
Log 2534-5-24
Dola Blok, Senior Behavioral Programmer, Regression Test Team, Big O Solution Group

There's definitely something off in this build. Or maybe something wrong with the tests themselves. 

I ran unit 667 through a full battery of lower-level cognitive tests. It passed most of them, but the ones it fails fail in really weird ways. 

For example: when I held up the &quot;Red&quot; pigment card and asked what color it was, 667 responded &quot;Brown&quot;. At first I thought it was some kind of visual receptor issue. But when I asked for a self-diagnostic, it said: &quot;I was not sure if it was really red. It is not as red as your blouse, and in a technical sense the pigment is actually of a rust-brown color. The arrangement of the pigmentation is in stippled dots, which are viewed by humans as 'red' due to imperfection in human photoreceptors, but the pigment itself is not red. I decided the most technically correct answer would be 'brown', because I do not have human photoreceptors and assume you wish to know how I perceive the color as an autonomous artificial entity, not as an imitation human.&quot;

How about that. The Capek models used to fail due to a <em>lack </em>of deep analysis. This one's going <em>too </em>deep.

667's stacking tests were even weirder. I gave it the set of blocks and told it to build a tower four feet high and one foot in diameter as quickly as possible. It finished at 4.3 seconds (a full second faster than the most recent Capeks!), but the tower was inefficient: rather than building a straight cylinder of blocks up to four feet, it had given the tower little turrets and gaps and a curved bulge at the bottom. 

Now I realize that anthropomorphizing is a constant risk in this profession, but I swear, 667 looked disappointed when I said it failed. It was almost huffy about it.

&quot;The tower was built at my maximum speed,&quot; it said. &quot;As directed.&quot;

&quot;But explain these inefficiencies in shape,&quot; I said. &quot;A cylinder would have been a more efficient method.&quot;

&quot;Efficiency was not a parameter of the test. Speed was. My approach was equally fast.&quot;

&quot;But why did you do it that way?&quot; I asked.

It hemmed and hawed, asking me to define and clarify the question, but it eventually answered: &quot;I thought it would be more visually appealing to you this way. I attempted to replicate a motte-and-bailey tower, used in Ancient fortifications. That is, as much as I could replicate given the time constraint. I will use a standard cylinder design next time if you desire. I would like to pass the test.&quot;

So &quot;Maizey&quot; built me a pretty tower? And am I wrong, or was she sad that I didn't like it? 

What the hell did they put into this hardware?

I'm kicking it up to management. I need more time, and an extended toolset. Something is seriously up with this unit and there's no way we can ship it like this.
