In dem in der FAZ erschienenen Text, Was der Künstlichen Intelligenz noch fehlt malt Kurzweil ein Bild, welches phantastischer nicht sein könnte: „Wie die vergangenen Jahre gezeigt haben, sind wir auf dem Weg, die Fähigkeiten des Neokortexes nachzubilden, bereits ein gutes Stück vorangekommen. Die letzten Defizite der KI lassen sich in verschiedene Hauptkategorien unterteilen, insbesondere: Kontextgedächtnis, Weltwissen und soziale Interaktion.“ Diese einleitenden Sätze setzen bereits eine Prämisse, welche sogenannte Künstliche Intelligenz, die eigentlich statistische Induktion genannt werden sollte, sprachlich in die Nähe zum menschlichen Denken rückt. Kurzweils erstem Satz widerspricht beispielsweise der renommierte französische Informatiker Francois Chollet, der die Deep-Learning-Bibliothek Keras initiiert hat, welche das Trainieren sogenannter Künstlicher Neuronaler Netze ermöglicht. Chollet hat deutlich gemacht, dass jene technischen Funktionen, die den derzeitigen Erfolg Künstlicher Neuronaler Netze ausmachen, wie zum Beispiel Backpropagation, Rectified Linear Units (Relu), Dropout Layers, Batch Normalization oder Layer Normalization vor allem der Optimierung von Rechenleistung dienen und in dieser Form im menschlichen Gehirn so überhaupt nicht existieren. Kurzweils Unterstellung, dass derzeitige Entwicklungen die menschliche Neokortex nachbilden, ist auch für die derzeit sehr erfolgreichen Bild-generierenden Modelle, sogenannte Transformer-Architekturen schlichtweg Unsinn. Ray Kurzweil lügt.

Er schreibt, dass es „letzte“ Defizite gäbe und nennt drei Beispiele: Kontextgedächtnis, Weltwissen und soziale Interaktion. Er behauptet also, der Hauptteil wäre bereits getan, und es sei nur noch ein Rest, ein paar Kleinigkeiten zu erledigen, um die Fähigkeiten des Neokortexes nachzubilden. Angesichts einer derart dreisten Verdrehung der Tatsachen, tut es weh ansehen zu müssen, dass die FAZ dieser pseudowissenschaftlichen Behauptungen überhaupt druckt. Kontextgedächtnis, Weltwissen und soziale Interaktion gehören zu den fundamentalen Problemen, welche die derzeitigen Statistikverfahren des Machine Learning von der Nachbildung menschlicher Denkfähigkeiten unterscheiden. Kurzweils Text ist durch und durch falsch und verdreht. Ray Kurzweil lügt

Literatur

Large Models of What? Mistaking Engineering Achievements for Human Linguistic Agency.
Abeba Birhane / Marek McGann https://arxiv.org/pdf/2407.08790

The empty brain. Your brain does not process information, retrieve knowledge or store memories. In short: your brain is not a computer.
Robert Epstein https://aeon.co/essays/your-brain-does-not-process-information-and-it-is-not-a-computer

On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜
Emily M. Bender, Timnit Gebru, Angelina McMillan-Major, Shmargaret Shmitchell, https://dl.acm.org/doi/10.1145/3442188.3445922

Unhype Artificial Intelligence! A Proposal to Replace the Deceiving Terminology.
Hunger, Francis https://doi.org/10.5281/zenodo.7524493.

First published on nettime mailing list on January 7, 2024 in reply to Olia Lialina: “I’ve built a simple AI”. Remembering the last AI spring https://pad.profolia.org/s/I_ve_built_a_simple_AI (read first)

Dear Olia,

thank you so much for your pointed observations. I would like to extend on both Sophia and Jarvis. Both seem to be driven by control and automation phantasies on the one hand and anthropmorphization on the other. “I’ve built a simple AI” already uses this “a AI”, ascribing agency (with the implication it would be human-like agency) to it, obscuring the human labor behind it and the machinic limitations it necessarily has.

The Team behind Sofia, for instance, consists of the CEO, the animation and interaction designers, the software enginieers, the content developers, robotic face architect, robotics engineers, a production supervisor, a language processing A”I” engineer, IT-Manager, office staff, events coordinator and freelance project workers according to their website. A team of at least 23 humans to get an elaborated deception, a version of a mechanical turk, running.

Interestingly Sofia is this embodied, female slave, while Zuckerbergs male voiced Jarvis is disembodied and omnipresent in his house.

Francis Hunger, December 7, 2022

Introduction

For the last time in human history the cultural-data space has not been contaminated. In recent years a new technique to acquire knowledge has emerged. Scraping the Internet and extracting information and data has become a new modus for companies and for university researchers in the field of machine learning. One of the currently largest publicly available training data sets to combine images and labels (which shall describe the images content), is Laion-5B, with 5,85 billion image-text pairs (Ilharco, Gabriel et al. 2021).[1]
The scope of scraping internet resources has become so all-encompassing, that researcher Eva Cetinic has proposed to call this form ‘cultural snapshot’: “By encoding numerous associations which exist between data items collected at a certain point in time, those models therefore represent synchronic assemblages of cultural snapshots, embedded in a specific technological framework. Metaphorically those models can be considered as some sort of encapsulation of the collective (un)conscious […]” (Cetinic 2022).[2] The important suggestion which Cetinic makes, is that these data collections are temporally anchored. The temporal dimension of these snapshots suggests that digital cultural snapshots taken at different times document different states of (online-)culture. So how will a 2021 snapshot differ from a 2031 cultural snapshot?

Consequences

Multi-modal models, like CLIP, trained on large-scale data sets, such as LAION-5B provide the statistical means to generate images from text prompts. In the CLIP Model, pre-trained models merge two embedding spaces, one for images and one for text-descriptions which with mathematical methods get layered together, so that the vectors in the one space, the image domain, align with vectors in the other space, the text domain, assuming there is a similarity between both, and one can translate into the other. In three short examples I’ll discuss some of the consequences of the underlying data for large-scale models from the perspective of cultural snapshots.

Interviews with Magda Tyzlik-Carver (curator), Adam Harvey (artist), Ulises Mejias & Nick Couldry (sociologists) and Matteo Pasquinelli (philosopher)

addressing questions of data, curating, artificial intelligence, data colonialism and face recognition data sets.

In the frame of »Training the Archive« (2020–2023), a research project that explores the possibilities and risks of Artificial Intelligence in relation to the automated structuring of museum collection data to support curatorial practice and artistic production.

Huge thanks to Inke Arns and Dominik Bönisch, who make it possible. https://trainingthearchive.ludwigforum.de/

Englisch mit deutschen Untertiteln

Working Paper Series 3, November 2021 https://zenodo.org/record/5705769 (PDF, open DL)

The concept of post-AI curating discussed in this Working Paper explores curation as a knowledge-creation process, supported by Pattern Recognition and weighted networks as technical tools of artificial ‘intelligence’.

It then examines several projects as case studies that approach curation using artificial ‘intelligence’: The Next Biennial Should Be Curated by a Machine from UBERMORGEN, Leonardo Impett and Joasia Krysa (2021) as a meta-artwork about curation and biennials; Tillmann Ohm’s project Artificial Curator (2020), which resulted in an automatically curated exhibition; and #Exstrange by Rebekah Modrak and Marialaura Ghidini et. al. (2017), which presents artworks as data objects on the eBay online platform.

German Version:

Kuratieren und dessen statistische Automatisierung mittels Künstlicher ‘Intelligenz’. Training The Archive, Working Paper Series 3, Oktober 2021 https://zenodo.org/record/5589930 (PDF, open DL)