Northwest Noggin was thrilled to receive an invitation from the Portland Art Museum, the seventh oldest museum in the United States, to present a workshop on neural networks at their free community “Evening for Educators” event in February, 2023.
LEARN MORE: Evening for Educators: Hito Steyerl: This is the Future
We’ve been honored to collaborate with PAM in the past on multiple projects, including the “Seeing Nature” exhibit in 2015-2016, contributing original Noggin panels with text on the neuroscience of vision for a “Nature of Seeing” gallery which accompanied this striking collection of landscape paintings from the Paul G. Allen Collection.
LEARN MORE: The Nature of Seeing @ the Portland Art Museum
LEARN MORE: The Nature of Seeing: Noggins and Art
LEARN MORE: Noggin @ Portland Art Museum
This is the Future
Sara Krajewski, the Robert and Mercedes Eichholz Senior Curator of Modern and Contemporary Art at the museum, introduced and deepened our understanding of the new Hito Steyerl exhibit, a compelling, immersive and powerful installation incorporating video, sculpture and sound. Go see it!
A central figure in the work is Heja, a Kurdish artist imprisoned in Türkije, who grows a stunning garden with the help of artificial intelligence, and hides it in a predicted future.
According to Sara, “Steyerl is one of the foremost artists offering critical reflections on the complexities of the digital world, global capitalism, and the implications of artificial intelligence (AI) for society.”
This is the Future will be at the Portland Art Museum through June 18, 2023.
LEARN MORE: Hito Steyerl: This is the Future
LEARN MORE: Turkish Artist Zehra Doğan Sentenced to Prison for Painting of Kurdish Town Attack
LEARN MORE: Kurdish artist flees Turkish prosecution to find creative sanctuary in Germany
LEARN MORE: Interview Heja Türk
Neural Noggin Networks!
“All you need is lots and lots of data and lots of information about what the right answer is, and you’ll be able to train a big neural net to do what you want.”
— Geoffrey E. Hinton
We tricked out the museum space with our hand-crafted pipe cleaner neurons, 3D-printed human brains, multiple skulls and of course some real noggin specimens, as a backdrop to our workshop discussion of artificial intelligence and its relationship to the neural networks in our own brains.
Our informed and enthusiastic Northwest Noggin presenters at the museum included Rebecca Chevrel, Brin Rey, Bradley Marxmiller, and Roman Cimkovich, undergraduates studying interdisciplinary neuroscience at Portland State University, and Dr. Denesa Lockwood from Oregon Health & Science University.
What is artificial intelligence?
Artificial intelligence (AI) is “…a field focused on automating intellectual tasks normally performed by humans…” writes RY Choi and colleagues in their review paper “Introduction to Machine Learning, Neural Networks, and Deep Learning.”
LEARN MORE: Introduction to Machine Learning, Neural Networks, and Deep Learning
Machine learning (ML) is a type of AI, where computer networks are given access to reams of data in order to extract descriptions of different categories, or predict what might happen given the presence of certain data, or prescribe specific steps to take in specific situations. The more data the network can train on, the more accurate the descriptions, predictions and prescriptions might be.
LEARN MORE: Artificial Intelligence (AI)
Today we’re all encountering machine learning through targeted ads, for example, based on what we type into the Google search bar, or what we like on Instagram. And corporations and governments are extremely interested in gathering more personal data about each person, because these systems are getting very good at predicting our behavior and selling us things (including candidates and policy choices). Research is also exploring how these technologies might improve medical diagnosis.
LEARN MORE: Unlocking the Power of Artificial Intelligence and Big Data in Medicine
LEARN MORE: Artificial intelligence, machine learning and the evolution of healthcare
LEARN MORE: AI, machine learning, and deep learning for clinical outcome prediction
In machine learning, there have been attempts to predict in advance what specific data might mean and require in response, and craft a fixed network that just responds to what we expect. But the world is not like that. More effective, and in some ways more “biological” systems, act like our own neural networks, driven by exploration of new data and reinforcement of responses that let us reach important goals.
The networks in these systems change, and learn with experience.
LEARN MORE: Spoken Language Identification Using Deep Learning
LEARN MORE: Reinforcement Learning for Precision Oncology
LEARN MORE: Deep learning for computational biology
Our own neural networks (our own brains) are made of linked cells called neurons. We have specialized sensory (or afferent) neurons, that convert (or transduce) a limited amount of energy (data) in the world, including a sliver of electromagnetic spectra (light), sound waves, mechanical pressure (touch), into electrical activity, which gets communicated chemically to other neurons across small gaps called synapses. Ultimately the network includes output – our behavior – effected through the electrical activity of motor (or efferent) neurons on muscles.
But in between, for humans, are vast numbers of “between” neurons, known as interneurons, that link together our selective sensory inputs and behavioral motor outputs, and bias our descriptions, predictions and prescriptions based on our own personal experience with the limited data we transduce from our world.
And our networks and synapses change dramatically with experience.
“Through practice, repeated signals have been passed along neural networks, strengthening synapses and thereby burning the skill into the circuitry.”
— David Eagleman
They exhibit extensive plasticity, particularly when encountering new, or unexpected data that doesn’t fit our existing predictions and biases. In fact, during childhood and adolescent development we lose more than half the neurons in our brains, ultimately wiring up neural networks adapted to particular data – the world we individually experience – to allow us to predict, survive, and hopefully thrive.
LEARN MORE: Synaptic Plasticity Shapes Brain Connectivity
LEARN MORE: The Cell Biology of Synaptic Plasticity
Making new networks!
NW Noggin Art Coordinator Jeff Leake led us in a little neuron crafting exercise, distributing piles of colorful pipe cleaners that we could bend and twist into dendrites, cell bodies and axons using our own individual neural networks – to share with each other and connect!
MAKE YOUR OWN BRAIN CELL: Pipe Cleaner Brain Cells!
Is your brain “overfitted?”
From Jeff: In the field of artificial intelligence and machine learning, a common problem is overfitting, where a model becomes too specialized to the data it has been trained on and fails to generalize well to new situations. To address this issue, researchers have developed techniques like introducing noise into the training data to increase a model’s ability to adapt to new inputs.
Neuroscientist and fiction writer Dr. Erik Hoel, has proposed an intriguing parallel between overfitting in AI and the human brain. In his “overfitted brain hypothesis,” Hoel suggests that our brains can also become too specialized to certain experiences, with necessarily limited sampling of what’s here, leading to distorted models of reality and cognitive biases.
To counteract this tendency, Hoel suggests that our brains use dreams as a way to introduce noise into our mental models and increase our ability to accommodate new situations. Dreams are a space where our brains can generate novel, often bizarre scenarios that challenge our existing mental models and force us to adapt.
“…dreams are a biological mechanism for increasing generalizability via the creation of corrupted sensory inputs from stochastic activity across the hierarchy of neural structures. Sleep loss, specifically dream loss, leads to an overfitted brain that can still memorize and learn but fails to generalize appropriately…”
— Erik Hoel
Hoel also proposes that art can serve a similar function in introducing noise and increasing mental flexibility. By presenting ambiguous or challenging elements, art can require viewers to approach the work in a more flexible way, accommodating multiple meanings and interpretations.
In “This is the Future” Hito Steyerl provides a compelling example of this idea. The exhibit’s video, installation, and narrative include numerous ambiguous elements that prompt viewers to approach the work with an open mind and adapt their thinking to multiple interpretations.
What data do you encounter in your daily life?
Do you further restrict and curate your experiences by, for example, only visiting certain websites, or going to the same places, or interacting with similar people?
Are your own neural networks ever challenged by meeting different people, going to new places (even within your own community), dreaming, considering unexpected approaches and ideas, or making and viewing art?
What garden are you growing?
LEARN MORE: The overfitted brain: Dreams evolved to assist generalization
LEARN MORE: From AI to abortion, the scientific failure to understand consciousness harms the nation
LEARN MORE:
Networks contemplating networks
And of course our accomplished undergraduates from Portland State University considered audience questions while examining both real and 3D printed brains!
Thank you Brin, Bradley, Denesa, Roman and Rebecca, and thank you Hana Layson, Sara Krajewski and everyone at the Portland Art Museum for a wonderful night of art, networks, plasticity and brains!
“Thank you all for the wonderful workshop (and fantastic photos)! I really appreciated how each of you connected the exhibition with neuroscience and how clearly and helpfully you explain these complex systems. People clearly loved the brains and the hands-on neuron building. Thanks and hope to see you all soon!”
— Hana Layson, Head of Youth and Educator Programs, Portland Art Museum
LEARN MORE: Hito Steyerl Exhibit Educator Resources @ Portland Art Museum
