Index

The Metacrisis is here

    Everything is connected (climate, food, security, biodiv)

× Self-evolving tech (inc. runaway AI)

× Coordination failure ("Moloch")

= Total risk for humanity and biosphere

Schmachtenberger: hypothesized attractors

1. Chaotic breakdown
2. Oppressive authoritarian control

Designing a third attractor

The Gaia Attractor

Goal: Resilience/stabilization of bio-socioeconomic system, subject to

• Multi-level system constraints (ex: planetary boundaries)
• Desirability (ex: preserve standards of living)
• Feasibility (ex: can be achieved from current initial conditions)

Complicators:

• Partial observability (info asymmetry)
• Scientific uncertainty
• Computability
• Control mostly via incentives & recommendations, not actuation
• ...

easy, right?

The Gaia Protocol

A decentralized hybrid AI-human system for planetary-scale decision support and automation

• Human and artificial agents interacting in an open network

• Facilitate consilience (model alignment) and coherence (goal alignment) at multiple levels/scales, up to global (survival of the system)

• Built-in incentives and governance

• Privacy-preserving

Index

Decentralized Digital Twins

Shared, verifiable, networked local models of real-world systems

Understand costs and benefits of strategies/projects

Better decisions: recommendations, negotiations, valuations...

Initial application: agroecology

• Distributed non-iid data

  • Sparse

  • Heterogeneous

  • Private

  • Potentially unreliable

• Localized, collaborative forecasting and planning

  • Personalized

  • Privacy preserving

• Recommendations, not direct control

Desiderata

A practical implementation tackling real-world challenges:

Federated active inference agents

Fangorn: engine used by each Ent to inquire, learn, plan, and allocate resources

Gaia Protocol: language used by Ents to independently interact with their environments and each other

Gaia Network: mesh of AI agents called Natural Entities (Ents) that act as proxies for real-world systems

If we do this right, the Gaia Network approximates a single composite agent (more on this later)

Fangorn Overview

Highlights

Automatic Model Selection

Match on:

• Configurations
• Available observable modalities

Hierarchical generative models

Hyperprior

\( p\left( \pmb{\phi}\right) \)

Prior

\( \prod_{n=1}^Np\left( \pmb{\theta}_n | \pmb{\phi}\right)\)

Agent 2

Agent N

Agent 1

\( \pmb{C}^1_t = \left(\pmb{c}^1_1, \ldots, \pmb{c}^1_t \right) \)

\( p\left( \pi^1 \right) \)

\( p_{\pmb{\theta}_1}\left( \pmb{s}^1_t|\pmb{c}^1_{t}, \pmb{s}^1_{t-1}, \pi  \right) \)

\( p_{\pmb{\theta}_1}\left( \pmb{o}^1_t| \pmb{s}^1_t \right) \)

Structured Variational inference

Ambrogioni, Luca, et al. "Automatic structured variational inference." International Conference on Artificial Intelligence and Statistics. PMLR, 2021.

P2P Federated inference

local data

\( \pmb{D}_n\)

local prior

\( p \left( \pmb{\theta_n|\pmb{\phi}}\right) \)

global posterior \( q\left(\pmb{\phi} \right) \)

local posterior

\( q\left(\pmb{\theta}_n \right) \)

local data

\( \pmb{D}_n\)

local prior

\( p \left( \pmb{\theta_n|\pmb{\phi}}\right) \)

global posterior \( q^{(i)}\left(\pmb{\phi} \right) \)

local posterior

\( q^{(i)}\left(\pmb{\theta}_n|\pmb{\phi} \right) \)

local data

\( \pmb{D}_n\)

local prior

\( p \left( \pmb{\theta_n|\pmb{\phi}}\right) \)

global posterior \( q\left(\pmb{\phi} \right) \)

local posterior

\( q\left(\pmb{\theta}_n \right) \)

local data

\( \pmb{D}_{n^\prime}\)

local prior

\( p \left( \pmb{\theta_{n^\prime}|\pmb{\phi}}\right) \)

global posterior \( q^{(i+1)}\left(\pmb{\phi} \right) \)

local posterior

\( q^{(i+1)}\left(\pmb{\theta}_{n^\prime}|\pmb{\phi} \right) \)

Node \( n \)

Node \(n^\prime \)

Gong, Jinu, Osvaldo Simeone, and Joonhyuk Kang. "Bayesian variational federated learning and unlearning in decentralized networks." 2021 IEEE 22nd International Workshop on Signal Processing Advances in Wireless Communications (SPAWC). IEEE, 2021.

Hyper-prior ConstrainTs

Academic studies

Expert

knowledge

Project context and data

Project assessment and  decision support

Project-level results

Global constrains on regression

Regression results

Global constrains on agents

Data Source reliability

Williams, Donald R., Stephen R. Martin, and Philippe Rast. "Putting the individual into reliability: Bayesian testing of homogeneous within-person variance in hierarchical models." Behavior research methods 54.3 (2022): 1272-1290.

\( p(y_{mj}|x) = \mathcal{N} \left(x, \sigma_{mj}^2 \right) \)

Source reliability

\( \rho_{mj} = \frac{\sigma_x^2}{\sigma_x^2 + \sigma^2_{mj}} \)

Simplified generative model

\( p(x) = \mathcal{N} \left( \mu, \sigma_x^2 \right) \)

Modality reliability

\( \rho_{m} = \frac{\sigma_x^2}{\sigma_x^2 + \frac{1}{M^2}\sum_j \sigma^2_{mj}} \)

\(m \in \{1, \ldots, M \} \) and \(j \in \{1, \ldots, K \} \)

Coming soon

• Category-theoretic model composition (Smithe, 2023)
• Category-theoretic model discovery (Sennesh et al, 2023)
• Machine-verifiable inference and planning (Kaufmann et al, unpublished)
• Automatic lit review (Kaufmann et al, unpublished)
• Integrating goals via constraints and satisficing (Simon, 1996)
• Knowledge economics (Hubbard, 2007)
• Multiscale ActInf agents via Theory of Mind and goal alignment (Kaufmann et al, 2021)
• ActInf over models of arbitrary HW+SW systems (Coy and Kaufmann, unpublished)
• Internalizing externalities with ActInf (Hesp and Kaufmann, unpublished)

Index

Case Study

AGROECOLOGICAL Modeling

• Data
  • Satellite images (remote sensing) 
  • In situ data (farms and forests)
  • Verbal attestation (farmers, auditors...)
  • Prior knowledge (meta-analysis)
• Models
  • Hemp farming
  • Agroforestry

Agroecological model

 \( \pmb{\varphi} \sim p\left( \pmb{\varphi}  \right) \)

Global parameters

• EVI properties
• Sprouting properties

\(\vdots \)

Latent states - Plant growth model

• Plant count
• Plant size
• Survival rate
• Growth rate

\(\vdots\)

Observables

• Plant count
• Plant size
• Vegetation indexes

\( \vdots \)

Local parameters

• Growth parameters
• EVI intercept
• EVI slope

\(\vdots \)

\( \pmb{y}_t^n \sim p \left(\pmb{y}_t^n | \pmb{x}_t^n , \pmb{\theta}_n  \right) \)

\( \pmb{\theta}_n \sim p\left( \pmb{\theta} | \pmb{\varphi} \right) \)

\( \pmb{x}_t^n = \pmb{f} \left( \pmb{x}_{t-1}^n, \pmb{c}_{t-1}^n, \pi_{t-1}^n, \pmb{\theta}_n \right) \)

Covariates

• Weather & physical risk

\(\vdots\)

\( \pmb{C}^n = \left( \pmb{c}_1^n, \ldots, \pmb{c}_T^n \right) \)

Policies

• Growth modulators

\(\vdots\)

\( \pmb{\pi}^n = \left( \pmb{u}_1^n, \ldots, \pmb{u}_T^n \right) \)

Meta-analysis model

 \( \pmb{\varphi} \sim p\left( \pmb{\varphi}  \right) \)

Global parameters

• EVI properties
• Sprouting properties

\(\vdots \)

Observables

• Experimental measurements
• Observational data

\( \vdots \)

Local parameters

• EVI intercept
• EVI slope
• Precisions

\(\vdots \)

\( y_k^d \sim \prod_i \rho_i \left( \pmb{c}_k, \pmb{\theta}^\prime_k \right)^{\delta_{y_k^d, i}} \)

\( \pmb{\theta}^\prime_k \sim p\left( \pmb{\theta}^\prime | \pmb{\varphi} \right) \)

Covariates

• Features
• Interventions

\(\vdots\)

\( \pmb{y}_k^c \sim \mathcal{N}\left( \pmb{y}| h\left( \pmb{c}_k, \pmb{\theta}^\prime_k \right), \pmb{\sigma}_k^2 \right) \)

Prior predictive

Multi-model Federated inference 

Academic studies

Expert

knowledge

Fangorn

Project-level

Agent Models

Entmoot

Meta-analysis Model

Global constrains on regression

Regression results

Project context and data

Multi-model Federated inference 

Academic studies

Expert

knowledge

Fangorn

Project-level

Agent Models

Entmoot

Meta-analysis Model

Project context and data

Posterior estimates of global hyper-parameters

Multi-model Federated inference 

Academic studies

Expert

knowledge

Fangorn

Project-level

Agent Models

Entmoot

Meta-analysis Model

Project context and data

Posterior estimates of source reliability

Policy InfErence

Estimate possible action sequences from available data

Project Assessment UX

Index

〞

...for policy purposes we do not need exact empirical measures of the optimal scale. If one jumps from an airplane it may be nice to have an altimeter, but what one really needs is a parachute.

- Edward Fullbrook

"But what about...?"

• Scaling to global goals?
• Internalizing externalities?

Scaling to Global Goals:
Agents made of Agents

• Collective of interacting ActInf agents performs approximate inference at the ensemble level
• Theory of Mind + goal alignment: improve CI performance, eliminate ambiguity by actively exploiting diversity

Kaufmann, Rafael, Pranav Gupta, and Jacob Taylor. "An active inference model of collective intelligence." Entropy 2021, 23(7), 830; https://doi.org/10.3390/e23070830

INTERNALIZING externalities: the Preservation Game

• Ent infers true system state and uses it to credibly incentivize long-term positive outcomes
• Works even with high uncertainty and misinformation
• Successful against a variety of bot strategies, including collusion 

Kaufmann, Rafael, Casper Hesp, 2022, unpublished

Partner Case Study

COVID-19 Policy Response

https://evolution.ml/demos/npidashboard/

Bringing it all together

• Interoperability: shared protocols to align decisions across contexts
• Reusability: libraries of diverse & tested components, standard "APIs"

Some of Our contributors (so far)!

Help us build the planetary brain

Sign up at gaiaconsortium.org and/or reach out to rafael.k@digitalgaia.earth!

Together, we can design, build, and learn what it takes to achieve planetary regeneration.