memo.tv

Machine learning is a way of building software that learns patterns from data instead of following rules a person wrote by hand. You use it when the relationship between input and output is too complex or too variable to specify directly — recognizing objects in photos, ranking search results, or predicting whether a transaction is fraudulent. The core idea: show the system many examples, let it adjust internal parameters to reduce its errors, then check whether it works on examples it has never seen.

Learning from data vs. hand-coded rules

In traditional programming, a developer writes explicit logic: if the email contains these words, mark it spam. In machine learning, you supply labeled examples and the model derives its own decision boundary. The trade-off is that the model's behavior depends on the data it saw — change the data, and the behavior changes.

The three main learning paradigms

Supervised learning covers most everyday applications. Unsupervised learning is used for clustering, compression, and anomaly detection. Reinforcement learning is harder to stabilize but is the approach behind recent work on scaling language-model reasoning — for instance, Qwen's GSPO research explicitly targets "stable and robust training dynamics" for RL at scale, noting that existing algorithms such as GRPO "exhibit severe instability issues during" training.

The core training loop

Every supervised model follows roughly the same cycle:

1. Collect and split data. Divide examples into a training set and a held-out test set.
2. Define a model. Choose an architecture with adjustable parameters (weights).
3. Measure error with a loss function. The loss quantifies how far predictions are from the correct answers.
4. Adjust parameters. An optimization algorithm nudges the weights to reduce the loss.
5. Repeat. Iterate over the data many times until the loss stops improving.
6. Evaluate on unseen data. Measure performance on the test set, not the training set.

The input is the data and the model definition; the action is repeated parameter updates; the expected result is a model whose error on new data is acceptably low.

Overfitting, underfitting, and why splits matter

• Underfitting: the model is too simple to capture the pattern — it performs poorly on both training and test data.
• Overfitting: the model memorizes the training examples, including their noise — it performs well on training data but poorly on test data.

This is why you never judge a model by its training accuracy. A held-out test set (or cross-validation) simulates the real world: data the model has not seen. If training error keeps falling while test error rises, you are overfitting.

Where deep learning and large language models fit

Deep learning is machine learning using neural networks with many layers. It is not a separate field — it is a subcategory that excels when data is abundant and patterns are hierarchical (images, audio, text).

Large language models are deep learning models trained on massive text corpora, usually with a self-supervised objective: predict the next token. That objective needs no human labels, which is why it scales. The Qwen family illustrates the breadth of the umbrella — its releases include a 20B image foundation model (Qwen-Image) for text rendering and editing, a safety classifier (Qwen3Guard) fine-tuned for prompt and response moderation, and RL research (GSPO) for training dynamics. All of these are machine learning systems; they differ in data, objective, and architecture, not in kind.

How to tell the paradigms apart in practice

Ask two questions:

1. Does the training data include the correct answer? If yes, it is supervised (or self-supervised, where the answer is derived from the data itself).
2. Does the model learn by taking actions and receiving feedback? If yes, it is reinforcement learning.

If neither applies and you are only looking for structure, it is unsupervised. Most real systems combine these — a language model may be pretrained with self-supervision, fine-tuned with supervised examples, and refined with reinforcement learning.

AI art is artwork in which machine learning or other AI systems play a substantive role in how the work is generated, behaves, or responds—not just a tool used to retouch a finished piece. It typically involves an artist choosing data, designing rules or training processes, tuning a model, and curating outputs, rather than pressing a single "generate" button. The term covers generative images, interactive installations, real-time performances, and web-based works. It overlaps with generative art and computational art but is distinguished by its reliance on learned models rather than hand-written rules alone.

Defining AI art and its boundaries

The label "AI art" is often applied loosely. A useful test is to ask what the AI system actually contributes.

• Generative art uses a set of rules or algorithms—often hand-authored—to produce variation. The artist writes the system; the system explores its own possibility space.
• Computational art is a broader category: any art made with computation, including simulation, data visualization, and procedural graphics.
• Machine learning art is a subset of AI art where the system learns patterns from data rather than following explicit instructions. This is where neural networks, training runs, and datasets enter.
• AI art is the widest of these terms, covering machine learning art plus systems built on other AI techniques (search, planning, agent-based behavior).

So a work can be generative without being AI-based, and AI-based without being purely generative—for example, an interactive installation that classifies a visitor's movement in real time.

Main creative paths

There is no single workflow. Most AI art falls into one of three approaches, and artists often combine them.

1. Training a model on chosen data

The artist assembles a dataset, trains a model (or fine-tunes an existing one), and works with what the model produces. The dataset is a creative decision: what is included, what is excluded, and how it is labeled all shape the output. This path tends to be slower and more resource-intensive but gives the artist control over the model's "vocabulary."

2. Using pre-trained models

The artist calls an existing model and directs it through prompts, parameters, or input images. This is the most accessible entry point and the source of much of the "one-click generation" narrative. The creative work shifts toward selection, iteration, composition, and critique rather than model building.

3. Treating the model as a live system

Here the model runs continuously as part of an installation, performance, or web piece. Inputs might be camera feeds, sensor data, sound, or audience interaction; the model's output changes in response. This is common in interactive installation and real-time performance work, where the artwork is the behavior of the system over time, not a fixed object.

What the artist actually does

Across all three paths, the artist's role is less about manual rendering and more about decisions that shape the system:

• Selecting and curating data — deciding what the model learns from.
• Designing rules and constraints — setting the boundaries within which the system operates.
• Tuning parameters — adjusting how the model behaves, often iteratively.
• Curating outputs — choosing which results are shown, and how.
• Framing and critique — using the work to comment on perception, agency, bias, or human-machine relations.

This is why AI art is often described as a collaboration between artist and system, with the artist setting the conditions rather than dictating every pixel.

Common forms of AI art

Memo Akten's practice, for example, spans interactive software, web-based work, and themes such as consciousness, dance, ecology, mind, and simulation—showing how AI and computational methods can be applied to very different subject matter rather than a single aesthetic.

How to tell whether a work is AI art

Ask four questions:

1. Is there a model? Does the system learn from data, or is it purely rule-based?
2. What data? Where does the training or input data come from, and who chose it?
3. What does the AI do? Does it generate, classify, respond, or drive behavior?
4. What is the artist's intent? Is the AI central to the concept, or incidental to production?

If the AI is doing meaningful work in generation or behavior, and the artist has shaped its conditions, it is reasonable to call it AI art.

Common misunderstandings and limits

• "It's just one click." Pre-trained models make generation fast, but the surrounding decisions—data, prompts, selection, context—are where the art is made.
• Copyright and data provenance. Training data is often scraped or sourced in ways that raise unresolved legal and ethical questions. These vary by jurisdiction and are not settled.
• Bias. Models reproduce patterns in their training data, including harmful ones. Artists and researchers often treat this as material to examine rather than a bug to ignore.
• Authorship and agency. Who is the author when a model produces the output? Practices differ, and there is no universal answer.

Understanding these limits is part of understanding the field, not a reason to dismiss it.

Memo Akten is a multidisciplinary artist, researcher, and computer scientist who works with code, data, AI, and emerging technologies to explore perception, consciousness, and human-machine relations. If you want to understand what he actually makes and whether his work is relevant to your interests in AI art, computational art, or new media art, the short answer is: he builds generative, interactive, and machine-learning-driven works that treat computation as a medium for asking questions about how humans and machines perceive and relate to each other.

The Three Roles, and Why They Matter Together

Akten's practice is usually described through three overlapping identities:

• Artist — he produces works shown as installations, software, and web-based pieces.
• Researcher — he investigates how perception, consciousness, and machine behavior can be modeled or provoked computationally.
• Computer scientist — he writes the code and builds the systems himself, rather than directing engineers.

The combination matters because his artworks are not illustrations of research findings, and his research is not separate from the artwork. The technical implementation is part of the argument. For example, a piece that uses machine learning to generate movement is simultaneously an artwork and a probe into how models interpret human gesture.

What He Works With

His stated materials are code, data, AI, and emerging technologies. In practice this places his work in several recognizable categories:

The site's featured projects section and a 2024 video introduction are the two main entry points for seeing this range in practice. The excerpted project keywords — interactive software, web-based, consciousness, dance, ecology, mind, simulation, ants — suggest the breadth: from simulation and swarm-like behavior (ants) to embodied human movement (dance) to ecological and consciousness themes.

The Recurring Themes

Three themes run across the work:

1. Perception — how seeing, sensing, and interpreting actually happen, in humans and in machines.
2. Consciousness — what it means to have an inner model of the world, and where the analogy between minds and models breaks down.
3. Human-machine relations — how people and computational systems shape each other, especially when the system is generative or adaptive.

These are not decorative topics. They determine the choice of technique: if you want to ask how a machine "sees" movement, you build a system that learns from movement data and let the audience watch it succeed and fail.

How to Explore His Work

If you want to judge whether his practice is worth following or citing:

1. Start with the 2024 video introduction on the site — it gives his own framing of the practice in a few minutes.
2. Then scan the featured projects — look for the medium (installation, web, software) and the theme (dance, ecology, simulation) that matches your interest.
3. Note the technique behind each piece — whether it uses machine learning, simulation, or interaction — because that is where the research and the art connect.
4. Use the project keywords as a map — terms like generative art, machine learning, interactive installation, and new media art describe both his methods and the communities his work sits within.

A practical example: if you are writing about AI art and need a practitioner who both builds models and writes about what they mean, Akten fits that gap — he is not only using AI tools but examining the perceptual and relational assumptions built into them.

Who This Is For

His work is most relevant if you are interested in:

• The intersection of art and machine learning, beyond image-generation demos
• Interactive and installation-based new media art
• Research-informed artistic practice where code is the primary material
• Questions about perception and consciousness approached through computational systems

If you are looking for conventional painting, commercial AI tooling, or purely theoretical writing without built works, this is a different kind of practice. The distinguishing feature is that the systems are made and shown, and the conceptual claims are tested through them.

Memo Akten is a multidisciplinary artist, researcher, and computer scientist who works with code, data, AI, and emerging technologies to explore perception, consciousness, and human-machine relations. If you are trying to place his practice — is he an engineer who makes art, or an artist who writes code? — the short answer is that he deliberately refuses the split. His work treats computation as a medium for asking questions about how minds, bodies, and machines perceive and relate to one another.

Three roles, one practice

The three labels in his own description are not separate careers:

• Artist — he produces works shown as installations, web-based pieces, and interactive software.
• Researcher — he investigates perception and consciousness, often through the same systems he builds.
• Computer scientist — he writes the code and works directly with machine learning and data.

The point of the combination is that the technical construction is part of the artistic argument. A model trained on data is not just a tool he applies to a finished idea; the behavior of the model is part of what the work is about.

The themes: perception, consciousness, human-machine relations

These three themes recur across his projects and are worth understanding as a set rather than as separate topics.

Perception concerns how a viewer or a system interprets what it receives. When a machine classifies an image or a sound, it is doing a version of what human perception does — and the differences are revealing.

Consciousness is the harder question underneath: what would it mean for a system to have an inner life, and how do we recognize it? He approaches this through simulation and behavior rather than through argument alone.

Human-machine relations is the practical layer: how we project intention onto systems, how we collaborate with them, and what that does to how we understand ourselves.

A useful way to see the connection: if you train a system on data and it produces something that feels alive or intentional, the interesting question is not whether it is alive, but why you experienced it that way. That gap between mechanism and impression is where much of his work sits.

How code, data, and AI function as materials

In this kind of practice, machine learning is not a shortcut for producing images. It is a material with specific properties:

• Data carries the biases and structure of whatever it was collected from, so the source of the data shapes the outcome.
• Models learn patterns rather than rules, which means their behavior can surprise the person who built them.
• Code makes the system reproducible and inspectable, which matters when the work is also a research claim.

For a concrete task, imagine building a piece that responds to a viewer's movement. You would need a sensing input (camera or microphone), a model or rule set that interprets that input, and an output (sound, image, or physical response). Each stage introduces its own kind of uncertainty, and that uncertainty is often the subject of the work rather than a problem to eliminate.

What his projects look like

His featured work spans several formats, which is typical for a practice that moves between research and exhibition:

The subject matter across these formats has included dance, ecology, mind, and simulation — for example, systems built around ants or other collective behaviors, where simple rules produce complex group activity. That kind of project is a good illustration of the general principle: the artist sets up conditions, and the system generates outcomes that neither the artist nor the viewer fully controls.

Where to start

If you want to understand his work directly rather than through description, the site offers two natural entry points:

1. Video Introduction (2024) — the fastest way to hear the practice described in his own terms.
2. Featured Projects — the concrete works, which show how the themes above appear in actual pieces.

Start with the video if you want the framing, then open the featured projects and look for the format that interests you most — interactive, web-based, or installation. The themes will read differently depending on which format you enter through, and that difference is itself informative about how he thinks about the relationship between a system and the person encountering it.

Being an artist means sustaining a working practice, not just producing occasional pieces. The clearest way to see the difference is through a painter like Holly Farrell, a Canadian artist who lives and works in Toronto. Her practice shows how subject matter, routine, and intent separate a working artist from a hobbyist: she paints the common tools of day-to-day existence, and she treats those still lifes as meditations on people and places she has known.

Working artist vs. hobbyist

The line is not talent or income. It is whether the work is organized around a continuing practice.

For Farrell, the through-line is explicit: nostalgia drives her subject matter. That single stated driver explains why her paintings return to ordinary objects rather than ranging across unrelated themes.

Still life as portraiture

The most useful idea in Farrell's practice is that still life can function as portraiture. She considers her still life paintings to be simple meditations on people and places she has known.

That reframes what a still life is for. Instead of a study of objects, it becomes a way of holding a person or place in view through the things associated with them. The objects are not the subject so much as the evidence of the subject.

How to read a still life this way

• Ask what person, place, or period the objects point to, rather than what they depict literally.
• Notice which objects are chosen and which are left out — selection is the portrait.
• Look at how ordinary the items are. Farrell's subjects are the common tools of day-to-day existence, which keeps the reference personal rather than monumental.
• Treat the painting as a meditation, not a document. It records a feeling about someone, not a likeness.

Why nostalgia and daily objects sustain a practice

Nostalgia is often dismissed as sentiment. In a working practice it does real structural work:

• It supplies an inexhaustible subject. Everyday tools and domestic objects are everywhere, so the artist never runs out of material.
• It creates consistency. A single emotional driver keeps a body of work coherent across years.
• It lowers the barrier to starting. The artist does not need a grand theme to begin a painting — only an object tied to a memory.
• It gives the viewer an entry point. Everyone has ordinary objects charged with personal meaning, so the work reads without explanation.

If you are building your own practice, this is the transferable lesson: pick a driver specific enough to repeat and broad enough to never exhaust.

What a studio-based painting career looks like day to day

A studio practice is less romantic and more administrative than it appears. Based on how Farrell's practice is described — living and working in Toronto, with painting as her occupation — the recurring elements are:

1. Studio time as a default. The studio is where the working day happens, not a place visited when inspired.
2. A settled subject territory. Time goes into exploring one area deeply rather than searching for a new theme each time.
3. Slow accumulation. A body of work builds piece by piece; no single painting carries the career.
4. Presentation and context. Work is shown and explained — the phrase "still life as portraiture" is itself a framing that helps viewers understand the intent.
5. Place as input. Living and working in one city means the people and places feeding the nostalgia are close at hand.

How to look at this kind of work

If you encounter still life painting that reads as portraiture rather than decoration, use these questions:

• What is the emotional temperature — warm, elegiac, matter-of-fact?
• Are the objects elevated or left plain? Plainness usually signals personal memory rather than display.
• Does the composition feel like a portrait's framing — a single subject given full attention?
• What would you need to know about the artist's life to complete the reading?

The last question is the point. Still life as portraiture asks the viewer to supply the relationship, which is why the artist's own statement — that these are meditations on people and places she has known — matters more than any visual analysis.

The takeaway

Being an artist, in this model, means choosing a subject you can return to for years, tying it to something you actually feel, and showing up in the studio to work through it. Farrell's practice demonstrates the whole chain: nostalgia as the driver, everyday objects as the material, still life as the form, and portraiture as the meaning.