Dan Greenberg - Backwards Thinking

Backwards Thinking Posted 2025-March

(This is a speculative essay. Read with caution as all of its thoughts are pure conjecture. I would appreciate any comments.)

Commonly, Reasoning is viewed as the "top" or "highest" function in intelligence. I certainly think about it that way. Since conscious thought is what we're aware of, it's certainly reasonable, albeit self-centered, to view it that way. But maybe that's not the right way to think about it. What if - at least initially - Reasoning developed as yet another tool of System 1 and only later became superior. Or did it ever?

Recently, I was having a discussion about AI architecture. Recently, there's been a move from single-model systems to systems of models, orchestrated by one. You will see things like "the [main] model first writes a task list to get the overall task done and then calls task-specialist models or other non-AI tools to get them done." Task-specific models might - say - write a bit of Python code or format data and then call an API. It's a good solution to handling complex assignments.

I started thinking about this in terms of ourselves and our brains. Normally, I think of this as sensors feeding "up" to a decision layer and then an action feeding "down", eventually to muscles. Instead, I applied a PC-like architecture. In a PC, there are various devices that can be driven and that can interrupt the higher functions when more information is available or needed. And there is an operating system that coordinates all of this activity and often adds to the base functionality. (Directory format on a disk drive comes to mind as an example of this.) Even without applications, a PC does quite a lot to keep itself functioning, respond to events, and more. In that sense, System 1 is our operating system, with the cerebellum committed to running it. And note that it calls specialized tools... like the premotor cortex or the nasal ganglia to execute commands or pre-process data. It's not a perfect analogy, but go with it for a moment.

What about applications? After all, that's what we see mostly and that's the purpose of having the PC. In this analogy, they are System 2, and they sit on top of the OS just as Reasoning sits on top of my model. But, applications for the most part came after the OS, just as Reasoning and the Frontal Cortex evolved later than the Cerebellum. There is a key difference, however: PC applications lead to benefit to the users while Reasoning leads to better survival for the individual.

Then, generalize. Suppose System 1's job is simple: keep things running. Survive. It gets a variety of inputs and takes actions accordingly, both from/through specialized units. It could evolve quite sophisticated behaviors to do so - including sequential task lists - and, in fact, we see a broad spectrum in mammals. But such a System 1 is reactive and tactical. What it lacks is a tool to be proactive and strategic.

And that would drive the evolution of System 2. But note that System 2 here is yet another tool that "the OS" is calling and coordinating. It's not a new boss brought in "on top". At least initially, in an evolutionary sense, it's simply a means for integrating behaviors over time and recommending strategies for higher survivability. As such, it's not surprising that it's "thinking slow."

From such a start, there's a through-line to where we are. For instance, self-awareness and consciousness certainly fit with the idea of a strategic aid to survivability. And the self-aware "self" feels that it's clearly the highest level of the being. But the question remains: is System 2 the boss or merely one tool of the OS? Does it matter?

Addenda (Because I am still thinking about this and there's no place better to put those thoughts)

Thinking some more about evolution... suppose we start with a creature that has just a cerebellum... just an OS. A lot of tactical behaviors - finding food, eating, mating, excreting (in a way that's harder for predators to track), fleeing danger - are prewired because they are elemental to the animal's survival. Notice, though, I said finding food and not hunting for food. This will be important in a moment.

So, this animal is pretty survivable, but it's all "in the moment" - all tactical. "I am hungry" triggers "find food" which triggers "eat." There's no long-term memory and therefore no long-term planning or strategy. The animal just is.

But some of the animals begin to have a bit more. Perhaps they start to remember good places to find food. "I am hungry" then leads to "where are some good places to find food" which leads to a more efficient "find food." Clearly this is advantageous, and long-term memory is relatively simple. In fact, there's been a recent finding that memories of events (like finding food) are tightly coupled to memories of places.

The need to find food and the ability to remember where/how to get it leads naturally to hunting for food. That is, hunting is a preplanned search for food. The proto-System 2 begins with simple planning/sequencing to achieve better hunting results. (I wonder how much of this planning is "visualization" versus a simple step-list.) Then, more adaptability evolves - scenario analysis and even rudimentary reasoning. Evidence of this exists in human brain structures.

Tangent (at this point): what is the purpose/advantage of vocalizations at this point? Do solitary animals vocalize much? Need to think about this, because...

And then we get to a point where pack behavior comes into play for more efficient hunting. Pack behavior brings with it two effects. Directly, pack behavior requires communication. This may or may not be verbal, but verbal communication can be used without line of sight, and so it is powerful. This further develops System 2, as elements/abstracts need to be constructed, communicated, and understood. Note also, that it drives awareness of others in relation to the self. That's a higher level of self-awareness.

The other effect is that predators in general, and pack predators in particular, drive prey species into pack behaviors as a defense. They require much less sophistication in tactics and communications, but the pressure is still there to develop these skills.

The higher levels of self-awareness tend to generalize: vocalization to woo a mate is a small step from vocalization to plan a hunt, and so on. Then there is the leap to completely abstract reasoning and thought. I wonder if that was driven by evolution, or a side effect of all of the above...

Google Sites

Report abuse