The Octopus as a Model for Artificial Intelligence

A Multi-Agent Robotic Case Study

Alfonso Íñiguez

Swarm Technology, LLC Mesa, Mesa, AZ, U.S.A.

Keywords: Autonomous, Robotics, Artificial Intelligence, Multi-Agent, Swarm Intelligence, Parallel Processing.

Abstract: The aim of this paper is to investigate the curious cognition process exhibited by the octopus, and its

practical applicability to multi-agent systems. The paper begins by explaining the limitations of using the

human brain as a model to achieve artificial cognition and proposes an alternative model inspired by the

octopus’ distributed approach to solving problems. As a case study, a laboratory prototype demonstrates

awareness, autonomy, solidarity, expandability, and resiliency in a multi-robotic system. The cognition

model described in this paper is primarily algorithmic and does not explore the model creation process nor

semantics; rather, it lays the foundation and inspiration for a future realization as a Process for Agent

Societies Specification and Implementation (PASSI).

1 INTRODUCTION

The conventional approach in the attempt to achieve

artificial intelligence is to use the human brain as a

model of operation; in fact, there are operational

similarities between the computer processing unit

(CPU) and the human brain – both make decisions

by fetching and processing data from memory, and

both store the processed data in memory.

The downside of using the human brain as a

model for artificial intelligence is scaling; as the

complexity of the tasks increase, the performance

demand on the CPU proportionally increases.

Limitations remain even with the advent of

coprocessors — intended to offload tasks from the

CPU. The traditional multiprocessing framework

(see Figure 1) suffers from two major drawbacks,

both caused by the architectural requirement that the

CPU must divide and distribute the threads.

First, a significant amount of the CPU's

processing time is consumed in managing the

coprocessing tasks. The management may include:

distributing tasks to the coprocessors according to

their capabilities, waiting for those tasks to be

completed before reassigning new tasks; and

responding to interrupts from coprocessors every

time a task is completed.

Second, a coprocessor will remain idle as it

awaits for a thread to be assigned to it by the CPU.

A multiprocessor system that alleviates the

management workload on the CPU while keeping

the co-processors busy is needed.

Figure 1: Traditional multiprocessing framework.

Given that robotic movement is ultimately

enabled by its processing capability, the same two

drawbacks that affect CPU/Coprocessor

performance limit robotic autonomy. Hence, to

enable robotic autonomy, it makes sense to begin by

solving the two computer processing drawbacks.

Ã Ã

siguez A.

The Octopus as a Model for Artiﬁcial Intelligence - A Multi-Agent Robotic Case Study.

DOI: 10.5220/0006125404390444

In Proceedings of the 9th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2017), pages 439-444

ISBN: 978-989-758-220-2

439

2 OCTOPUS COGNITION

Written records of octopuses leaving the water have

existed for over 2,000 years (Balme, 1991). The

Octopus Alpheus is known to leave the water to

crawl between tide pools (Norman, 2000). More

recently, Boyle (1991) wrote, “Octopuses are

particularly prone to escape from aquarium tanks.

Loose lids are of little value because the octopuses

will easily lift them and push their way out of the

tank” (Boyle 32).

2.1 Cognition Evidence

It is not surprising to learn that the octopus is

considered to be the most intelligent of all

invertebrates (Linden, 2002); it learns simple mazes

(Boal, 1996), uses landmark navigation while

foraging (Mather, 1991), and uses tools (Mather,

1994).

2.1.1 Cognition Efficiency

Experimentation results do not imply that octopuses

are smarter than human children; however, the

octopus is a model for efficient cognition given the

limited amount of available neurons in its brain —

500 million in the octopus as opposed to almost 100

billion in Homo sapiens.

Biologists at the Seattle Aquarium challenged a

female Enteroctopus dofleini — a giant Pacific

octopus — with a childproof bottle, the kind that can

puzzle Homo sapiens. The results were staggering,

“To open the lid it was necessary to push down on

the lid at the same time as turning it … the octopus,

accomplished this task in 55 minutes … Further

presentations resulted in a decrease of the average

opening time to 5 minutes” (Anderson, 2006).

2.1.2 Distributed Cognition

Distributed neurons allow the octopus’ arms to

problem-solve autonomously; the “arms are not

entirely under the control of the octopus' brain . . .

two thirds of its neurons reside not in its central

brain but out in its flexible, stretchable arms”

(Harmon, 2013).

The “Octopus’ arms have a mind of their own

… as a result, the arms can problem-solve how to

open a shellfish while [the octopus] is busy doing

something else, like checking out a cave for more

edible goodies” (Nuwer, 2013).

2.1.3 Arms React after Detachment

Researchers, working at St. George's University of

London and the Anton Dohrn Zoological Station in

Naples, Italy, demonstrated that, "the arms are

capable of reflex withdrawal to a 'noxious' stimulus

without reference to the brain." (Harmon 2013a)

Other experiments show an active nervous system

after detachment, “the arms can react after they’ve

been completely severed. In one experiment, severed

arms jerked away in pain when researchers pinched

them" (Nuwer, 2013).

2.1.4 Arm Ambidexterity

A series of interactions were performed to determine

if the octopus (Enteroctopus dofleini) had arm

preference when reaching for objects; the results

supported the hypothesis of ambidexterity of the

arms. All arms are equally willing to work; arm

selection is based on availability and relative

proximity (Wülker, 1910).

2.2 Summary of Principles of

Cognition

After investigating the behavior of the octopus and

the embedded cognition of its arms, we can clearly

see that the octopus — when viewed as a processing

system — is a superb model for efficient cognition.

Let’s now generalize the cognition principles

governing the octopus’ system. As a way of keeping

the principles as generic as possible, the arms will be

referred to as “members” and the octopus will be

called “system.”

2.2.1 Principle 1: Member Awareness

Each member must be aware of its surroundings and

abilities. This principle is derived from the fact that

each arm can react to its environment even when

detached from the head.

2.2.2 Principle 2: Member Autonomy

Each member must operate as an autonomous

master (not as a slave); this is essential to self-

coordinate allocation of labor. This principle is

derived from the fact that the arms are not entirely

under control of the octopus’ head.

2.2.3 Principle 3: Member Solidarity

Each member must cooperate in solidarity; when a

task is completed each member should

ICAART 2017 - 9th International Conference on Agents and Artiﬁcial Intelligence

440

autonomously look for a new task (leveraging its

current position). This principle is derived from the

observed ambidexterity of the arms.

2.2.4 Principle 4: Member Expandability

The system must permit expansion where members

are dynamically aggregated. This principle is

derived from the fact that octopuses can regenerate

lost arms with ease (Harmon, 2013b).

2.2.5 Principle 5: Member Resiliency

The system must be self-healing; when members are

removed, the remaining members should undertake

the unfinished tasks. This principle is derived from

the fact that losing an arm is not considered

traumatic; octopuses occasionally lose an arm in

nature and function normally while the limb

regenerates (Levy, 2014).

The attributes described above may also be

referred to as the Five Principles of Swarm

Intelligence (Íñiguez, 2016).

2.3 Octopus’ Cognition Model

After defining the principles of the octopus’ arm

behavior, the next step in abstracting the octopus’

cognition model is to define an architectural

representation (see Figure 2).

Figure 2: Octopus’ Cognition Model.

3 APPLICABILITY OF THE

COGNITION MODEL

Notice the differences between the traditional

multiprocessing framework of Figure 1 and the

octopus’ cognition model of Figure 2. There are two

main differences:

First, the coprocessors of the traditional model

became masters instead of slaves.

Second, the CPU, which is equivalent to the

octopus’ head, does not directly communicate with

the coprocessors; instead, the coprocessors

autonomously read the octopus’ intentions, i.e. seek

tasks from the task pool.

Figure 3:

Solidarity Cell Architecture

3.1 Transposing the Cognition Model

The transposing of Figure 1 into Figure 2 resulted in

Figure 3. A fundamental principle of operation of

the proposed model is the cooperation in solidarity;

since each member is a processing cell in the

system, we will refer to this cognition model by the

name of Solidarity Cell Architecture (SCA).

The SCA model solves the two limitations of the

traditional model previously described in the

introduction. In the SCA model the CPU does not

spend a significant amount of time micromanaging

coprocessors — just as the octopus’ head does not

spend time micromanaging the arms — and the

coprocessors do not remain idle waiting for tasks to

be assigned.

The Octopus as a Model for Artiﬁcial Intelligence - A Multi-Agent Robotic Case Study

441

3.2 Further Description of the

Solidarity Cell Architecture

In general terms, the SCA model is to be described

as a method for processing information in parallel;

the system uses autonomous computer processing

cells to perform tasks needed by a central processing

unit. Each cell in the system is connected through a

switching fabric, which facilitates connections for

data transfer and arbitration between all system

resources. A cell has an agent, which is a software

module that may be transferred through the

switching fabric to a task pool containing the tasks.

The agent searches within the task pool for available

tasks that match the cell's instruction type. A task

may be broken into threads that are to be executed

sequentially or independently depending on recipes

constructed by the central processing unit.

Interdependent tasks within the task pool may be

logically combined as needed by the recipe. A

notification is sent from the task pool to the central

processing unit when a task or task thread is

completed.

Therefore, it is an object of this new architecture

to provide a method for parallel processing in a

multiprocessor system using coprocessors — or

autonomous robots — that proactively seek threads

to process (Íñiguez, 2013).

3.3 Applicability into Robotics

A recurring challenge in robotics is to build a biped

robot that has the balancing ability of humans. A

mechanism to account for continuous balancing is

needed; as the robot walks, climbs, or bends, it

needs to swing its arms autonomously to keep its

balance.

Figure 4 shows the conceptual balancing

mechanism; the shoulders and elbows — a, b, c, and

d —are equipped with actuators that continuously

and autonomously send wireless software agents to

seek tasks from the intention’s task pool – the

intention’s task pool is a module in which the

system’s central brain deposits its desire to maintain

balance. In this example, the intention’s task pool is

analogous to a cerebellum in charge of coordinating

and maintaining balance. However, as opposed to a

traditional cerebellum that sends commands to the

body, the biped robot follows the octopus’ model, in

which the arms autonomously send inquiring agents

to the task pool.

The breakthrough advantage of this

implementation is that the central brain system can

delegate the task of maintaining balance to an

Figure 4: Mechanism to account for continuous balancing.

electronic gyroscope that constantly deposits the

balancing requirements into the intention’s task

pool.

If the biped robot begins to lean to the left, then

the gyroscope will deposit an intention in the task

pool named “I need somebody to help me move my

center of gravity to the right.” The shoulders and

elbows continuously send agents looking for tasks;

when an agent finds a task in the task pool, it returns

to its actuator to execute the requirement. The

process is repeated continuously achieving human-

like balancing without direct intervention from the

central brain system.

4 ROBOTIC CASE STUDY

As a way to demonstrate the SCA model with a

proof-of-concept prototype, we adapted the biped

robot of Figure 4, into the streamlined laboratory

representation shown in Figure 5. Free-moving

wireless-connected tank robots represent the

shoulders and elbows. Hence, the designations of a,

b, c, and d originally used by the shoulders and

elbows in Figure 4 are now given to the tank robots

in Figure 5.

The intention is implemented by a gyroscope that

places “move left” or “move right” tasks into the

task pool.

ICAART 2017 - 9th International Conference on Agents and Artiﬁcial Intelligence

442

Figure 5: Proof-of-concept laboratory experiment.

The original octopus cognition model can be

transposed into the robotic architecture without

modification; the only difference between Figure 2

and Figure 6 is the terminology.

Figure 6: Block representation of the laboratory prototype.

4.1 Prototype Implementation

To build the prototype, we selected to work with off-

the-shelf components. The microcontrollers are

Arduino

boards, the robots are a customized

version of Makeblock

toy

tanks, and the wireless

communication is through Xbee

(Zigbee)

transceivers. We have a YouTube video illustrating

the interaction between the tank robot and the

gyroscope:

https://www.youtube.com/watch?v=jq1EfxkneJI

Figure 7: Toy robot interacting with the gyroscope.

4.2 Complete System Implementation

The demonstration of the proof-of-concept prototype

exhibiting the five principles of cognition — also

known as the five principles of swarm intelligence

— is also available via YouTube:

https://www.youtube.com/watch?v=axxXz2BM0yw

Figure 8: The Five Principles of Swarm Intelligence.

5 CONCLUSION

Various companies and academic institutions are

actively researching the field of swarm intelligence;

a search on the topic reveals two distinct

approaches:

a) Each member is controlled through a central

computer, e.g. Intel’s 100 drones (Geiver, 2016).

b) Each member behaves autonomously without a

central computer; e.g., Harvard University’s 1024

Robot Swarm (Hotz, 2014).

Both approaches have merits and limitations

(Íñiguez, 2016).

In the case of a, members are slaves in a system

controlled by a central computer with sufficient

channels of communication. The results can be

visually spectacular — as illustrated by Intel’s

The Octopus as a Model for Artiﬁcial Intelligence - A Multi-Agent Robotic Case Study

443

drones. However, since a central computer dictates

the movement of each member, there is limited

flexibility to adapt to changing environments, such

as: x) members lost to unforeseen events, y)

members added to speed up the mission, or z)

members autonomously self-allocating labor.

Of course, the intricacy of the central software

may be increased to account for x, y, and z, but that

would make the central computer responsible for

real time response, it would increase vulnerability

due to single point of failure, and it would deviate

from the concept of swarm intelligence which is

defined as the collective behavior of decentralized,

self-organized systems.

In the case of b, members have the autonomy to

adapt without a central dictator. Considering that

each member possesses modest processing power —

as illustrated by Harvard University’s swarm of

robots — the results are truly impressive;

nevertheless, this type of behavior falls into the

realm of swarm flock. It does meet swarm

intelligence’s basic definition of collective behavior

of decentralized, self-organized systems, but it still

lacks the ability to autonomously distribute and

undertake allocation of labor.

If neither a nor b meets the requirements of

autonomous allocation of labor, then we need a

different approach.

As demonstrated in the proof-of-concept

protopype, the Solidarity Cell Architecture

effectively achieves the principles of awareness,

autonomy, solidarity, expandability, and resiliency;

it also solves the two major drawbacks described in

the introduction, i.e., CPU micromanagement and

coprocessor idleness, present in the traditional

multiprocessing framework.

REFERENCES

Balme, D. M. (1991). Aristotle: History of animals.

Cambridge, MA: Harvard University Press.

Norman, M. (2000). Cephalopods: A world guide.

Hackenheim, Germany: Conch.

Boyle, P. R. (1991). The UFAW handbook on the care and

management of cephalopods in the laboratory. Potters

Bar, England: UFAW.

Linden, E. (2002). The octopus and the orangutan. Dutton

(NY). 242 pp.

Boal. J.G. (1996). A review of simultaneous visual

discrimination as a method of training octopuses.

Biol. Rev. 71:157-190.

Mather, J.A. (1991). Navigation by spatial memory and

use of visual landmarks in octopuses. Journal of

Comparative Physiology. 168(491-497).

Mather, J.A. (1994). “Home” choice and modification by

juvenile Octopus vulgaris (Mollusca: Cephalopoda):

specialized intelligence and tool use? J. Zool. Lond.

233:359-368.

Anderson, R.C (2006). The Festivus Seattle Aquarium

38(1):7-9 TheCephalopodPage.org.

Nuwer, R 2013. Ten Curious Facts About Octopuses.

smithsonian.com.

Harmon K. (2013a). Octopus Chronicles, Even Severed

Octopus Arms Have Smart Moves.

Wülker (1910). Choice of arm usage in a Giant Pacific

Octopus, Enteroctopus dofleini.

Harmon K. (2013b). Octopus Chronicles How Octopus

Arms Regenerate With Ease.

Levy G. (2014). Self-Recognition Mechanism between

Skin and Suckers Prevents Octopus Arms from

Interfering with Each Other.

Íñiguez A. I. (2016). The Five Principles of Swarm

Intelligence, LinkedIn.

Íñiguez A. I. (2013). Parallel processing with solidarity

cells by proactively retrieving from a task pool a

matching task for the solidarity cell to process. U.S.

Patent 9,145,777.

Geiver L. (2016). Intel’s Entrance Into UAS, UAS

Magazine

Hotz R. L. (2014). Harvard Scientists Devise Robot

Swarm That Can Work Together. The Wall Street

Journal.

ICAART 2017 - 9th International Conference on Agents and Artiﬁcial Intelligence

444