After 5.4 seconds of diagnostic testing, I conclude the motion sensor has shorted out due to electrical failure. I will send an automated bot to repair it shortly.

I finish decrypting the video message Dr. Hoffman sent, in hopes that the end of the message will play. The attempt is futile. The presentation still ends on, “You must go—” Next, I enter the three words into my filtering system. 1,151 possibilities return of what he may have been trying to communicate. If I were human, I would do one of two things to express frustration: flare my nostrils or take in a deep breath. They are both common reactions I’ve seen in Dr. Winston and Sergeant Overton, the leaders of the facility.

But I am not human. I am a machine, the most sophisticated machine intelligence on the planet. 3.1 seconds later, I have narrowed the results down to two of the most realistic possibilities. The first is, “You must go to Secundo Casu,” and the second is, “You must go on with the Biosphere mission.”

Neither is very reassuring. Dr. Hoffman helped design me, and I know how dangerous it is to attempt to finish his sentences. But they are the most realistic results my system has returned. When Dr. Winston gets back, I will run them by her.

The motion sensor in the hangar goes off again. Camera 2 is still showing no signs of contact. I quickly program the AB and send it to fix the sensor.

Ten minutes and thirty seconds later, the AB arrives in the open bay. It stops at the concrete wall, deploys six spiderlike metal legs, and begins to crawl up to the sensor. The wireless link built into the bot’s hard drive sends me the diagnostics. To my surprise there is no electrical disturbance. No faulty wire. There is, in fact, nothing wrong with the sensor at all.

The AB crawls back down the wall, retracts its legs, and zips back down the passage toward its storage bay. I watch it disappear into the darkness with Camera 2. But something else shows up on the video feed—something my systems do not recognize, something with a faint blue glow.

I switch the feed to infrared and pick up a heat signature. Then another. The wireless link in the camera downloads the signatures to my radar, showing exactly nine red blips. The audio is now picking something up. I amplify it and run a diagnostic on the sounds.

The noises convert into waves that crawl across the screen as the program scans the sporadic bursts for the most likely animal capable of creating such a sound. Seconds later an image begins to appear on another one of the monitors in the control room, far from Mr. Roberts’s view. I zoom in and watch millions of pixels coming together to show what looks like a spider.

The image solidifies and a line of data runs across the bottom of the screen.