7
$\begingroup$

A question about swarm intelligence as a potential method of strong general AI came up recently, and yielded some useful answers and clarifications regarding the nature of swarm intelligence. But it got me thinking about group intelligence in general.

Here organism is synonymous with algorithm, so a complex organism is an algorithm made up of component algorithms, based on a set of instructions in the form of a string.

Now consider the Portuguese man o' war, not a single animal, but a colonial organism. In this case that means a set of animals connected for mutual benefit.

And physalia physalis are pretty smart as a species in that they've been around for a while, I'm not finding them on any endangered lists, and based on their habitat it looks like global warming will be a jackpot for them. And they don't even have brains.

Each component of the physalia has a narrow function, colony organism organism itself has a more generalized function, which is the set of functions necessary for maintenance and reproduction.

{Man o' War} ⊇ { {pneumatophore}, {gonophores, siphosomal nectophores,vestigial siphosomal nectophores}, {free gastrozooids, tentacled gastrozooids, gonozooids, gonopalpons}, {dactylozooids}, {gonozooids}, {gastrozooids} }

  • What types of applications qualify as "compound intelligences"? What is the thinking on groups of neural networks comprising generally stronger or simply more generalized intelligence?

I recognize the underlying problem is ultimately complexity and that "strong narrow AI" is, by definition, limited, so I use "generalized" and omit "strong" because because human-like and superintelligence are not conditions. Compound intelligence is defined as a colony of dependent intelligences.*

Utility software is often a form of expert system that manages a set of functions of varying degrees of complexity. There's currently a great deal of focus on autonomous vehicles, which would seem to require sets of functions.

Links to research papers on this or related subjects would be ideal.


Portuguese Man o' War (oceana.org)

The Bugs Of The World Could Squish Us All

$\endgroup$
  • $\begingroup$ The main problem in analyzing intelligence of swarm intelligence is due to their genetic link...they have become intelligent by evolution, whereas we humans pass our knowledge and sometimes become intelligent by practicing, learning. $\endgroup$ – DuttaA Apr 4 '18 at 5:20
  • $\begingroup$ @DuttaA per the reference question, I was careful to avoid the term "swarm", because, even though emergence is tagged, I'm thinking more about extensions of classical expert systems, as opposed to bot networks. (Glad I got at least one upvote though--wasn't sure what the reaction would be. But we haven't had a lot of theoretical questions lately, so I figured, "why not?".) $\endgroup$ – DukeZhou Apr 4 '18 at 17:31
  • 1
    $\begingroup$ this is one of the most underrated fields...we need to explore more in this area....not many books also exist on these topics....i wonder why people are ignoring this topic as such $\endgroup$ – DuttaA Apr 4 '18 at 18:13
1
$\begingroup$

Swarm intelligence, compound intelligence, or group intelligence may emerge as an important concept as AI develops toward higher complexity. Whether these terms should be considered synonymous is doubtful.

Compound features in biology are the result of control in differentiation during the development of an organism from a single cell. Compounding in biology is a single function performed across like elements.

Swarms are a result of distinct organisms operating in proximity. Swarming in biology is similarity in complex independent behaviors that appear coordinated but are generally engaged in as a defense from predators.

Group intelligence may be distinct from compound intelligence in that adjacent units may be in agreement or opposition, as can be case with intelligent beings in groups. The agreement in the group that a variety of models is permissible. We call them opinions, but they are distinct matches of models to problems that produce different projections and suggest different selections from among the group's options.

The terms general and strong appear in this question and many others and it appears that, mainly for historical reasons, may continue to frustrate clarity.

All intelligence is general in that it learns a generality from specific experiences and then applies that generality to future scenarios to achieve objectives within those scenarios. What makes the application of the generality intelligent is that it is expected to work because it has been working for similar scenarios.

All intelligence is specific in that it is limited to scope of what generalities have been discovered. Certain techniques are significantly more general that others because they are allegedly domain independent. We call this mathematics.

Consequently, the development of artificial intelligence is not a path from specific to general but one of discovering generalities and applying them to specifics. One could say that the most general thinking of mathematics applied to the computer is the primary activity of applied artificial intelligence, and the capabilities that emerge are a more sophisticated as more sophisticated mathematics is represented in working software and hardware.

The generalities take on greater complexity so that they can apply to a greater number of specific scenarios. That is not a gain in strength but a gain in the breadth of potential application.

Physalia physalis is a symbiotic colony of organisms of four types, the pneumatophore (or float), dactylozooids (long tentacles), gastrozooids (feeding tentacles), and gonozooids which produce reproductive gametes. To avoid breaking the historic conception of animals, these organisms are called polyps. The venom of the colony does not come from any of those organisms but rather from another symbiotic organism, cnidocytes, which attach to the tentacles and are released under strictly controlled conditions.

In addition to all this symbiosis, there are several species of fish that use the colony for shade and protection that the colony lets swim among the tentacles and that have formed partial immunity to the cnidocytes. These complex symbiotic networks are not fully understood, but they seem to operate well and create sustainable inter-specie systems.

It's not likely that the organisms, within their lifetimes adapt or remember, but the DNA of each organism has involved in a way that resembles intelligence in that the colony and its symbionts have adapted to the ocean's surface and its biology.

The idea that the system of the biosphere is the first example of broad intelligence is likely, in that design excellence has emerged from an evolutionary processes. It is not altogether ridiculous to propose that human intelligence is a higher speed approach in neurons to the slower speed of DNA replication in larger organisms. Because of the metabolic requirements of growth slows evolution for larger organisms with a greater cell count, these larger organisms may have needed to develop a way to achieve the nimble adaptivity of their lower ancestors.

Neurology facilitates the approximation of some aspects of evolution and may have been the most attainable natural solution to reacquire nimble adaptivity.

Attempting to apply these various ideas to the current and ongoing development of autonomous vehicles reveals a gap in understanding. We don't yet have the mathematics developed to understand how compound, swarm, or group intelligence can be used in the laboratory to accomplish well defined problems in controlled execution scenarios. That is probably a prerequisite to using these ideas in vehicle control.

The system designs of future cars may be like a colony of independent or semi-independent components that each have a role and purpose. Compound, group, and symbiotic designs are likely to develop. The current activities of vehicles on the road or in the air near airports is like a swarm, so the application of that idea is obvious: Avoid collisions.

$\endgroup$
0
$\begingroup$

With the increase of both unit capacity turbines and size of wind farm, the safe operatio wind farm has received growing attention. In ma factors that affect the safe operation of wind far being struck by lightning is an important aspect. intelligent lightning monitoring system is used fo surveillance of wind turbine generators, which ca real-time and accurate monitoring of the lightnin current waveforms, amplitude, time of occurrenc number of lightning strokes and all the other imp parameters of lightning thus providing an effectiv monitoring and analysis tool to quickly locate fa location on wind turbine generator and cause of malfunction and a theoretical basis for the desig lightning protection system of wind turbines. Thi examines the principle and main methods of the turbine generator-matching lightning monitoring and combined with the specific research project designs and implements a high-precision and multifunctional intelligent lightning monitoring sy based on the theory of Rogowski coils.

$\endgroup$
  • 1
    $\begingroup$ I think you are referencing two different types of control systems in a symbiotic relationship? If so, could you add a synopsis to clarify? $\endgroup$ – DukeZhou Oct 23 '18 at 20:19
-1
$\begingroup$

I did work on compound intelligence because that is the direction that Google is trying to go. I couldn't find any basis for it. In other words, having a collection of AI expert systems does not seem to provide any collective intelligence. You would also need some kind of control program that could decide which system to use. Currently, Google relies on the user to choose. If Google was able to create an independent control program it would already be out. This does not seem to be a matter of complexity or code tweeking but a fundamental limit with AI.

$\endgroup$
  • $\begingroup$ Great info. Thanks for posting! PS- is there a formal term for this concept? (in some sense, I was more thinking about integrating neural networks as components in expert systems, but also tried to make the question fairly general.) $\endgroup$ – DukeZhou Apr 5 '18 at 18:48
  • 1
    $\begingroup$ I don't know if there is a formal term; I called it the Swiss Army knife strategy. It is possible that neural networks could be components in a future cognitive system but you can't build one using NN alone. $\endgroup$ – scientious Apr 5 '18 at 19:54
  • $\begingroup$ Yes, I did look at social insects. There is no actual intelligence there. I wrote a paper on information storage in non-replicating and replicating systems (like DNA). Viruses do not exhibit any intelligence as a group. You seem to be confusing intelligence with adaptive or shotgun environmental reactance. $\endgroup$ – scientious Oct 24 '18 at 6:10

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.