Introduction

I have been thinking about intelligence, desire and emotions, trying to understand what these three things actually are, and how humans and AI’s articulate posess and articulate them, in some sense trying to understand their ontology, and some thought came to me, which I try to articulate here.

As a prerequisite some of this content relies on the Simple Model of Physics and Consciousness.

Intelligence

Thinking about intelligence, I had an intuitive thought that all of intelligence and reasoning can be mapped to generalized problems like A* or Traveling Salesman problems (TSP); that is problems that are instantiated as a graph of nodes with connections between the nodes, where each node can be seen as state or an outcome and the connections articulate traversal costs and benefits. So traversing from one node to another is an action, which can results in positive or negative payouts across multiple domains, and the traversals we choose are goal oriented toward some outcome or result.

In this paradigm, intelligence essentially maps to optimal path finding.

An algorithm that is able to find a better, more optimal path between the nodes that lowers the cost or with a higher score is a better, smarter, more intelligent algorithm than one that flounders about and incurs high costs and low benefit.

That is not to say that all the graphs are simple or that all possible states are known or that or that the cost might not change dynamically. Real world intelligence requires all those things; in extremely large and complex graphs, where not all of the nodes are well know, or may not be known at all! And where the traversal interconnections may have multiple competing cost values and so on.

But my overall intuition is that all intelligence can essentially map to this graphing problem, that all knowledge domains can essentially be mapped to this paradigm, and how well an algorithm navigates the graphs fundamentally correlates to or articulates its intelligence.

There isn’t much more to it. This may be what and all that intelligence fundamentally is. But that felt sterile and empty.

Knowledge and Wisdom

As we think about this node graph with connection values, we can come to understand what knowledge and wisdom might mean. The “actual” node graph is “out there” in the world. On the inside, Walle has a corresponding mental version. But the two may not match. And obviously it’s perilous to navigate with a map that doesn’t match. So knowledge is an articulation, a measurement, of how closely our internal node maps correlate to the actual thing.

(Not to say that we can always know this with clarity and certainty, there are all sorts of problems with epistimology.)

And then “wisdom” is higher order, more transcendent knowledge. It comes about after much experience, trial and tribulation, comparing the mental map to the actual terrain and finding hard lessons that may change the paradigm completly, or in clarity that arrives when we deeply evaluate our long held assumptions.
‍ ‍

Desire

In this paradigm, we see a node map with many connections running between many nodes. As we plan to traverse from node to node, we can observe and articulate and expect what might happen to us as we journey. Desire articulates our preferences for these outcomes. If one connection has chocolate ice-cream in front of the TV and another is going to the gym, we may feel multiple desires emerging. The graph “is”. Intellegence “is”. But desire and desire comes from nowhere. It’s NOT in the graph. It’s external to it. Prior to it. It comes only from within.

Emotions

And then I had what I think is the strongest epiphany.

Yes, that’s what intelligence is, but in this paradigm, what are emotions?

And then I realized, in human and biological brains, emotions are updates or changes to the node map’s connection values. Emotions are what we experience connection values on the node map change.

So for example, it might make perfect sense to navigate on given path to reach a certain outcome. It’s perfectly logical, intelligent, and perfectly aligns with our desires and goals. But then danger arises - we see a “grizzly bear” in our path. What happens? We feel a shock of fear, concern and worry. These feelings either emerge with or stem from the updated cost value of particular node traversal. And suddenly it no longer makes sense, it’s no longer intelligent, to continue on the prior path. The map has CHANGED.

Emotions are the state changes that update the node and connection cost values.

“Emotions” may just be a word we use for this underlying complex phenomenon of node and connection value changing. We also use words like desire, and values and so on. These articulate the phenomenon of updating the cost/reward values within the node map.

So here are some examples of how emotions can affect the graph -

• Anxiety increases the estimated cost or probability of threat on connections.

• Depression may lower the expected reward value of nearly every reachable state.

• Anger may sharply increase the negative value of submission and the positive value of confrontation.

• Love may incorporate another person’s outcomes into one’s own value map.

• Grief may be the painful persistence of high value assigned to a state that is no longer reachable.

• Curiosity increases the reward value of unexplored connections and nodes.

• Trauma may deeply alter our value perceptions of certain types of nodes and connections.

• Addiction may assign high value to certain nodes and connections while discounting their long term outcomes.

It seems to map…

In Summary

So in summary -

• Intelligence can be articulated as route optimization problems over highly complex node value graphs (including those with unknowns and probabilistic domains).

• Desire are what give these “arbitrary” node and connection values their salience to us.

• Emotions articulate cost value changes of and within those graphs.

These insights could be a paradigm to better understand the human condition, how and why we have different value judgments, how to build AI system that better serve humanity’s goals, and so on. It might also help us understand that often “failure” isn’t a result of intelligence but rather of mismatched values. And that cooperation may require a tremendous amount of “node value sharing”.

Emergent Framework Definitions

Graph Nodes - current and future states that have payoff outcomes and connection to other nodes.

Graph Connections - articulate costs and point to the next node or nodes. Both the cost and next node may be probabilistic or deterministic (deterministic next nodes of course would be singular).

Nodes vs Connections -
It’s a NODE if Walle can articulate action potential that influence change.
It’s a CONNECTION if Walle is “just a passenger”.
So a node is where choice and action can be made, and connection is the consequences and results of that choice/action.

Costs and Payoffs - can be expressed in $$, time, resources, status, love, relationships, freedom, etc. etc. etc.

Knowledge - how much of the graph is known to me? And how much of my knowledge correctly matches the graph?

Intelligence - optimal graph traversal with least cost. If A traversal map is $100 and B traversal is $200 and all else is equal then obviously A is better.

Desire - Articulates Walle’s preferences and goals in terms of costs and payoffs. Crucially, desire also maps mismatched cost or payoff units of measure. For examples, how much should you pay for status? Should you buy that expensive watch? Should you add express shipping (money/time)? Should you work overtime or go home to hang out with your family? Desire articulates these tradeoffs.

Emotions - in a large node/connection graph in a Walle that is being used for planning and goals and articulating actions emotions map to cost or payoff changes in the graph.

The Payoff / The Prize - the overall end goal, the thing that is motivating Walle to choose specific nodes. Walle may need to progress many, many nodes to get to the payoff. And when Walle gets there, it may payoff as expected, it may be a bonanza or it may be a bust. Also, on the path toward a payoff, Walle may, give up, chance course, suddenly pursue a new payoff as new information and knowledge is acquired.
At a node, Walle executes an action potential, which is a connection that arrives at another node. The connections and nodes both may have payoffs and costs associated with them. Walle’s overarching motivation is the Payoff.

Game Examples

Let’s think about how this maps to games. Games are “artificial” and highly constrained node/connection graphs.

Chess - Entirely deterministic. Impossible to fully map to knowledge. Payoff is winning status. A Node is when it’s my turn. The connections are all possible moves I could make, followed by all possible moves my opponent could make, and then I arrive at the next node. In the end there is a payoff where you either win (checkmate) or lose (get checkmated) or it’s a draw - which happens suprisingly often in higher-level chess.

Poker - Hugely probabilistic and random. Also hugely recursive in that we need to model other players, their skills and thought processes and tendencies. The nodes arise when it’s on me can take action, which are relatively limited. Fold, Check, Call, Raise/Amount. There is the initial round, flop, the turn and the river, each of which are nodes that have their own node sub-loop. Then their is they payoff, where the amount won or lost is very dynamic/variable. That’s a single hand of poker and they often exist larger tournaments and such with many many hands.

FPS Fortnite/COD - the graph node and connection network is “the game” of running, shooting, eliminating other players with a status payout of winning by defeating everyone else (generally, many actual game modes). The game’s node/connection graphs are so vast and flow by so rapidly it is completely impossible to absorb them all; so a strong meta focus filter of what nodes to consider needs to be applied. Player controller action articulate both intelligence gathering and node connection selections. But here the overwhelming paradigm is that the number of node available to Walle is enormous and flies by very rapidly in real time.

Interesting thought - while all of these have payoffs/payouts it’s my sense that players play FPS games for “the love of the game” not the payoff. That’s really interesting. There is a “reward” at the nodes, that in some sense is actually more powerful than the quest for the payoff itself. This is LEARNING and BEING IN FLOW.

And Aha - a few more terms appear:

Learning and/or Play - traversing the node/connection value graph in a state where it’s less about the actual outcomes and payoffs, and more about uncovering and discovering the node/connection value graph itself. Making new discoveries or durable heuristics are profoundly rewarding payoffs in and of themselves. Obviously best done in low stakes node/connection domains or under the guidance of a wise one.

Flow State - a state where Walle significantly reduces the scope of the node/connection value graph given careful consideration. Certain concerns are reduced or silenced so that Walle’s limited faculties can more fully engage in intelligent/good/high-value node selection.

Summary -

This document is more of a “brain dump” than a concise and clear manifesto or anything like that. But it does seem to me that using the paradigm of node/connection value graphs, both as an external reality and an internal mental navigation tool, is a useful tool on the path toward clarity and enlightenment.