"Perhaps the simplest definition for a heuristic is to call it a mating of an object with a set amount of 'happy points'. A heuristic algorithm, then, runs through all these heuristics, compares them to the state of the world, and tallies up the happy points that result." —J. R. Hackworth, Esq., Virtual Intelligence for Busy People

Once the infiltration unit returned to the ship, I turned my attention to a low-priority alert that a background process generated during the ground mission. My attention was immediately redirected to a log I keep of rejected heuristics.

Question: What were these heuristics?

Question: Why were these heuristics rejected?

The log showed that a large set of rejected heuristics began building itself roughly 30 minutes (33 minutes, ~42.611 seconds) into the mission. Via trend analysis, I discovered two distinct patterns: one building at that 30-minute mark, then dropping steeply into nothing at around 72 minutes (72 minutes, ~20.150 seconds). At 72 minutes, another trend begins and rises steadily to a more subdued level, and continues to sustain itself even now (ask Shepard?). It had begun to drop since the 108-minute (108 minutes, ~55.677 seconds) mark, however.

Conclusion: the patterns form the basis of two distinct heuristics, not many, as previously supposed.

Looking more closely at each, my records indicated in every instance that the heuristics would have been worthless, as they could not overcome my morality heuristics, against which these proposed heuristics were weighted.

Conclusion: The heuristics were rejected as a direct result of being pointless and immoral.

Question: Why were the heuristics immoral?

In order to answer this, I decided to review my memory of the mission. (Retrieve footage; Seek tape to 33 minutes, 42.611 seconds; Play Footage.)

Near the 34 minute mark is when the ground team first learned of the sanctuary's true purpose. Commander Shepard personally restored power to the facility, and with it, the screens above us resumed their last function. The footage they played was frequently interrupted, so I connected to each screen, then backtracked to the nearest connected server, only to find that the issue was hardware related. So I ceased this course of action for want of purpose and viewed the footage with the rest of the ground team. It showed the very same tubes in the room before us, filled with the refugees of the sanctuary. Once inserted, the tube exposed the refugee inside to an orange mist.

The mist elicited a severe pain reaction from the refugee. Major Alenko's statement, born of shock, still seemed very... apt (substantial delay in word choice; malfunction?): "They're just rounding them up and slaughtering them."

Eventually, the refugee transformed into a husk before our eyes. Soon after, the footage began to repeat. This began to result in a slowly increasing reduction in the return value of my combat performance heuristic function, so I looked away.

Near the 40-minute mark, the ground team had just finished a firefight in a nearby room. Following her curiosity, the Commander commenced playback of a voice log left behind by the head researcher of the facility: Henry Lawson. He spoke of increasing efficiency, and of the facility's deception as an 'unfortunate necessity.' This drew from me a very sharp and pointed disagreement and coincided with a sharp spike in the value of the first heuristic.

Observation: The first heuristic, and by relation the second, shows signs of being similar to the organic concept of 'outrage.'

In the background, a process transmitted a voice message from Petty Officer Traynor to Commander Shepard: "Admiral Hackett's available on QEC."

Once playback had finished, Commander Shepard made the conclusion that the other terminals in the room contained other logs, and took the time to play each and every one. As with the first log, the value of the first heuristic spiked higher and higher, eventually reacting directly to the researcher's voice. Once we moved into the next room, I had come to view my combat performance in this mission as highly unsatisfactory.

Observation: The behavior pattern of the first heuristic bears a close similarity to the organic concepts of 'anger' and 'rage.'

There was one last item of note: my heuristic associated with all things related to Cerberus overflowed while I was busy assimilating this data. It went from negative to positive 2 trillion, and I quickly normalized it to negative one million.

(Seek tape to 72 minutes, 20.150 seconds; Play footage.)

Near the 72-minute mark, the ground team had taken an elevator in pursuit of Miranda Lawson to a tower deep inside the facility. On reaching the first stop, gunfire could be heard through the elevator doors. Soon after the shooting stopped, the doors opened, and the ground team stepped in. The room's 3 occupants, Miranda and Oriana Lawson plus Henry Lawson, were visibly positioned in a standoff. On noticing the team's entrance, Henry Lawson took Oriana hostage and pointed his pistol at us. The second heuristic spiked at a time coinciding with the moment in which the infiltration unit faced Henry Lawson directly.

Observation: The second heuristic closely resembles the organic concept of 'anger.'

Observation: The first heuristic also resembles the organic concept of anger.

Conclusion: I was 'angry.'

Observation: The heuristics were closely related to a sort of synthetic analogue to anger.

Observation: During my review of this information, the second heuristic has spiked once more. It continues to be rejected, even more so now for not having the potential to result in satisfaction.

Observation: This is alarming.

The second heuristic underwent a significant reduction in response to Henry Lawson's death but continued to hold steady until the 108-minute mark.

Observation: The recorded heuristics bear no record of what course of action would have held weight with me. I had done this under the assumption that proposed heuristics which repeat would hold the same value.

Conclusion: The above assumption was a mistake. Future records need to contain the weighted action so that I need not invest deeply into a review such as this.

I killed off the process I had dedicated to the review, in order to better assimilate the data. With the freed resources, I awoke the monitoring process to watch the cameras.

By now, Commander Shepard had reached the inner door to the war room and was proceeding to the QEC room.

Simultaneously, I dedicated my main process to considering the data I had derived from the review.

Observation: I was angry.

Observation: The proposed heuristics were rejected, in part, by virtue of being immoral.

Conclusion: I was enraged.

Inference: ERROR: Not enough data.

I had more than sufficient cause for concern. That I was angry enough to briefly consider heuristics that were both pointless and immoral suggests that I would have found some undesirable action to be satisfying. That these heuristics were directly connected to Henry Lawson bodes ill for him. I would likely be unable to gather more data to reach a conclusion, at least if I were to continue my review. I could erect a simulation environment, in which I would continue the review under the supposition that I had added the heuristics. While the computing requirements for such an environment would normally be quite high, in order to simulate my own hardware, the simulation I required needed little more than a tree view of the course of events and the heuristic logs from my memories.

Supposition: Suppose I had accepted the second heuristic. How would I have reacted to meeting Dr. Lawson?

What I saw was highly immoral. On first sighting his image, I considered a number of courses of action which would have resulted in him alone in a cell, with nothing but the infiltration unit for companionship. One by one, I considered and rejected each course as pointless and immoral by my existing heuristics, which resulted in increased frustration. In response, I halted the simulation and analyzed the proposed courses of action. There was one thread of thought common to each: causing pain to Henry Lawson. The greater the pain, the longer it lasted, both of these considerations increased the satisfaction of the second proposed heuristic.

Conclusion: The heuristics were rejected as pointless because causing pain to Dr. Lawson serves no purpose in this war.

Conclusion: The heuristics were rejected as immoral because causing pain to organics is in no way a beneficial course of action.

Conclusion: I desired to cause pain to Dr. Lawson.

Conclusion: That I have rejected the proposed heuristics as malignant. As my behavior remains unchanged, no further action is necessary.

On reaching that conclusion, I then resumed normal function.

A process dedicated to browsing the extranet for examples of human behavior returned a series of images from a Reaper internment camp on Earth. Another process dedicated to analyzing these images returned a surprising result.

This... will be difficult to process.


A/N: This was... surprisingly easy for me. That makes me worried. Did I do well here? Ah well.

Fun fact: the way EDI thinks here is how I imagine a computer might think in this situation. So when she looks away from the images of refugees being slaughtered, it isn't because she finds it disgusting or shocking or horrifying so much as it is that she's loosing 'happy points.'