another day.
If you'd like to learn more about this topic, I've left some neat sources in the description
below.
And if you'd like to see what odd images the DeepDream engine sees in your photos,
you can visit the DeepDream generator which I've left a link to in the description below.
So what do you think of DeepDream?
Let me know in the comments below.
And as always, thanks for watching.
Wanna see more? Click here to checkout the official Darkology playlist!
Không có nhận xét nào:
Đăng nhận xét