FOLI RFP: How to mitigate AI-driven power concentration

General Discussion
1

How to mitigate AI-driven power concentration

We’re offering up to $4M to support projects that work to mitigate the dangers of AI-driven power concentration and move towards a better world of meaningful human agency.

It’s not like I haven’t been trying to head this problem off for over 3 decades now

(if not longer)

And it isn’t like I haven’t approached related Institutes (Lifeboat, Metaculus, Hansen, etc. as well as dissident right influencers like Murray, Derbyshire and Sailer – all of whom should have been eager) with the relevance of the prize awards for Algorithmic Information Criterion for macrosocial dynamics model selection (like during the pandemic).

My cynicism reaches inexhaustable extremes in this mismatch between “influence” and competence – to the point that I seriously want to tear the whole edifice down some point in the 1800s if not 1700s and start over.

But, ok, let’s say that I can set aside my attitude and entertain the possibility that Future of Life Institute is, like many others, starting to realize that the existing structures have screwed the pooch – not the least of which is the Jewish elite that has taken over from the old WASP elite without really understanding what it took to build that society in the first place and now finding themselves caught between calling those they want to take over “Nazis” and being called “Nazis” themselves. There is some evidence (ie: Trump) that there is an increasing “Come to Jesus” moment among the elites. And I did predict this clear back in the early 1990s when my proposal to replace the 16th Amendment with a net asset tax that funds the replacement of the welfare state with a citizen’s dividend was met with intense Jewish hostility precisely because it would have accomplished what the Future of Life Institute wants in the present RFP.

But it is clearly too much to ask people to take more than one change in policy in their thinking at a time, let alone 3 (Repeal 16th, Institute net asset tax, replace welfare state with citizen’s dividend) in order to get to a new stable state.

sigh

OK, so here’s what GPT4 did with my Hume’s Guillotine writeup to reformat it into a response to the FLI’s RFP. Whaddya think? I’ve got other things to do with my life than beg influencers that don’t want to be influced because they haven’t yet reached their “Come to Jesus” moment:

Proposal for Prize Awards in Macrosocial Dynamical Modeling

HumesGuillotine

Cynic: A blackguard whose faulty vision sees things as they are, not as they ought to be. — Ambrose Bierce’s Devil’s Dictionary

Introduction

This proposal outlines a series of competitions aimed at fostering rigorous ethics in Artificial General Intelligence (AGI) based on Hume’s Guillotine, which separates what is from what ought to be. Our focus is on macrosocial dynamical modeling using the Algorithmic Information Criterion (AIC) as a fundamental tool for understanding and mitigating AI-driven power concentration. The foundation of this approach is that to achieve any objective, including mitigating power concentration, a robust model of macrosocial dynamics is essential.

Background

Artificial General Intelligence (AGI) unifies is with ought. In Marcus Hutter’s AGI theory, AIXI, Algorithmic Information Theory provides the is through scientific knowledge, while Sequential Decision Theory provides the ought by applying this knowledge to decision-making. The Algorithmic Information Criterion (AIC) stands out among information criteria for model selection as it focuses on lossless compression of data, thereby avoiding confirmation bias and ensuring the integrity of scientific discovery.

Objectives

Our proposal aligns with the Future of Life Institute’s (FLI) request for proposals aimed at mitigating the risks of AI-driven power concentration. We emphasize that achieving these objectives necessitates a sound understanding of macrosocial dynamics through rigorous modeling. Our approach involves competitions that drive the development and validation of these models, ensuring they are robust, unbiased, and capable of informing effective interventions.

Methodology

Competitions

The proposal includes a series of competitions centered on:

  1. Data Collection: Curating comprehensive datasets that encompass diverse aspects of social dynamics.
  2. Model Development: Utilizing the AIC to develop models that accurately represent macrosocial dynamics without omitting critical data, including what might traditionally be considered noise or measurement errors.
  3. Model Evaluation: Employing objective scoring based on the AIC to evaluate the models’ predictive power and alignment with observed data.

Impact on AI-Driven Power Concentration

Effective mitigation of AI-driven power concentration requires an in-depth understanding of the underlying social structures and dynamics. By leveraging the AIC, our approach ensures that models are not only scientifically robust but also capable of identifying and predicting biases and power imbalances within social systems. This knowledge is crucial for designing policies and interventions that promote equitable distribution of AI benefits and prevent the concentration of power.

Alignment with FLI’s Vision

Our proposal directly addresses the topical focus areas outlined by FLI, including:

  • Public AI Development: Proposing models and frameworks for AI that operate outside traditional corporate structures, ensuring greater public control and accountability.
  • Safe Decentralization: Developing governance models for AI that prevent the proliferation of high-risk systems while promoting decentralized control.
  • Effectiveness of Open Source: Evaluating how open-source AI initiatives can mitigate power concentration and proposing strategies for their effective implementation.
  • Income Redistribution and Incentive Design: Exploring mechanisms to ensure equitable distribution of AI benefits and designing incentive structures that prioritize social good over profit maximization.

Expected Outcomes

  • Enhanced Understanding of Social Dynamics: Providing robust models that elucidate the complexities of macrosocial interactions and power structures.
  • Informed Policy Development: Offering insights that guide the creation of policies aimed at preventing AI-driven power concentration and promoting equitable distribution of AI benefits.
  • Increased Public Engagement: Encouraging public involvement in AI development and governance through transparent and accountable modeling practices.

Conclusion

This proposal underscores the necessity of robust macrosocial dynamical modeling for any initiative aimed at mitigating AI-driven power concentration. By leveraging the Algorithmic Information Criterion, we ensure that our models are scientifically rigorous, unbiased, and capable of informing effective policies and interventions. We invite FLI to support this initiative, driving forward the development of equitable and sustainable AI systems.

4 Likes
2

Maybe this makes me a “blackguard”, but it is not clear that the avenue (the difficult & treacherous jungle trail?) back to a sustainable society can be based on computational models.

Right & wrong – no computation needed there. Good & bad? Effective & ineffective? The computational modeling approach has given us such disasters as the Anthropogenic Global Warming Scam, the Covid Scam. It is time to get back to basics! Fortunately, the Coming Collapse will end reliable electric supply in the West, which will leave us with no alternative than to go back to basics in analyzing our problems & opportunities.

3 Likes
3

Hence my reference to the 1800s if not 1700s, but let’s for just a moment reflect on another notion of “back to basics”:

Occam’s Razor applied to “computational models”.

This is not yet been tried as an alternative to killing off billions of people.

I’d like to at least give those billions that chance… however small it might be.

3 Likes
4

Very cool to see FOLI support this.

2 Likes
5

Taking time out from debunking Einstein, I have been working and reworking this introduction since the deadline is tomorrow:

We propose to provide a critical component of any effort to mitigate centralization of power: A Foundation World Model (FWM). Unlike prior efforts such as World 3, FWM will be the result of offering sociology a financial incentive, treating it as a learning system that incorporates humans as well as machines, subject to the most principled loss function: The Algorithmic Information Criterion (AIC, not to be confused with the less principled Akaike Information Criterion) for model selection. The AIC is the most principled loss function due to its foundation in Algorithmic Information Theory’s proof as the gold standard information criterion for inductive inference. Large language models have demonstrated foundation models are superior to specialized models in specialized task benchmarks. Indeed, the phrase “world model” has come to characterize the reason foundation models are superior in specialized tasks: They account for more variables. By contrast, the social sciences focused on specialized models, i.e. the prediction of so-called “dependent variables”. In the absence of a coherent incentive, the social science potential of Moore’s Law in the age of Big Data produced profound dysfunction in sociology, such as “p-hacking”. Attempts to address this dysfunction with replication studies involving multiple analysts being presented with the same data to compare analysis methods has foundered on the lack of a principled information criterion for dynamical model selection. That the AIC is ideally suited for the age of Moore’s Law and Big Data in such multi-analyst replication studies has entirely escaped the social sciences. While this remained in mere potential civilization has careened blindly toward a potential global catastrophe, only one variable of which is centralization of power. The FWM will open our eyes to the road ahead.

Comment on Illya’s mischaracterization of fine tuning:

An ideal Kolmogorov complexity model of the entire corpus would if prompted with a minimum context question respond as truthfully as it could without any additional fine tuning. So the fine tuning is really a ham-handed adaptation to the failure to achieve that ideal. Nevertheless what he says about achieving a world model is good as far as it goes.

2 Likes
6

Probably not going to make it. They demand an EIN from a non-profit and although I have a religious corporation in the state of Iowa, I don’t have it registered with the IRS and the online registration for an EIN with the IRS decided not to vend one. So now I’m on hold for like an hour with a status code “101” whatever that means to those of us who haven’t read Orwell.

2 Likes
7

It wouldn’t have mattered if I had an EIN as the requirements for a “fiscal sponsor” are above and beyond tax exempt status with the IRS. Here’s the resource I was referred to:

https://fiscalsponsordirectory.org/

To get a feel for where all the grants are going, here’s the list of their “service categories”:

Search by service category Animals • Arts & Culture • Children, Youth & Families • Disaster Relief • Drug Treatment • Economic Development • Education • Environment/Sustainable Growth • Faith Based/Religious • Festivals & Events • Food Insecurity • Health/Nutrition • Homelessness • Housing • International Development • LGBTQ • Mental Health • Open Space/Parks • People of Color/Minorities • Political Advocacy/Social Justice • Social Services • Transportation • Veterans • Women • Youth Development

I smell Melinda French.

1 Like
8

And, of course, wouldn’t you just know it…

“math”
heh…

2 Likes