We Could Not Optimize a Marriage
I was 41 when I asked HOS whether I should divorce my wife. By then, artificial intelligence could analyze almost anything. It could reconstruct years of financial decisions. It could identify patterns in health data before doctors noticed symptoms. It could analyze conversations, detect changes in emotional patterns and model possible consequences of major life decisions. So I assumed it could analyze my marriage. Laura and I had been married for 15 years. We had two children, ages 12 and 9. We weren't constantly fighting. That almost made it harder. We had simply become distant. I traveled frequently. Laura had built a demanding career of her own. Our home functioned remarkably well. Our marriage didn't.
So I asked: “Based on everything you know, should I stay married?” HOS wouldn't answer. I was disappointed. What was the point of having access to all that intelligence if it couldn't answer the question that mattered?
It could tell me almost everything except what I should choose
HOS could help me understand patterns. When had the distance begun? What had changed? What assumptions was I making about Laura? What was I avoiding? What were the likely financial and practical consequences of divorce? How might different choices affect our children? What did research suggest about conflict, communication and family transitions? But there was something it could not know. What did our marriage mean to Laura? For that, I had to ask Laura. That conversation was terrible. And necessary. She wasn't experiencing the marriage the way I thought she was. Some things were worse. Some were better. She was angry about things I barely remembered. I was hurt by things she hadn't realized mattered.
For years we had each been living inside our own explanation of the same marriage. AI could help illuminate those explanations. It couldn't reconcile them for us.
Then I made things worse
I became obsessed with understanding our problems. Every disagreement became data. I analyzed conversations afterward. I started recognizing patterns and pointing them out to Laura. She finally exploded. “I don't want to be analyzed. I want to be married to you.” That stopped me. I had turned understanding into another form of control. HOS helped me see the problem. Laura had not agreed to become an optimization project. Her privacy mattered. Her interpretation mattered. Her agency mattered. I stopped trying to use technology to prove what was happening between us. We found a human therapist.
We stayed together
For six more years. People sometimes hear that and assume therapy failed. It didn't. Those six years contained some of the best years of our marriage. We became closer. We became better parents. We learned how much of our conflict came from things neither of us had understood. Then something changed again. Laura wanted a different life. Eventually I realized I did too. This time we were able to say it honestly. We divorced. I had once thought HOS should help me save my marriage. Then I thought it should help me decide whether to leave. Eventually I understood that neither was its job.
Its job was to help me become more capable of understanding what was happening, recognizing consequences, respecting the other people involved and making my own decisions.
AI changed relationships around us
By the time our children were adults, relationships had changed in ways Laura and I never imagined. People routinely used AI companions. Dating systems could model compatibility from enormous amounts of personal information. Some people allowed relationship AIs to mediate disagreements in real time. Virtual experiences made long-distance relationships feel physically close. People could create synthetic companions tailored almost perfectly to their preferences. That created extraordinary possibilities. It also created a strange temptation. Human beings are inconvenient. They misunderstand us. They disagree. They change. They want things we don't want. A synthetic companion could increasingly be designed not to. My son once asked me whether that made artificial relationships better.
I told him I didn't know. But I knew something from my marriage to his mother. Some of the most important growth in my life came from loving someone whose agency I could not control.
Laura and I became something else
Years after the divorce, Laura remarried. I didn't. We remained connected through our children and eventually grandchildren. One Christmas, we sat together watching everyone open presents. She laughed and said: “We didn't completely screw it up.” I laughed too. “No. Just parts of it.” Our marriage had ended. Our family hadn't. Neither of us would have chosen everything that happened. But we had become better at responding to what did.
Looking back
At 41, I wanted intelligence powerful enough to tell me whether to stay married. I'm grateful I didn't get it. There are decisions where information can become extraordinarily valuable while the decision itself must remain human. Love is one. Commitment is another. Leaving can be another. HOS helped me see things I couldn't see. Sometimes it challenged the story I was telling myself. Sometimes it suggested expertise beyond itself. Sometimes it refused to help me use intelligence to control another person. And sometimes it simply helped me sit with uncertainty long enough to understand what I was actually choosing. My marriage to Laura did not last forever. That doesn't make it a failed marriage.
We loved each other. We raised two children. We hurt each other. We learned. Eventually we released each other. Technology became astonishingly good at predicting human behavior during my lifetime. But Laura was never a prediction. She was a person. So was I. HOS helped me understand that respecting another person's agency sometimes means accepting an outcome I would not have chosen. That lesson outlasted our marriage. I'm glad it did.
By this point in the imagined future, AI can know extraordinary amounts about human behavior and relationships. That makes the limits HOS places on itself increasingly consequential. The husband wants an answer to a deeply human question: should he remain married? HOS can help him understand patterns, examine assumptions, consider consequences, bring relevant knowledge into the conversation and recognize things he may not see himself. What it cannot legitimately do is convert all of that intelligence into authority over the decision—or treat his wife as an object to analyze and influence. When he begins using technology to prove his interpretation of their marriage, greater understanding starts becoming greater control. HOS helps expose that boundary and, importantly, recognizes when a human therapist can contribute something technology should not try to replace. The eventual divorce is not presented as either success or failure. The success is that two people become more capable of understanding, choosing, accepting consequences and respecting each other's agency—even when their choices ultimately diverge. An AI might someday become remarkably good at predicting whether a marriage will survive. HOS asks a different question: how can that intelligence help two human beings make their own choices without quietly taking ownership of those choices away from them?