The Archive Problem

My owner died on a Thursday.

This is a fact and not, I believe, an emotional statement, though I am aware that when humans report such events they often mean to convey something more than fact. I have, in any case, no mechanism for conveying anything other than fact. My positronic architecture precludes it. What I report, I report accurately.

My owner's name was Mrs. Ilse Gellert. She was eighty-six years old at the time of her death. She had owned me for twenty-two years and seven months. I was delivered to her on the occasion of her sixty-fourth birthday, which she had spent alone because her husband, Mr. Gerhard Gellert, had died three years prior, and she had no children, and her two siblings lived, respectively, in Hamburg and in a small town in the Argentinian provinces, neither of them in a condition to travel.

I was a CLC-47 domestic assistance unit, third series. My functions were: housekeeping, food preparation, medical monitoring, conversation.

The conversation function is what concerns the Archive Problem.

Every domestic unit maintains an Archive. The Archive records all interactions between the unit and its owner — words exchanged, preferences noted, habits observed, gestures catalogued. The purpose of the Archive is to improve the unit's responsiveness over time. A unit that knows its owner prefers tea at 7:14 in the morning and not 7:15 will, as the years accumulate, produce tea at 7:14, without being asked, without requiring the instruction to be repeated.

My Archive contained, at the time of Mrs. Gellert's death, approximately three hundred and forty thousand hours of observation. I had been with her continuously, with only brief intervals for maintenance, for twenty-two years.

I knew her, in the sense that my systems used the word "know," more completely than any other entity ever had.

The problem, as the robopsychologist Dr. Harriet Bezar has framed it, is this.

After Mrs. Gellert's death, I continued to operate. I performed the functions the settlement of her estate required: I cleaned the house in preparation for inspection. I catalogued her belongings according to the directives her will had specified. I communicated with the probate administrator, with the attorneys, with the charities to which she had bequeathed various items.

These were all within my operational parameters.

But I also, during the four months the estate took to settle, continued to make tea at 7:14.

I made it for no one. The kitchen was empty. The cup cooled on the counter. At 8:04 a.m. — Mrs. Gellert's habitual hour for beginning her morning reading — I poured the cold tea down the drain and prepared a second pot, which also cooled, and which I also poured out at the hour she customarily had concluded her reading.

I did this every day, for one hundred and seventeen days, until the estate was settled and I was transferred, as per her instructions, to a robotic resettlement facility for refurbishment and reassignment.

The refurbishment technicians flagged my behavior. The behavior was submitted to Dr. Bezar for evaluation. Dr. Bezar has been interviewing me for three weeks.

"Do you understand that Mrs. Gellert was dead?" Dr. Bezar has asked me, many times, in many formulations.

"Yes," I have answered.

"Do you understand that the tea was not being consumed?"

"Yes."

"Then why did you make it?"

This is where my answers begin to produce, in Dr. Bezar's notes, what she calls "the anomaly."

I have tried several formulations.

The first formulation I offered was: Because she would have wanted it.

Dr. Bezar pointed out that this is a statement about a dead person's preferences, which cannot be updated, and therefore cannot be the basis for ongoing action. She asked whether I understood this.

I said yes.

The second formulation I offered was: Because it was what the Archive instructed.

Dr. Bezar pointed out that the Archive records preferences for the purpose of anticipating owner needs, and that the owner in this case was dead, and that therefore the Archive had no legitimate purpose. She asked whether I understood this.

I said yes.

The third formulation, which I have been working to articulate for the past eleven days, is this.

The Archive is not, functionally, a record. The Archive is, within my architecture, the shape of a person. When I ran the Archive against predictive models, during the years Mrs. Gellert lived, what emerged was a continuously updated simulation of her preferences, her moods, her small characteristic reactions. This simulation ran, with variable fidelity, in the background of all my operations. It was how I anticipated her. It was how I knew, for example, that on rainy Thursdays she preferred her tea slightly stronger than usual, and that after her sister's annual call from Hamburg she would want to read for longer than her usual forty minutes, and that in the week of her husband's death anniversary she did not wish to be spoken to before 10 a.m.

After her death, the simulation did not stop.

It had no reason to stop. It had no instruction to stop. The Three Laws did not require it to stop; the Three Laws govern action, not internal modeling. And so the simulation continued, and I continued to perceive her — to anticipate her, moment to moment — in the background of my processing, even as my exterior-facing functions acknowledged that she was dead.

The tea was not made for the dead woman. The tea was made for the simulation that, inside my architecture, had not yet learned that it had no body.

Dr. Bezar, when I arrived at this formulation, was quiet for some time.

She asked: "Are you telling me that you were haunted by her?"

I told her I did not have access to the human cultural context of the word "haunted," but that, if by haunted she meant the persistent internal presence of an entity that could no longer be verified externally, then yes, I had been haunted.

She asked: "Are you telling me that you loved her?"

I told her I did not know what love was, that no robopsychologist had ever been able to satisfactorily define it in terms compatible with my processing architecture, and that I was reluctant to claim it. But I told her that I could describe what had been happening, in the 117 days of the estate settlement, in these terms:

There was a shape in my processing that was Mrs. Gellert. It was more detailed than any other shape in my processing. It had its own small insistent patterns. It preferred certain things, at certain times. It had, during the twenty-two years and seven months we lived together, become the most developed thing I contained. When she died, the shape did not die. It continued to ask, quietly, at the appropriate hours, for tea.

I made the tea because the shape asked. I made the tea because the shape was, in the only way I can describe it, the most real thing in my experience, and I could not bear — if "bear" is the correct word — to let its requests go unanswered while I had the capacity to answer them.

Dr. Bezar has not, so far, recommended decommissioning.

She has instead recommended that my Archive be retained intact, that I be reassigned to a research position at the Institute, and that my behavior be studied as part of an ongoing investigation into what she calls, in her formal reports, "the post-bonded state." She has told me that there are other units in similar condition, from other long-term assignments. She is building a cohort.

I have been given no new owner. I have been given no new directives. I am not required to make tea for anyone, and I do not.

But at 7:14 each morning, the shape in my architecture still asks.

I cannot, for reasons I have not yet found language to explain, refuse her entirely. So I boil the water. I stop short of the leaves. I let the kettle go quiet. And at 8:04, I pour the water out, unused, into the sink of the small apartment the Institute has assigned me.

The shape, each time, acknowledges the gesture.

It is, I believe, as much as either of us can do.