On Fri, Jun 12, 2026 at 4:20 PM Matt Mahoney <[email protected]> wrote:
> ...My issue was about data. What data would you compress > >From the Hume's Guillotine README: Yes, people can *and will* argue over what data to include or exclude -- but the Algorithmic Information Criterion traps the intellectually dishonest by making their job much harder since they must include *exponentially* much more data that is biased towards their particular agenda in order to wash out data coherence (and interdisciplinary consilience) in the rest of the dataset. The ever-increasing diversity of data sources *identifies* the sources of bias -- and then starts predicting the behavior of data sources in terms of their bias, *as* bias. Trap sprung! This is much the same argument as that leveled against conspiracy theories: At some point it becomes simply impractical to hide a lie against the increasing diversity of observations and perspectives. (and earlier) Out of all so-called "Information Criteria" for model selection, the Algorithmic Information Criterion is the best we can do in scientific discovery of causality *relative to a given set of observations*. "Many analysts, one dataset" <https://www.socialjudgments.com/docs/Silberzahn_Uhlmann_Martin_et_al_in_press_Many_Analysts.pdf>, the current best practice in meta-scientific research hence scientific ethics, is one step shy of the Algorithmic Information Criterion for causal model selection. That step? Understand that the best analysts will be able to *losslessly compress* the "one dataset" best. This is how you make competition fair, objective and optimal between schools of thought. > to answer questions like "how many weeks after conception should > *abortion* remain legal?" Or "what is the optimal *immigration* policy?" > Again from the README: the real world isn't *static* nor composed of disconnected systems that have a single dependent variable like "*crime*" or "*poverty*". > Even if clean data existed and Kolmogorov complexity was computable > Matt, this repeated mention of "computable" by someone of your stature is really quite disturbing. I repeat: No one has ever asked a scientist to *prove* that his theory is the best of all possible models relative to all data under consideration. Don't you understand why I always say the same thing in response to the "uncomputability" of Kolmogorov Complexity? Do you think it doesn't pertain to that issue? Why must I repeat myself on this? Why isn't it glaringly OBVIOUS? My take: There is an institutional block against being held to account by an objective metric. Why? No conspiracy is required. All that is required is the incentive and everything self-organizes around the goal: Promote specious pedantry like this "uncomputable" trope and hope no one notices you have no clothes! > , the output is still just a few bits. Your choice of the language > dependent constant could give you whatever answer you wanted. > The solution to both of these objections is an inevitable consequence of the prior passage about incorporating data from a variety of competing sources. The "many analysts one dataset" projects are a small step toward such transparency. With a proper scale of data (which The Laboratory of the Counties <https://github.com/jabowery/HumesGuillotine/tree/master/LaboratoryOfTheCounties> may or may not provide -- I'm in the midst of determining that -- but in any case, it is only there to point the way) the KC of THAT dataset would dwarf the Hutter Prize "constant" choice of UTM fiction (ie: x86 instruction set). We should be worried. I just saw a study showing fertility collapse > coinciding with the introduction of the iPhone around 2009. This is going > to accelerate with the introduction of artificial people. > I agree but that's only a proximate cause over which we can argue until we're turned to protoplasm by the hypersonic tsunami. When I say "we" I mean to poiont to people like Raj Chetty and his proprietary access to government data. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T5b182c3fd14f61e9-M58bab79f8b95f712e1f43b12 Delivery options: https://agi.topicbox.com/groups/agi/subscription
