One of Levin's main points is to describe agency/intelligence as a continuous spectrum rather than an on/off thing so within the framework of thought that this paper exists in, it's no longer meaningful to treat the answer to the 'is it intelligence?' question as having a boolean answer.

I completely agree with this myself (and have for a long time before I even read any of Levin's frankly amazing work) and I think of the answer to this as more like a float/real-numbered thing - the amount of consciousness/intelligence/agency as a fraction of overall energy usage or something maybe? And that probably will lead to one constantly having to try to work out where the heck zero and one are all the time eh? heheh : )

I think it's fun and fascinating as well though for sure, and I think that even stuff as simple as a reaction-diffusion simulation, can actually contain some tiny elements of agency (just like this paper does with it's self-sorting cells!) Who cares what the scale is, right?, it's the same phenomena at the tiniest scales in my opinion, that led to life, that led to humans.

Whatever the purpose of this "sliding scale" is, it has little to do with the property of intelligence we are concerned with generally.

When we say one species is more intelligent than another, one breed of dog, or one person -- we aren't describing a difference in their organisational structure. And when we want to build intelligent systems we aim to build things that have specific capacities, not that have this sort of abstract organisation which is entirely orthogonal to these capacities.

One dog is more intelligent than another if it can read the intentions of its owner (theory of mind), coordinate in its environment (eg., open doors, etc.), plan more extended actions, pretend/fake actions to confuse the owner/other-dogs, and so on.

These ranges of capacities do not follow from an abstract 'organisational' description of the dog. The great pseudoscience of this 'computer science' thinking is that abstracts to a degree of description that is almost universal, then claims to make fine-grained distinctions.

That the earth-and-moon are 2, and the tree-and-bird are 2, does not mean the earth-and-sun and the tree-and-bird are sharing in any capacities at all. To instantiate an abstract description implies almost nothing.

Yes I agree there's little point in instantiating arbitrary abstract structures, however that isn't what I was suggesting. The related structures and properties identified in Levin's work are specifically those of the 'agential' type (going from molecular-networks/cellular scale right up to and past human-level): Goal-following meta-rules is what would be in the 'class', not just measurements or information attached/correlated to something because some numbers happen to be equal, but a type of memory of intrinsic meta-self-serving micro-behaviors that actually do inform the micro-elements how to behave.

I think that self-organization creates the possibility for the natural nucleation of agential-behavior (akin to crystal-formation) and when the system is also replicative overall, it might cause itself to happen again too! (and down the rabbit-hole we go!)

And I wonder whether it's really ok to just claim there is no micro-organizational structure that gives rise to this meta-goal-following, I mean, this is what Levin's work is all about! My own consciousness/'intelligence' is brought about by many smaller agents (my cells), and I think there's really no sharp categorical barrier here - goal-seeking (and hitting, for the winners of evolution) are pretty clear strategies from the molecular-scale to the blue-whale-sized (and humans too I reckon!)

If you mean to say that materials which have the property of self-replication with variation are essential to intelligence, then I agree with you.

One finds, in reality, that crystals do not meet this requirement -- the only known chemistry to provide it is a highly specific subtype of carbon chemistry called biology.

Yeah I think we are mostly on the same page for sure, I would nearly go so far as to say that self-replicating with variation is 'essential' for intelligence.. and I do utterly love evolution as an idea - but maybe it's just what happened-to-produce the main examples of intelligence we've observed so far in recent history on Earth? (personally I don't think it's just humans that are intelligent either though! so many animals and probably other things are extremely smart too! Just very hard to measure!)

And yeah I was only using the idea of a crystal as an analogy, all replicators (all life?) is a bit like a kind of smooshy-space-and-time-crystal in a way though right? Especially the multicellular kind!!

It's the regeneration/continuation of information of who-knows what type? All types! Obviously there's genes and Levin's vmem and other epigenetic stuff on the biological-side, but now there's also youtube, video-games, recipes, traditions? all memetic-replicators, or meta-memetic ones that control which other memes you allow in your life? I like to remember that all of them are subject to the rules-of-evolution too - there's a success for the memes themselves if they are getting us to carrying them forward!! Can be wildly different types of memories accross epic amounts of time too? Bloody amazing to think about I reckon! Evolution FTW!

> The great pseudoscience of this 'computer science' thinking

You spelled Nobel-level science wrong.

Most of Levin work is not easy to understand or appreciate. Our university lab has been doing cooperative intelligence for decades. The insight in some of his work is revolutionary. He shows why randomness is much more intelligent then most scientist think.

I think his work in biomechanics is easier to grasp than this, because the level of cognition he's talking about and working with there is (much?) closer to ours than what's described in this paper. It's easy to say there's no "there" there if you're unfamiliar with how his work is being successfully applied elsewhere.

There is a different between intelligence and cognition.

Any system which exhibits robustness against errors must to some degree be modeling its environment, and analyzing internal and/or external states in order to make better decisions.

I consider any such system to have a degree of intelligence, even if it lacks what we recognize to be cognition or conscious awareness.