ML and Deep Learning to automatically create AGI?

[yogaman]

I absolutely agree that subcortical structures are critical to motivation and behavior. I’m encouraged that you agree that every drive must have termination signal.

I agree that there can be different satisfaction triggers for lizard-level drives. I believe at some level of development there is a recognition of how the pleasurable sensation that a full belly creates is at least a little like other pleasurable sensations, for example, safe sleeping quarters. When the organism is “idle”, it can pursue one of the behaviors on its pleasurable list.

I suggest that in humans there is a more synthetic signal, one that gives rise to the “Aha!” sensation in mental problem solving. This signal promoted the behavior we call curiosity.

I didn’t discuss, but I believe the emergence of the Aha! signal is related to the need to process “mental” objects that emerge in social relationships, the most complex of which seem to be human. It’s not surprising (to me) that only humans might evolve the signal so highly.

My proposition only requires that signal and enough “down-time” from primal drives for the emergence of curiosity.

Curiosity expands the goal map greatly, but the goal map was populated before there was curiosity; before there were brains, organisms had goals.

Edit:
I see you’ve edited again. Could you maybe please start a new post for new information while we’re typing back and forth?

I also edited for typos.

[yogaman]

Oh - and that “ah ha” thing? Making the connection between drive and solution.

Yes. Evolve that, and you get curiosity which is pretty close to GI, right?

[Bitking]

Sorry about the multi-edit thing. When I am out walking it is difficult to check resources and I enter data that is all related in sprints. I happen to be out walking the trails at night at this moment.

Also - when I am driving around doing chores I have to set the phone down ( no distracted driving) and I close the edit window then.

Also - I spell check after posting and often see I missed a point I had intended to cover so it get added. Like this line.

[Bitking]

As far as curiosity - lizards explore and establish territory.

This drive may be highly developed in humans but it is still an old-brain thing.

Any critter that has maps to goals will act to populate those maps with basic need goals like food, water, and shelter.

We humans are such chauvinists and tend to ignore that our most prized abilities are found in some form or other throughout the animal kingdom.

[yogaman]

I usually agree very strongly that human chauvinism blinds us to the capabilities of animals with less cortex, but here you seem to be suggesting that lizard’s explorative behavior is equivalent to GI? I don’t agree.

Agreed, explorative behavior arises in lizards and in people when there are no more pressing drives. And lizards can apply skills from one activity to another in some limited ways. But only humans have generalized and developed curiosity to the extent that we have, at least for tool-making.

Something is responsible for the difference in scope that makes human-scale curiosity roughly equivalent to what we’re calling GI. If it’s not evolutionary refinement of the “Aha!” signal, what do you propose is responsible for the vast difference?

Edit: I hit send too soon. Sorry. I’d delete this post if I could. I’m hoping I can sneak in some edits before you see this. Sigh. Violating my own request in the first post after I requested it. Very sorry for the hypocrisy.

[Bitking]

No harm, no foul; I look forward to seeing your edit - when you do delete the edit thing and I will delete this bit. As for this post - I’m done now.

I am in no way suggesting that the drive to explore and the resulting curiosity is equivalent in lizards and humans. There is a continuum and this is clearly more developed in humans. Some of this is cultural and not in the hardware. I offer the example of speech and all it bring to the human mental machinery. Without speech and the mental tricks it brings you lack the ability to form and manipulate certain mental constructs.

I do see AGI as far more than curiosity; I see that AGI will be a cluster of traits that must be taken together to seem intelligent.

• Intentionallity.
• build in behaviors to act as a framework for development.
• Memory. (it seems that several kinds will be needed)
• Task switching (with a minimum set of goals)
• Self awareness.
• Emotions of some sort (necessary for judgement)
• I’m sure if you think about it you will add a few more things to this list.

I have been posting on various aspect of this for a long time on these topics in this forum. I posted pointers to about half of them in this thread.

You may also want to read the thread " Two Types of Hierarchies". There is some discussion relevant to this topic there.

As far as the “AH ha” experience, see Global Neuronal Workspace.
To see this in it’s proper place, see this post.

[bkaz]

Something is responsible for the difference in scope that makes human-scale curiosity roughly equivalent to what we’re calling GI

I think that something is hierarchically recursive pattern representation, enabled by neocortex.
Feedback of these hierarchical patterns is what guides exploration, physical or introspective.

[yogaman]

@Bitking: Thanks for your encouragement to post the salient points from our recent emails. I think these are the points I made.

Solving the problem/query that curiosity generates produces an evolutionarily advantageous, behavior-reinforcing response. When curiosity’s reinforcement response is made more powerful curiosity can motivate behavior in a way that emotion does.

It’s possible to see this Aha!-response addiction driving some aspects of brain structure. The neural infrastructure that supports the evolutionary development of curiosity must support vast imagination capabilities as the range of consideration grows, optimally engaging as much of the neural population as allowed by physical constraints. Thus layer II/III’s long-range connections. But I do not make that claim at this time.

GWT seems fairly successful at modeling consciousness’s apparent serial nature, but people can hold conversations while folding laundry, so behavior is not entirely serial. I do need to study more about GWT.

Last, note that individual corvids, elephants, and top-of-food-chain predators, which are other notably curious animal groups, also have significant “free” time during which evolution could have sharpened the Aha! response.

[Bitking]

@yogaman

Solving the problem/query that curiosity generates produces an evolutionary advantageous, behavior-reinforcing response. When curiosity’s reinforcement response is made more powerful curiosity can motivate behavior in a way that emotion does.

I see curiosity as a drive to populate maps. In “lower” animals that is simple spatial maps of the environment (food, water, shelter, cached food) , predictor/prey facts, communications calls, and perhaps some social environment. (birds & pecking order?) These are learned extension of the built-in genetic programming.

As the number of interconnected maps increase you have the ability to form higher dimensional internal maps where the same type of objects have more abstract extents. These maps have the unique feature that you can explore them without having to move your body. These maps also need to be populated, hence the thing we call play and curiosity - the need to populate features about goal objects. As our representations become more abstract the definition of a goal object expands.

The Interpretation of GWT is a general mechanism. I don’t see that it has to engage the entire cortex with every activation. There are in excess of 100 maps, with two almost identical halves of the brain available. I could see local pockets of a few maps coalescing on some cluster of features & goal, with a different pocket elsewhere processing some other cluster of features & goal.

I think that the AH-HA feeling happens when a large number of maps fall into enlightenment with some (internally) important goal. These features could well be feature and relation aspects distributed through higher dimensional internal representations.

In case I did not make the point clearly - play and curiosity are drives like food, water, shelter, and reproduction.

[yogaman]

play and curiosity are drives like food, water, shelter, and reproduction.

I agree that play fills maps, but I see play as a behavior motivated by curiosity.

Curiosity seems in a different class from primal drives such as food/water/shelter (shelter includes fleeing threats, right?). If those drives are not satisfied, the organism dies, yielding no (more) progeny. If reproduction is not satisfied before the end of an individual’s life, likewise, no progeny. If curiosity is not satisfied, the probability of direct impact to the individual’s reproductive success is minimal.

Consider also: In the simplest animals, enough information about how to select among possible behaviors seems to be available from birth. One might say their “maps” are pre-filled. (Maybe someone should carefully observe hydra for signs of play to test this idea.)

Curiosity as a motivation for play requires unfilled neural maps (assuming that neural maps are the repository for drives in more cortical animals). Simpler brains apparently have fewer maps to fill than human brains. It seems logical that the curiosity “drive” would be correspondingly weaker in those animals, unlike primal food/water/shelter/reproduction which would seem to be just as important to simple and complex organisms.

So it seems to me that the unique characteristics of curiosity as a motivator for behavior require an explanation for those differences. This has motivated my proposal for its amplification via the Aha! signal in free-time species.

Your suggestion to allow for >100 maps seems to me to be an excellent enhancement to GWT, again given that I admit need for further study of GWT. If you have also considered how those maps interact, please share your thoughts.

I’m not sure you’ve significantly clarified the Aha!/AH-HA signal by saying it “happens when a large number of maps fall into enlightenment (perhaps you meant alignment here?) with some (internally) important goal”, more than just saying it signals the solving of the problem/query. What am I missing?

(I’m also not sure the antecedent for “These” in “These features could well be feature and…” or whether it’s more than tangential to your multi-map proposal.)

Finally, I’m making assumptions about what you mean by maps. Would you kindly point to or summarize a decent definition?

[Bitking]

Cortical maps 101:

Cortical map

Region is a different name for a map:

List of regions in the human brain

And more about how they are connected:

Grids into Maps!

This is a companion to the post HTM Columns into Grids Sure Grids are cool - News Flash- so are Maps! The reinforcement within and between maps goes a long way towards a general solution of the binding problem. One of the “breakthroughs” for me is that the cortical.io people have formed the SOM all in a single batch. While very powerful this is not biologically plausible; the brain learns online as data is presented to it. I am thinking that with an attractor model that is formed as the co…

[Bitking]

The “standard” model of the cortex has a bidirectional flow of connections, one from the sensory areas towards the temporal lobe and frontal lobes, and the other roughly from the frontal lobes to the central sulcus with many projections back to the sensory areas. These projections from the frontal lobe are associated with goal states, and the sensory areas, perception of the environment.

[gmirey]

Reading you with some interest.

Curiosity seems in a different class from primal drives such as food/water/shelter (shelter includes fleeing threats, right?). If those drives are not satisfied, the organism dies, yielding no (more) progeny. If reproduction is not satisfied before the end of an individual’s life, likewise, no progeny. If curiosity is not satisfied, the probability of direct impact to the individual’s reproductive success is minimal.

From what @Bitking posted here

This paper seems related to the current question of what initial functions you would give your evolving AGI to start the evolutionary process:
https://link.springer.com/content/pdf/10.3758%2Fs13414-019-01760-1.pdf

You may get another answer. Here expressed as a theory of a very basal eumetazoan

Here it becomes useful to make a distinction between two hierarchical levels within the nutrient control system (Fig.3B), which probably emerged more than 750 Mya. The ANS possessed chemical receptors sensitive to the nutrient content of the external world, which secreted hormones when that content was high. This defined a new lower-level impetus, “hunger + presence of food”, which motivated patterns of oscillatory activity at the blastopore that resulted in water intake.
[…]
If the food was depleted before the hunger impetus was eliminated, the animal would now be in a state of “hunger + no food”, which would motivate oscillatory patterns at the blastopore that resulted in propulsion and, thus, long-range locomotion. This is analogous to “long-range exploration” and tended to have the effect of bringing the animal to sites of higher nutrient concentration
[…]
Although the scenario described above is hypothetical,widespread evidence supports many of its features, including the signals involved in the internal causal pathways. In early eumetazoans, the high-level state was governed by the ANS,which in vertebrates became the hypothalamus, whereas specific behavioral policies (such as locomotion) were controlled by the BNS, which became the rest of the nervous system.Even in modern animals, much of the high-level control is still regulated by the hypothalamus. The general state of satiation is signaled by serotonin
[…]
The shift between local exploitation and long-range exploration is governed by the neurotransmitter dopamine, whose secretion is related to food intake and which stimulates behaviors leading to area-restricted search in very diverse species, including nematodes, flatworms, mollusks, and vertebrates (Barron, Sovik, &Cornish, 2010; Hills, 2006). Indeed, as was proposed by Hills et al. (2015), the original role of dopamine may have been the arbitration of motor behavior between local exploitation and long-range exploration, and this was only later elaborated into the neurotransmitter’s many other roles in the vertebrate nervous system, such as the signaling of positive reinforcement (Schultz, 2004) and the modulation of novelty seeking

[Bitking]

If curiosity is not satisfied, the probability of direct impact to the individual’s reproductive success is minimal.

Success in exploration and mapping the location of food and water has a huge impact on the success in survival and would be a highly conserved trait.

Likewise - development of survival skills using play while still under the protection of a parent must offer a substantial survival advantage.

I could see this as the root of the extra-genomic transmission of information - culture.

[yogaman]

Success in exploration and mapping the location of food and water has a huge impact on the success in survival and would be a highly conserved trait.

I certainly agree that food and water are required for reproductive success. But your comment overlooks the question: Is that exploratory behavior motivated by a primal food/water drive or by no-pending-primal-drives, human-level, drive-like curiosity?

An organism that will die without immediately locating food or water would not seem to be curious; it seems to be hungry or thirsty, i.e., motivated by primal drives that must have existed long before curiosity emerged. I don’t see the example as particularly relevant to GI, unlike how I see human-level curiosity with its evolved Aha! signal.

We agree that play behavior fills maps. I agree that information discovered during play behavior builds culture, although I see that some cultural “truths” arise from survival-based drives as well. Have you adopted my view that play is a behavior motivated by curiosity, that is, only when primal drive urgency is minimal?

<Unfortunately, I may not be able to dedicate as much time as I would like to this discussion during the next couple weeks due to prior commitments. Hope I can, but life.>

[Bitking]

An organism that will die without immediately locating food or water would not seem to be curious; it seems to be hungry or thirsty, i.e., motivated by primal drives that must have existed long before curiosity emerged. I don’t see the example as particularly relevant to GI, unlike how I see human-level curiosity with its evolved Aha! signal.

This seems like a chicken or egg proposition.

If the critter is exploring when it is not hungry or thirsty and maps out what is in its territory is that curiosity or some aspect of hunger? Is that exploration without any other currently active drive pure curiosity?

When you search through your memory of location and goals and they match up to your current drive (hunger/thirst/shelter/mate) so you experience some reinforcing signal (AH-HA!) is that internal search part of those drives or some sort of separate general mechanism that services all drives?

I see that you have identified curiosity as a key feature of AGI.

Speech evolved from the intersection of signalling calls and object representation. Once you have shared naming and cultural reinforcement the rest of speech mechanisms fall into place. I bring this up because many of the traits that we would like to attribute to an HGI are in fact properties of the motor programming of speech recognition and subsequent mimicking/production (loaded programs?) and are not actually what might be considered pure hardware.

I place our use of the basic exploration to populate mental maps in the same general category. Yes, in humans it is a very developed trait, but it is an elaboration of a very basic drive - not something that is novel to human level intelligence.

[cezar_t]

An organism that will die without immediately locating food or water would not seem to be curious; it seems to be hungry or thirsty, i.e., motivated by primal drives that must have existed long before curiosity emerged. I don’t see the example as particularly relevant to GI, unlike how I see human-level curiosity with its evolved Aha! signal.

Hi, I just need to chime in on the curiosity topic. The way I see things, curiosity is not a drive but a mechanism in service of a drive. A hungry lizard is curious about how to find a grasshopper or any other food satisfying bug, once no more hungry its curiosity shifts towards finding a mate or a shiny spot on a rock to roast in the heat of the sun. Whatever “needs” are in line, curiosity serves them most important first.
Curiosity is the “make wish happening” mechanism of the brain.
And of course the lizard has an “Aha!” moment when it is certain that whatever shady feature in environment its senses spotted as “potential grasshopper” is confirmed to be a grasshopper through further examination: “Aha! it tastes like a grasshopper”.

PS. and of course it

1. builds learning experience once the “shady feature” was confirmed to be grasshopper it increases grasshopper-probability of that feature and
2. once a behavior path is learned, curiosity is no longer needed, lizard reacts from learned experience, just grabs the “feature that was proved to be grasshopper”, without bothering to pay much attention.

[yogaman]

I see we’re struggling with eggs and chickens and definitions, and I’ve done my part to make it a muddle. I apologize.

I would call the mechanism in service of primal drives “exploratory behavior”. Presumably there is a short list of possible EBs in primitive animals. Once an organism has developed enough unfilled neural map space, the possibility of having behavior be non-primally motivated, I would say it has the possibility of higher-level curiosity, assuming it can find moments without primal drive urgency.

No question, as @bkaz highlighted, that there are termination signals for exploratory behavior motivated by hunger/thirst in species lacking cortex. I would distinguish the drives that motivate those behaviors from human-level (or its near antecedents) curiosity.

The way I’d like to see human-level curiosity is as the not-quite-drive-level motivation for what to do when no primal drive is active, i.e., it leads to play, a behavior that fills in the available map space, behavior that terminates when the Map-Space-Is-Populated Aha! signal arises, the MSIP Aha! being qualitatively different from the Aha! signal for a primal drive.

I wish for a better, shared vocabulary. Can someone suggest some clarifying terms?

[cezar_t]

I disagree even Archimedes “Evrika” moment (which is old Greek for “Aha”) would not have happen through aimless playing in the bathtub, his play was motivated by a focused drive, a wish to understand what makes some things float and others sink.

[yogaman]

I didn’t mean to suggest that human-level curiosity couldn’t motivate focused investigation. In fact, that’s kind of my point: simpler editions of curiosity-as-motivation drive play behavior to fill in maps. Significantly more evolved, human-level curiosity motivates qualitatively different behavior, given the vast unfilled map space that “demands” filling.

Edit: typo.