Part 6 : Designing an Artificial Life Form : Considerations

6.1. Sensory types

The lifeform should have sensors to detect stimuli types that are necessary for its survival in that specific environment. More the types of relevant stimuli, the more feedback it can get from environment/internal sensors, increasing the chances of its longer survival.

6.2. Motor actions

The lifeform should be able to interact with its environment and influence its environment through 'motor actions'. (eg: locomotion)

6.3. Embodiment

The lifeform should be 'embodied' , ie., should have a perceived physical form. Such a perceived physical form makes it possible to create a sense of 'self' ,by demarcating self-generated stimuli and environmental stimuli.

The interaction of a lifeform with its environment through motor actions creates a change in sensory stimuli.(Eg: movement in a specific direction could create a change in visual input) The lifeform then learns the association between the executed 'motor action' and the resultant change in sensory stimuli from the environment.^[4]

When a lifeform starts its 'virtual life' in a new environment, it could execute all possible motor actions at random (motor babbling) to create the associations between the motor-actions and the resultant change in stimuli. Without embodiment, this association-creation is not possible. For example, in cases of lifeforms with limbs, the motor-action of moving a limb also generates proprioception sensory stimuli, which is internal stimuli that encodes relative position of a limb with respect to the body.

6.4. Reflex actions

Innate connections are required for incorporating 'primitive reflexes' like food-seeking and danger-avoidance.

Such primitive reflexes are required for initial survival until more complex behaviour is learnt using environmental/internal feedback. Some reflex actions can be made to fade over time (eg: fading of suckling , grasping reflexes in humans) after associations are formed to generate the equivalent voluntary behaviour

6.5. Autonomous Subsystems

An Artificial Lifeform should be made to feel 'virtual pain' in order to shape its behaviour. To aid its survival, certain critical motor-actions should not be controlled directly by the Artificial LifeForm.(eg: actions equivalent to breathing,digestion) To achieve this, autonomous subsystems should be incorporated in the Artificial Lifeform's design.

Taking a Human analogy, the "human digestive system" is an autonomous subsystem. When stale food is ingested, the digestive system triggers sickness and causes pain, discouraging future ingestion of stale food. The human-digestive-system is autonomous in that it cannot be forced not to cause pain. It has its own independent sensors,and an independent ability to trigger pain neurons.

The autonomous subsystem should therefore independent of free will and powerful enough to trigger pain or satiation, thus shaping behaviour. Such autonomous subsystems enable implementation of motivation, positive reinforcement and negative reinforcement.

An Artificial Lifeform should therefore have several autonomous subsystems that trigger pain when undesirable stimuli is detected by the subsystem (eg: detecting proximity to fire) The pain should be inflicted on the Artificial Lifeform ONLY by an autonomous subsystem within the Artificial Lifeform. Inflicting and stopping pain should be possible only by the Artificial Lifeform’s autonomous subsystem and not by the free-will of Artificial Lifeform. Similarly, triggering an "urge" and signalling "satiation" of the urge should be done only by the Artificial Lifeform’s autonomous subsystem and not by the free-will of Artificial Lifeform.

Such autonomous subsystems would help to :

Trigger quick reflex actions on certain stimuli , bypassing free-will (eg: analogous to human spinal reflex of pulling back arm on touching a flame)
Ensure continuous activation of highly critical motor sequences : It is important that some motor-actions be initiated involuntarily . For example, it is dangerous if life-critical actions such as breathing have to be initiated voluntarily. (Eg: Forgetting to breathe .) Other bodily functions such as digestion, involves a sequence of actions ( acidity increase in stomach, periodic contraction of stomach muscles, etc.) which should not be directly controllable by the lifeform's will.

When an Artificial Lifeform performs an undesirable motor action, the undesirable action will be detected by one of its autonomous subsystems , and pain is delivered to the Artificial Lifeform, thus dissuading similar undesirable behaviour in the future.

Various types of 'virtual pain' could be created, corresponding to different types of negative stimuli ( hunger, thirst, physical pressure). Various levels of pain could be created, because an incremental scale of pain could stop undesirable behaviour at earlier stages.

6.6. Inflicting pain on an Artificial Lifeform

'Virtual pain' can be designed to cause the following changes to the Artificial Lifeform : ^[5]

Pain could inhibit the motor-action neurons, reducing its responsiveness
Pain could prevent or intermittently block the flow of sensory data from the environment
Pain could trigger specific reflex actions, preventing other intended motor actions(eg: pain neurons trigger loss of balance maintenance, forcing robotic AI to bend/fall)
Pain could impair involuntary actions of its subsystems (eg: impair the equivalent of digestion)
Higher levels of pain could cause shutdown of all sensory information flow from environment
Higher levels of pain could cause shutdown of all motor action capabilities,effectively putting it in a "machine coma"

6.7. Creating a motivation system

Research has shown that non-satiation of an urge (eg:hunger) encourages more activity in pursuit to satisfy the urge.^[6]

An Artificial Lifeform can be designed with neural circuits that execute various combinations of possible motor-actions until it finds the means to satiate its urge. A need (food) creates an unsatiated feeling(hunger), which triggers discomfort(pain),which in-turn triggers some sort of action.

Therefore, when the Artificial Lifeform performs a desired set of motor action sequences (eg: eats virtual food), its own autonomous subsystem (eg: digestive system equivalent) detects the virtual food, inhibits the hunger/pain neurons, thus providing satiation.

By introducing different types of needs (food,water) and corresponding feelings(hunger,thirst), the Artificial Lifeform can :

be motivated to explore its environment
obtain positive feedback/reward when it performs a desirable action
obtain negative reinforcement(pain) when it performs an undesirable action.

Using such a motivation system and feedback system from its autonomous subsystems, its behaviour can be shaped.

Interactive examples

A plausible motivation circuit : with Positive reinforcement

In this example :
1. neuron U is the "urge" neuron triggered by an autonomous subsystem (eg: triggered by Hunger)
2. neuron S is the "satiation" neuron ,triggered by an autonomous subsystem, when the urge is satisfied(eg: signal that stomach is full)
3. neuron C1 and neuron C2 are the two possible voluntary motor actions that the Artificial Lifeform has the option to trigger.
4. neuron A is the decision neuron, which on firing, will cause either neuron C1 or neuron C2 to fire, with equal probability.

This is made possible by a fluctuating connection from U to A, which provides input signal of a random intensity to A, and the random signal is passed on to B1 and B2 which have different thresholds: A weak signal to A causes only B1 and C1 to fire ; a strong signal to A causes B1,B2,C2 to fire .

5. Let us say that, in this case, neuron C2 is the correct motor action neuron to be initiated for satiating the urge.
6. When neuron U fires, followed by neuron A, either C1 or C2 will be triggered.
7. In the scenario where neuron C2 fires, a satiation signal neuron S is activated by the autonomous subsystem.
8. As soon as the satiation neuron S fires, it creates a new dynamic connection in the recently activated pathway A=>B2=>C2
9. The next time A fires, the dynamic connections A=>B2 and B2=>C2 increase the likelihood of C2 firing, and the "reward" pathway A=>B2=>C2 is strengthened and used everytime for urge satiation.

A plausible motivation circuit : with Negative reinforcement

In this example :
1. neuron U is the "urge" neuron triggered by an autonomous subsystem (eg: triggered by Hunger)
2. neuron P is the "pain" neuron ,triggered by an autonomous subsystem, when an undesirable motor action is executed. (eg: proximity to heat)
3. neuron C1 and neuron C2 are the two possible voluntary motor actions that the Artificial Lifeform has the option to trigger.
4. neuron A is the decision neuron, which on firing, will cause either neuron C1 or neuron C2 to fire, with equal probability.

5. Let us say that, in this case, neuron C2 is a neuron that triggers an undesirable motor action .
6. When neuron U fires, followed by neuron A, either C1 or C2 will be triggered.
7. In the scenario where neuron C2 fires, a pain signal neuron P is activated by the autonomous subsystem.
8. As soon as the pain neuron P fires, it creates a new inhibitory dynamic connection in the recently activated pathway A=>B2=>C2
9. The next time A fires, the inhibitory dynamic connections A=>B2 and B2=>C2 decrease the likelihood of C2 firing, and the alternative pathway A=>B1=>C1 is used.