A person with their hands on a laptop keyboard is looking at something happening over their screen with a worried expression. They are white, have shoulder length dark hair and wear a green t-shirt. The overall image is illustrated in a warm, sketchy, cartoon style.  Floating in front of the person are three small green illustrations representing different industries, which is what they are looking at.  On the left is a hospital building, in the middle is a bus, and on the right is a siren with small lines coming off it to indicate that it is flashing or making noise.  Between the person and the images representing industries is a small character representing artificial intelligence made of lines and circles in green and red (like nodes and edges on a graph) who is standing with its ‘arms’ and ‘legs’ stretched out, and two antenna sticking up.  A similar patten of nodes and edges is on the laptop screen in front of the person, as though the character has jumped out of their screen.  The overall image makes it look as though the person is worried the AI character might approach and interfere with one of the industry icons.

A data scientist had found that their work (the algorithm depicted on their laptop screen) has ‘jumped’ out of the screen and threatens to cause problems with a variety of different industries. Here a hospital, bus and siren could represent healthcare, transport and emergency services. The data scientist looks shocked and worried about what trouble the AI may cause there.

AI across industries

A laptopogram on a plain background with very simplistic black-outlined blocks stretching across the centre, almost end to end. The blocks are linked, but not solidly, often leaving small gaps between the objects suggesting they could still shift around and recombine.

This image shows a diagram of a neural network algorithm, created by a machine learning researcher to communicate the system’s architecture. Diagrams like this one are often drawn as a series of connected boxes. They depict cognitive functions as a linear procedures, comprised of basic building blocks that can be recombined to one’s heart’s content. The image is a ‘laptopogram’, created by exposing photographic paper using a computer screen and developed in the artist’s bathtub. The process preserves a digital artifact of AI research in silver crystals, returning a physical dimension to sterile data. Dust, scratches, and the marks left by the artist’s hands draw a connection to the role of the researchers' subjectivity in making AI.

Neural network diagram

Three colorful GPUs with their packaging cleanly removed laying on a white surface

The GPU (Graphics Processing Unit) is an essential part of modern AI infrastructure. It’s a special type of chip or electronic circuit that is used in mobile phones, personal computers, game consoles, workstations and cloud computing infrastructure. Its close cousin is the CPU (Central Processing Unit). GPUs were originally designed to process images and render graphics. Nowadays they’re used for other computational tasks as well such as training neural networks in deep learning. GPUs can process many chunks of data simultaneously which is called parallelization. As of 2018 the companies NVIDIA and AMD control nearly 100% of the GPU market. These photographs are known as “die-shots“. A die is a small block of semiconducting material, such as Silicon, that serves as the supporting material on which an integrated circuit is fabricated. Die-shots are close-up photographs of computer chips, from which the “packaging“ is removed, usually by undergoing a quite dangerous etching process involving sulfuric acid and high temperatures. The typical bright rainbow colors in die-shots do not represent the actual colors of the chips, which are mainly gray and silver. The artist has used a combination of external light sources, polarising filters on the camera lense and image post production to create the colourful effect. This picture shows three NVIDIA Turing Chips (TU104, TU106, TU116)

GPU shot etched 2

An person is illustrated in a warm, cartoon-like style in green. They are looking up thoughtfully from the bottom left at a large hazard symbol in the middle of the image. The Hazard symbol is a bright orange square tilted 45 degrees, with a black and white illustration of an exclamation mark in the middle where the exclamation mark shape is made up of tiny 1s and 0s like binary code. To the right-hand side of the image a small character made of lines and circles (like nodes and edges on a graph) is standing with its ‘arms’ and ‘legs’ stretched out, and two antenna sticking up. It faces off to the right-hand side of the image.

An illustrated person looks up at a large hazard symbol, which has a character representing data science and AI ‘standing’ next to it. The AI character is made up of nodes and edges (like a network diagram). The hazard symbol is the General Data Hazard from the Data Hazards project, and is a large exclamation mark made up of binary code.

Here, the person is considering the Data Hazard and how they could use it to help them manage the risks the AI has been causing.

Managing Data Hazards

A block of silicon (also known as a mono-crystal) placed on a plain black background and photographed in HD to make its rich, reflective and complex surface visible.

This image shows a silicon mono-crystal on a neutral background, a crucial component of AI manufacture: a block like this would be sliced into 12 inch diameter wafers to form the base of CPUs. Picturing silicon visually asserts AI’s materiality and illustrates that the ‘mining’ of AI is not purely metaphorical (e.g. data mining) but also a literal, material undertaking. Responsible for sustaining innovation within  ‘Silicon valley’/wider tech, silicon enables the packing of greater computational/processing power into a single surface allowing for more efficient data movement, the co-location of memory and processing and parallelisation of function, necessary for machine learning. Catherine Breslin, the photographer, operates within the AI supply-chain first-hand in her work as a machine-learning voice engineer and consultant, previously involved in building Amazon Alexa.

Silicon on Black 1

A photographic rendering of a young black man standing in front of a cloudy blue sky, seen through a refractive glass grid and overlaid with a diagram of a neural network.

This image shows a young black man wearing a black coat staring past the camera in front of a blue cloudy sky. The scene is refracted in different ways by a fragmented glass grid. This grid is a visual metaphor for the way that new artificial intelligence (AI) and machine learning technologies can be used to extract and analyse behavioural and demographic data in innovative ways. Some of the grid squares reveal graphical interpretations of the man that exceed the capabilities of human vision, indicating how cutting edge technologies offer ways to augment traditional human understandings of complex phenomena. A neural network diagram is overlaid, familiarising the viewer with the formal architecture of AI systems.

Quantified Human

A young woman sitting in front of a computer in an office while another woman standing next to her is pointing at something on her screen. They are both dressed in casual clothes - a bright blue top and furry waistcoat, and a denim jacket.

Data annotation is becoming a professional field of its own, with many annotators working on it as full-time employees, following complex guidelines, and using a variety of different tools and platforms. What used to be considered 'menial' and 'low-skilled' work, is today a nascent field with its own complexities and skills requirements, which involves extensive training and specializing in different modalities: video labeling, medical imagery labeling, 3D LiDar labeling, and so on. It is important to recognize the value that data annotators bring and their expertise, in order to uplift them from the 'ghost work' position in which they are now.

A trainer instructing a data annotator on how to label images

A laptopogram based on a neutral background and populated by scattered squared portraits, all monochromatic, grouped according to similarity. The groupings vary in size, ranging from single faces to overlapping collections of up to twelve. The facial expressions of all the individuals featured are neutral, represented through a mixture of ages and genders.

This image shows face images from the Olivetti faces dataset, created at AT&T Laboratories Cambridge in the early 1990s. The portraits are scattered across the image and grouped by similarity. The image is a ‘laptopogram’, created by exposing photographic paper using a computer screen and developed in the artist’s bathtub. The process preserves a digital artifact of AI research in silver crystals, returning a physical dimension to sterile data. Dust, scratches, and the marks left by the artist’s hands draw a connection to the role of the researchers' subjectivity in making AI.

Data flock (faces)

Abstract microscopic photography of a Graphics Processing Unit resembling a satellite image of a big city

The GPU (Graphics Processing Unit) is an essential part of modern AI infrastructure. It’s a special type of chip or electronic circuit that is used in mobile phones, personal computers, game consoles, workstations and cloud computing infrastructure. Its close cousin is the CPU (Central Processing Unit). GPUs were originally designed to process images and render graphics. Nowadays they’re used for other computational tasks as well such as training neural networks in deep learning. GPUs can process many chunks of data simultaneously which is called parallelization. As of 2018 the companies NVIDIA and AMD control nearly 100% of the GPU market. These photographs are known as “die-shots“. A die is a small block of semiconducting material, such as Silicon, that serves as the supporting material on which an integrated circuit is fabricated. Die-shots are close-up photographs of computer chips, from which the “packaging“ is removed, usually by undergoing a quite dangerous etching process involving sulfuric acid and high temperatures. The typical bright rainbow colors in die-shots do not represent the actual colors of the chips, which are mainly gray and silver. The artist has used a combination of external light sources, polarising filters on the camera lense and image post production to create the colourful effect. This picture shows a “NVIDIA GeForce GTX 280”

GPU shot etched 5

A photographic rendering of a young houseplant against a neutral background, seen through a refractive glass grid and overlaid with an single neuron from a neural network diagram.

This image shows a houseplant against a neutral background. The scene is refracted in different ways by a fragmented glass grid. This grid is a visual metaphor for the way that artificial intelligence (AI) and machine learning technologies can be used to extract and analyse biological data in innovative ways, and the repetitive and self-similar structure of these systems. A single neuron from a neural network diagram is overlaid, revealing an augmented data view of the plant.

Plant

Silicon on Black 2

Silicon Closeup

Abstract microscopic photography of a Graphics Processing Unit resembling a floor plan or fractal art

The GPU (Graphics Processing Unit) is an essential part of modern AI infrastructure. It’s a special type of chip or electronic circuit that is used in mobile phones, personal computers, game consoles, workstations and cloud computing infrastructure. Its close cousin is the CPU (Central Processing Unit). GPUs were originally designed to process images and render graphics. Nowadays they’re used for other computational tasks as well such as training neural networks in deep learning. GPUs can process many chunks of data simultaneously which is called parallelization. As of 2018 the companies NVIDIA and AMD control nearly 100% of the GPU market. These photographs are known as “die-shots“. A die is a small block of semiconducting material, such as Silicon, that serves as the supporting material on which an integrated circuit is fabricated. Die-shots are close-up photographs of computer chips, from which the “packaging“ is removed, usually by undergoing a quite dangerous etching process involving sulfuric acid and high temperatures. The typical bright rainbow colors in die-shots do not represent the actual colors of the chips, which are mainly gray and silver. The artist has used a combination of external light sources, polarising filters on the camera lense and image post production to create the colourful effect. This picture shows a “Radeon R9 290” by AMD.

GPU shot etched 1

A laptopogram displaying a dataset as cloud-like clusters of black blobs on a neutral background. There are three larger collections, almost resembling a map, with some data points leaking out into the negative space.

This image shows a machine learning dataset visualized spatially, in cloud-like clusters according to visual similarity of the data. Although visualizations like this one always simplify and fail to represent the data’s true complexity and nuance, they guide the researchers’ intuitions for their subject matter. The image is a ‘laptopogram’, created by exposing photographic paper using a computer screen and developed in the artist’s bathtub. The process preserves a digital artifact of AI research in silver crystals, returning a physical dimension to sterile data. Dust, scratches, and the marks left by the artist’s hands draw a connection to the role of the researchers' subjectivity in making AI.

Data flock (digits)

Abstract microscopic photography of a Graphics Processing Unit in rainbow colors

The GPU (Graphics Processing Unit) is an essential part of modern AI infrastructure. It’s a special type of chip or electronic circuit that is used in mobile phones, personal computers, game consoles, workstations and cloud computing infrastructure. Its close cousin is the CPU (Central Processing Unit). GPUs were originally designed to process images and render graphics. Nowadays they’re used for other computational tasks as well such as training neural networks in deep learning. GPUs can process many chunks of data simultaneously which is called parallelization. As of 2018 the companies NVIDIA and AMD control nearly 100% of the GPU market. These photographs are known as “die-shots“. A die is a small block of semiconducting material, such as Silicon, that serves as the supporting material on which an integrated circuit is fabricated. Die-shots are close-up photographs of computer chips, from which the “packaging“ is removed, usually by undergoing a quite dangerous etching process involving sulfuric acid and high temperatures. The typical bright rainbow colors in die-shots do not represent the actual colors of the chips, which are mainly gray and silver. The artist has used a combination of external light sources, polarising filters on the camera lense and image post production to create the colourful effect. This picture shows a “Radeon R9 Nano” by AMD.

GPU shot etched 3

A photographic rendering of medical swabs and pills in a jar seen through a refractive glass grid, overlaid with a diagram of a neural network.

This image shows two labelled swab tubes with red caps and a glass jar full of blue and white pills, floating against a grey background and refracted in different ways by a fragmented glass grid. This grid is a visual metaphor for the way that new artificial intelligence (AI) and machine learning technologies can be used to extract and analyse medical data in innovative ways. Some of the grid squares reveal graphical interpretations of the objects that exceed the capabilities of human vision, which indicates how cutting edge technologies offer ways to augment traditional human understandings of complex phenomena. A neural network diagram is overlaid, familiarising the viewer with the formal architecture of AI systems.

Medicine

Two digitally illustrated green playing cards on a white background, with the letters A and I in capitals and lowercase calligraphy over modified photographs of human mouths in profile.

These illustrated cards are the backs of an original playing card deck assembling 56 A and I words and chess diagrams, part of a larger experimentation of analogue game investigations about A.I. It explores an alternative redefinition of A.I as units of a social language before mathematical construct. Its purpose is to look at some of the principles of machine learning like interpretation, classification, prioritization, as components of human interaction, before its encoding. Hands, mouths and handwritten typefaces are chosen as imagery to visualize human sources and choices expressed through our technical architectures. The associative and generative modularity of a playing card deck opens up a more timeless representation of A.I, to consider our relationship to technology differently.

Handmade A.I

Plastic figure resembling a human who sits on a table infront of a laptop in a dark room. Long shadows disseminate a gloomy mood.

Clickworkers are the invisible workforce behind today's AI systems. Without actual humans who do the tedious labor of labeling datasets, machine learning and its many real-life applications would be impossible. This image shows 3D-printed figures who work at a computer in an anonymous, dark environment. They are anonymized, almost de-humanized.

Clickworker 3d-printed

Plastic figures resembling humans who sit at tables in front of laptops. the lack of background makes their envoronment look bleak.

Clickworkers are the invisible workforce behind today's AI systems. Without actual humans who do the tedious labor of labeling datasets, machine learning and its many real-life applications would be impossible. This image shows 3D-printed figures who work at a computer in an anonymous environment. They are anonymized, almost de-humanized.

Clickworker Abyss

A woman and a man sitting in front of a computer screen, pointing at something on the screen and talking, with a colourful stncil design on the wall behind them.

Data annotation for AI is a subjective activity which involves a lot of interpretative work and communication between the annotators and the data scientists. Sometimes, when you are labeling a training dataset for AI, you bump into a lot of edge cases: occluded objects, blurry images, groups of many objects together, etc. How you annotate them can be crucial for ensuring that the resulting AI system is robust in cases of uncertainty. This is what we call the 'long tail' of AI edge cases and where human involvement is most needed.

Data annotators discussing the correct labeling of a dataset

A black keyboard at the bottom of the picture has an open book on it, with red words in labels floating on top, with a letter A balanced on top of them. The perspective makes the composition form a kind of triangle from the keyboard to the capital A. The AI filter makes it look like a messy, with a kind of cartoon style.

This image refers to Natural Language Processing, known as NLP, a sub-field of AI enabling computers to understand language. Chatbots, voice assistants, automatic translation, text summarization, speech recognition and synthetic voices are applications that fall under NLP. To be able to communicate we need to have a common language. The symbols, letters and words are the connection between the human and the AI, using these we can communicate with each other. This forms the basis of language models (and wider machine learning), embedded within statistical processes of pattern recognition. Words are placed, as in this image, within a symbolic taxonomy. The letter is a part of a word, which is a part of a sentence, which is a part of a context, which creates meaning. Yet these systems have no real understanding. The illusion of understanding emerges from the pattern they have learned. The single letter 'A', missing its counterpart 'I' could represent the actual lack of intelligence in these systems. The image was generated by applying a neural network-based technique called 'style transfer' using the mobile app Prisma.

Letter Word Text Taxonomy

Two women, one of whom is wearing a colorful hijab and looking down smiling, are sitting in front of computers and typing.

Data annotation is a key activity required in order to prepare training data for Artificial Intelligence. Manual data annotation is especially needed when dealing with unstructured data: images, videos, and text. For example, if we want to train an AI model to recognize animals on images, we will need the help of data annotators who would draw bounding boxes or add tags to each image depending on what animal it depicts. Data annotation work is being performed by hundreds of workers around the world who use the job as a means of livelihoods.

Data annotators labeling data

An illustration in intense colors in a gloomy mood showing a collage of two mirrored cars, street signs and mathematical symbols

This picture deals with self-driving cars, environment sensing and neural networks. A computer vision system implemented in a car needs to distinguish between important street signs and other non-important elements of the environment like chimneys.

Autonomous Driving

An illustration of three „pixelated“ cupboards next to each other with open drawers, the right one is black

The cupboard as a metaphor for a classification system/algorithm with the drawers representing different categorization classes. Black cupboard: The classification system itself can be a black box, meaning that the mechanisms by which a data item is classified into a specific class are often hidden even to the developers of these systems.

Classification Cupboard

An illustration containing electronical devices that are connected by arm-like structures

This image highlights AI as an extension of human capabilities. Embedded in a variety of devices it can enrich our lives as a tool or partner that adapts to our needs.

Human-AI collaboration

A computer generated photographic style image showing a piles of distorted books with some surreal landscape features in the immediate foreground, such as a kind of beach and games board. The closed books, many stacked, have a range of different colours (with blue and red palate colours most dominant) with the white pages visible as they are mainly placed flat. The books merge into each other in an impressionistic, digitally blurred way, and rising out of them and taking up the main part of the image is a huge undefined concrete structure rising out of the books. It has more, larger mulitcoloured digitally impressionist books heaped on top of it in a huge pile. This structure dominates the image like a monument with a pale grey and light blue cloudy sky natural sky in the background. In the distance, another monument can be vaguely seen, a needle shape.

Image created for eleuther.ai using a text-to-image AI algorithm. The prompt given was “a monument surrounded by piles of books”, to represent the eleuther.ai dataset called “The Pile”, which constitutes a large, diverse, open source dataset made of many smaller sets, including sources such as Wikipedia. The resulting image of a pile of books emerging from a stone structure works well to communicate the amounts of textual data that go into contemporary AI language models like GPT-3. The image therefore references language processing, language data sets, and also how neural networks can be used creatively, like the artist did here with BigSleep.

A monument surrounded by piles of books

Nine small images with schematic representations of differently shaped neural networks, a human hand making a different gesture is placed behind each network.

This image is made of screenshots of an educational video explaining neural networks. It can be seen as a metaphor for the idea of „explainable AI“ (XAI), a field of research that deals with the „black box“ problem of Machine Learning: Until now we often can’t really say how a neural network comes to its decisions and which are the underlying rules. Explainable AI seeks to find methods for making the AI decision making process more transparent and understandable.

Explainable AI

8 photos of the same tree taken at different times of the year in 2 rows of 4, the last photo is highly pixelated. A pattern of random white blocks run across the image from the left and become aligned on the right.

This image represents the process of machine learning in object recognition. The trees are training data and the white blocks/pattern which become more aligned indicate progress in the machine's objectives. The pixelated tree is a reminder that machines do not see like we humans see.

Trees

A photographic rendering of a succulent plant seen through a refractive glass grid, overlaid with a diagram of a neural network.

This image shows a low angle of a succulent plant against a hazy sky. The scene is refracted in different ways by a fragmented glass grid. This grid is a visual metaphor for the way that artificial intelligence (AI) and machine learning technologies can be used to extract and analyse biological data in innovative ways, and the repetitive and self-similar structure of these systems. A neural network diagram is overlaid, familiarising the viewer with the formal architecture of AI systems.

Nature

A photographic rendering of a smiling face emoji seen through a refractive glass grid, overlaid with a diagram of a neural network.

This image shows a 3D rendering of a smiling face emoji against a black background. The scene is refracted in different ways by a fragmented glass grid. This grid is a visual metaphor for the way that artificial intelligence (AI) and machine learning technologies can be used to extract and analyse social data in complex ways, and the repetitive and self-similar structure of these systems. A neural network diagram is overlaid, familiarising the viewer with the formal architecture of AI systems.

Social Media

A photographic rendering of a simulated middle-aged white woman against a black background, seen through a refractive glass grid and overlaid with a distorted diagram of a neural network.

This image shows a portrait of a simulated middle-aged white woman against a black background. The scene is refracted in different ways by a fragmented glass grid. This grid is a visual metaphor for the way that artificial intelligence (AI) and machine learning technologies can be used to simulate and reflect the human experience in unexpected ways. A distorted neural network diagram is overlaid, familiarising the viewer with the formal architecture of AI systems.

Virtual Human

A banana on a table and an image of a banana on a laptop on the same table. Each of the two bananas has a white frame around it with the word ‚Banana‘ sticked on top of it

Bounding boxes are commonly used in AI research to signify where a computer vision algorithm has detected an object in an image. Here the artist has played with this aesthetic: The bounding boxes are 3D-printed frames positioned in the physical environment around objects. In this case there is a physical banana and a photo of a banana on a screen. The object detection doesn’t make a difference between these two.

Ceci n'est pas une banane

A banana, a plant and a flask on a monochrome surface, each one surrounded by a thin white frame with letters attached that spell the name of the objects

Bounding boxes are commonly used in AI research to signify where a computer vision algorithm has detected an object in an image. Here the artist has played with this aesthetic: The bounding boxes are 3D-printed frames positioned in the physical environment around objects. Sometimes the objects stick out of their frame.

Banana / Plant / Flask

A window of three images. On the right is a photo of a big tree in a green field in a field of grass and a bright blue sky. The two on the left are simplifications created based on a decision tree algorithm. The work illustrates a popular type of machine learning model: the decision tree. Decision trees work by splitting the population into ever smaller segments. I try to give people an intuitive understanding of the algorithm. I also want to show that models are simplifications of reality, but can still be useful, or in this case visually pleasing. To create this I trained a model to predict pixel colour values, based on an original photograph of a tree.

The work illustrates a popular type of machine learning model: the decision tree. Decision trees work by splitting the population into ever smaller segments. I try to give people an intuitive understanding of the algorithm. I also want to show that models are simplifications of reality, but can still be useful, or in this case visually pleasing. To create this I trained a model to predict pixel colour values, based on an original photograph.

Decision Tree

Decision Tree reversed

A window of three images. On the right is a photo of a man in a dark t-shirt and light trousers walking in a narrow street. The curving street is lined with brightly coloured terraced houses facing straight onto the pavement, with green trees in large pots on the other side. The images left and in the middle consist of multiple colored semi transparent polygons, made from the key colours of the mans clothes, the houses and the plants. These polygons were randomly initiated, then randomly mutated. However, only if the mutation made the image look more like the original it was kept, otherwise the image was reverted to the previous step. The image on the right is the original, the image on the left is an early stage of the evolution process, the image on the right is a late stage of the evolution process.

The work illustrates a popular type of algorithm called a genetic evolution. The image consists of multiple colored semi transparent polygons. These polygons were randomly initiated, then randomly mutated. However, only if the mutation made the image look more like the original it was kept, otherwise the image was reverted to the previous step. The image on the right is the original, the image on the left is an early stage of the evolution process, the image on the right is a late stage of the evolution process. Inspired by Roger Johansson's work on genetic programming evolutino of mona lisa.

Man reversed

A yellow window with three images. On the right is an under water photo of a single bright blue and yellow tropical fish. The two on the left are simplifications created based on a decsion tree algorithm. The work illustrates a popular type of machine learning model: the decision tree. Decision trees work by splitting the population into ever smaller segments. I try to give people an intuitive understanding of the algorithm. I also want to show that models are simplifications of reality, but can still be useful, or in this case visually pleasing.

Fish

Fish reversed

A grey window with three images. On the right is a zoomed in photo of a red sports car with a white background. The two on the left are simplifications created based on a decsion tree algorithm. The work illustrates a popular type of machine learning model: the decision tree. Decision trees work by splitting the population into ever smaller segments. I try to give people an intuitive understanding of the algorithm. I also want to show that models are simplifications of reality, but can still be useful, or in this case visually pleasing.

Images of AI library

AI across industries - Yasmin Dwiputri & Data Hazards Project

Neural network diagram - Philipp Schmitt

GPU shot etched 2 - Fritzchens Fritz

Managing Data Hazards - Yasmin Dwiputri & Data Hazards Project

Silicon on Black 1 - Catherine Breslin

Quantified Human - Alan Warburton

A trainer instructing a data annotator on how to label images - Nacho Kamenov & Humans in the Loop

Data flock (faces) - Philipp Schmitt & AT&T Laboratories Cambridge

GPU shot etched 5 - Fritzchens Fritz

Plant - Alan Warburton

Silicon on Black 2 - Catherine Breslin

Silicon Closeup - Catherine Breslin

GPU shot etched 1 - Fritzchens Fritz

Data flock (digits) - Philipp Schmitt

GPU shot etched 3 - Fritzchens Fritz

Medicine - Alan Warburton

Handmade A.I - Alina Constantin

Clickworker 3d-printed - Max Gruber

Clickworker Abyss - Max Gruber

Data annotators discussing the correct labeling of a dataset - Nacho Kamenov & Humans in the Loop