tanteOct 15, 2023

Mitchell Baker from Mozilla tells the German news agency that we shouldn't "leave AI development to the tech-giants", that somehow "the training data should be controlled by users".

And that does on a surface level sound nice. User control and criticizing tech giants. Great. But it accepts the tech-giants' narrative: That "AI" is inevitable and that all data *has to* be dumped into these systems, that unfairness is a tech issue to solve with more data. That is not true.

We can also decide not to build these huge and wasteful statistical systems.

Show threadtanteOct 15, 2023

Sure there are useful machine learning tools, I spend parts of my day building some, but these overreaching, all-encompassing, general models are a mistake. They promise things that they'll never deliver, are untestable, unverifiable, massively wasteful, express systemic violence (and all the other structural issues @timnitGebru , @mmitchell_ai, @emilymbender and Angelina McMillan-Major have written about).

If there is something in machine learning, it is in small, focused, testable networks.

Show threadRhymer

@tante @timnitGebru @mmitchell_ai @emilymbender

Let's see if the essential questions are discussed on *todays* panel (appr. 19:00 CEST) with #RobertHabeck @BMWK and #EthicsInIT Prof. Judith Simon from #HamburgUniversity. and #JonasAndrulis from #AlephAlpha

https://social.bund.de/@BMWK/111222555658901593

Wirtschaft & Klimaschutz (@[email protected])

Attached: 2 images Sie haben Fragen zum Thema Künstliche Intelligenz? Beim 4. #Transformationsgespräch wird Bundesminister Robert #Habeck mit Prof. Dr. Judith Simon und Jonas Andrulis auf Fragen aus der Community eingehen. Stellen Sie jetzt Ihre Fragen in den Kommentaren! #KI Livestream und mehr Informationen hier: https://www.bmwk.de/Redaktion/DE/Livestream/2023/10/2023-10-16-4-gespraech-zur-transformation/4-gespraech-zur-transformation-wie-gelingt-gute-kuenstliche-intelligenz.html 

social.bund.de
Oct 16, 2023 at 4:51amWeb
Show threadtanteOct 16, 2023
@rhymer @timnitGebru @mmitchell_ai @emilymbender @BMWK again the question of the "if" is already answered. It's merely about the how. That is exactly what I was talking about.
Show threadtanteOct 16, 2023
@rhymer @timnitGebru @mmitchell_ai @emilymbender @BMWK especially given that Mr. Andrulis' company got some press recently due to their refusal to filter the (racist) outputs of their large language model. What kind of panel is that gonna achieve?
Show threadRhymerOct 16, 2023

@tante

Wouldn't that be a very good reason to discuss the "if"-question, in that panel with the party involved present?

@timnitGebru @mmitchell_ai @emilymbender @BMWK

Show threadRhymerOct 16, 2023

@tante

The panel organizers asked publicly for questions from the interested public and to-be audience. I do not know if they (would) have accepted the "if" question to be discussed tonight.

@timnitGebru @mmitchell_ai @emilymbender @BMWK