I shall now declare unto you in full this knowledge, both phenomenal and numinous. This being known, nothing farther shall remain for you to know- says Lord Krishna to Arjuna’

Last time, I had shown that our minds consist of a large no of abstract concept like hunger, fear, build blocks etc. We had shown that these could be represented by goal trees. Each region of the brain can be though of as a collection of unintelligent agents. Each such agent can be a part of multiple goal trees like hunger, fear etc. since our minds have a limited no of such agents, but can perform a large no of tasks. Today, I shall make it my aim to make clear the nature of destructiveness, and pain which humans feel and how it might work in a society of agents.

Destructiveness:

Consider, a society of agents in the following scenario:

Suppose the child is playing, and starts to feel sleepy. What happens in that situation? If the machine is in some step of the play’s goaltree, how does it terminate? Does the child absolutely pause playing with the blocks and automatically proceed to sleep. No, in fact the child might kick the blocks in frustration and then proceed to sleep.

Such fluctuations in emotions might be assigned to a random fluctuations of the child’s self. However, I argue that this can be modelled through a computational procedure. The child can be considered on two levels. On external level, he is an entity who impacts the world through his arms. On an internal level, he is a society of agents whose outputs govern the outside interactions. When the child gets frustrated, his outside body performs the action. This sends a ‘garbage collection’ interrupt to the internal “play’s goal tree”. Rather than switching to the sleep agent instantaneously, such a machine then waits for the child to physically topple all the blocks. Then, the system starts to move into the sleep goal tree.

However, each goal tree must possess two copies of itself. The first copy would signify our physical body is performing a particular action (for eg, the child is actually playing the block). The other copy represents that the same action , is being performed in the machines brain but not physically, (for eg, the child ‘thinks’ that he plays with the block’). Thus I shall call these goal trees as a ‘physical goal tree’ and a ‘dream goal tree’.

Once an agent takes over, i.e. sleep, it signifies a physical action. Thus the machine then activates the physical goal tree of the ‘sleep’ agent. Then, the ‘dream goal tree’ of other activities shall also become activated simultaneously. The ‘dream goal tree’ and ‘physical goal tree’ signify two important concepts of human interpretation:

  1. the physical actions we perform
  2. the thoughts we have.

Also, we would have a mechanism for switching between these ‘physical goal trees’ and ‘dream goal tree’.For this, consider the situation that the child stacked only one block in his tower and went to sleep. His dream goal tree for building blocks is then activated. In his brain, he thinks that he has already stacked second and third block on top of each other. However, when he wakes up he has to start physically again from the second block, even when in his brain the third block had been placed. I shall handle this with a synchronization process between the dream goal tree and the physical goal tree.

Another aspect of this synchronization is when perturbations in the dream goal state, could in fact be applied directly to the physical world. For eg, suppose the child has to solve the arithmetic problem. He thinks on it and comes with an approach. When he switches to his physical action, (i.e. picking up the pen and actually deriving the answer), this switch remains seamless. Whatever the child has come up in his dream goal tree, forms a direct input knowledge for the physical goal tree.

Coming back to my initial example, an interrupt in the play state requires the mind to wait for the child to topple blocks over, before switching to the sleep state. To outside, the action seems a random outburst of the childs self. But, what this serves as is a satisfaction criteria for an internal ‘self’ agent, which acts as a sort of garbage collection process to free the mind’s internal agents. Therefore, my next attempts shall be on how to model this self.

On the dynamics of Pain:

It has dawned upon me that human emotions lie on the opposite ends of a continuous spectrum. Our abilities to feel emotions is continuous, but our words for what those emotions mean only quantify the opposite extreme ends of that scale. However, from a computation perspective this continuous traversal of emotions can in fact be modelled as the physical actions of an agent.

Consider for example the meaning of pain. Pain numbs us. We want it to stop. Thus, all our actions are on coming out of it. This signifies a goal tree, where we are continously in the traversal loop. We never come out of it, till we finish it. In-fact the pain’s goal tree holds highest precedence of other mind agents like sleep, and even tends to permeate into them (hence the account for disturbed thoughts during sleep).

Now consider pleasure. It is opposite of pain. We want it to continue. From a machines perspective, we are still stuck in the limbo of traversing the goal tree. Next, consider a general situation in between. We receive a news to which we tend to react neutrally. Therefore, this news doesn’t impact us and hence allow us to come out of the goal trees naturally, and other agents to take over. Hence, it is my understanding that the emotions can be modelled on a continuous scale, provided we know two things.

The first thing is to know which goal tree a particular group of emotions belong to. Next, each such emotion is directly linked to the amount of time we shall spend in that goal tree. At the same time, a parallel contest shall occur between several parallel agents (like eat, sleep etc.). Once the goal tree gets exhausted, or forcefully interrupted, the machine then proceeds to the dominant agents goal tree.

Finally, I argue that the emotions can in fact be interpreted on this constant scale. For eg, sound/silence, interest/boredom, hunger/not hunger depend on these continuous scales. What then remains is to model the self, which could be thought of as a collection of consciousness and super consciousness agent. But, the nature of this self has yet to reveal itself to me, which i hope that the time shall make clearer.

rajat credits: minsky,watson,csail and mila