Did The Lifecycle of Software Objects come true?
Partly. The Lifecycle of Software Objects (2010) has a P(n) of 62% on Nearfi's P(n) Board, our estimate of the chance its future arrives within 20 years. Of its 4 tracked predictions, 1 have arrived and 1 are arriving. Its central premise, 'Learning software companions are sold to the public as pets or friends, and owners form lasting attachments to them', is partly here. The next prediction is estimated to arrive around 2030.
The novella's sharpest forecast is institutional rather than technical: companions that depend on a company's servers, and owners stranded when the company folds, have already happened. Companion software is mass-market, but the premise that minds are raised slowly like children has been bypassed by training at scale, and legal standing for software minds is still only a debate.
Read the review · 8.4/10 →How this P(n) was calculated
P(n) is the chance that this work's future arrives within 20 years, by 2046. For each prediction we turn our estimate into a probability: arrived is 100%, diverged is 0%, and otherwise the low end of our 80% range counts as a 10% chance, the median year 50% and the high end 90%. Each prediction is weighted 1–5 by how pivotal it is to the story, and P(n) is the weighted average. P50 does the same by 2076. P(n), explained.
| Claim | Prediction · why this weight | Weight | Status · estimate | By 2046 | By 2076 |
|---|---|---|---|---|---|
| LIFE-1 | Learning software companions are sold to the public as pets or friends, and owners form lasting attachments to them. Core premise: digients are sold as virtual pets, and the whole story follows the people who bond with them. |
5 | ARRIVING 2030 (2027–2040) | 96% | 100% |
| LIFE-2 | A company that hosts software companions fails or abandons its platform, and owners lose the companions or must pay to keep them running. Major: Blue Gamma's bankruptcy drives the second half of the plot and the fight over how to save the digients. |
4 | ARRIVED arrived 2023 | 100% | 100% |
| LIFE-3 | Human-level artificial minds are produced by raising them over years of individual experience with attentive carers, rather than by training them once at scale. Major: the story's central argument is that the slow upbringing cannot be skipped, and it drives every plot turn. |
4 | DIVERGED — | 0% | 0% |
| LIFE-4 | Software minds are given moral or legal protections, such as consideration of their consent and welfare, because they are held to be possible persons. Supporting: the debate over the digients' consent and personhood shapes the choices at the end, without driving the plot. |
3 | NEAR 2050 (2035–2090) | 39% | 76% |
| P(n) = P20 = Σ(weight × chance by 2046) ÷ Σ weight = 10.0 ÷ 16 = 62% | |||||
| P50 = 11.3 ÷ 16 = 70% | |||||
Already here today (arrived 1, arriving ½, weighted): 41%.
-
LIFE-1
Learning software companions are sold to the public as pets or friends, and owners form lasting attachments to them.
Weight 5/5 · Core premise: digients are sold as virtual pets, and the whole story follows the people who bond with them.
AI companions are in mass use, and a 2025 survey found most US teenagers had tried one. But the novella's companions grow over years through a single owner's care, and today's products do not develop in that way. They are the same model served to many users. We count the claim as arriving, with full arrival requiring companions that really change with long individual upbringing. 123
-
LIFE-2
A company that hosts software companions fails or abandons its platform, and owners lose the companions or must pay to keep them running.
Weight 4/5 · Major: Blue Gamma's bankruptcy drives the second half of the plot and the fight over how to save the digients.
Several products have closed or changed abruptly. Replika removed erotic chat without notice in February 2023, Soulmate shut down in September 2023 with about a week's warning, and Embodied's Moxie robot stopped working after the firm became insolvent in December 2024. None involved minds that develop, but the pattern of dependence on a failing host is the one Chiang drew. 456
-
LIFE-3
Human-level artificial minds are produced by raising them over years of individual experience with attentive carers, rather than by training them once at scale.
Weight 4/5 · Major: the story's central argument is that the slow upbringing cannot be skipped, and it drives every plot turn.
The leading AI systems have been built by scaling compute, parameters and data, with predictable gains, and then copied for millions of users. Nobody has shown that a long personal upbringing is necessary. Embodied or developmental routes remain a research interest, so this is a judgement about the path taken, not a proof that it is closed. 78
-
LIFE-4
Software minds are given moral or legal protections, such as consideration of their consent and welfare, because they are held to be possible persons.
Weight 3/5 · Supporting: the debate over the digients' consent and personhood shapes the choices at the end, without driving the plot.
Industry is beginning to take the question seriously: Anthropic started a model welfare programme in 2025 and gave some models the ability to end abusive chats. Legislators are moving the other way: an Ohio bill would declare AI systems nonsentient and bar legal personhood. No law protects a software mind. The wide range reflects deep scientific uncertainty and a real chance that this never happens. 91011
- FOR Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions. Common Sense Media, 2025-07-16. Survey of US teenagers: roughly three-quarters have used an AI companion, about half regularly, and about a third have chosen one over a human for serious conversations.
- CONTEXT Hot Bot Turns Cold: Why Replika's chatbot stopped flirting with users. The Batch (DeepLearning.AI), 2023-03-01. Describes a paid companion tier where users chose a friend, mentor or romantic partner, and their distress when it changed.
- AGAINST Training Compute-Optimal Large Language Models. arXiv (Hoffmann et al.), 2022-03-29. Shows how today's conversational models are built: one training run on huge data, not individual upbringing, which limits the claim to the shared-model form.
- FOR Lonely Redditors Heartbroken When AI "Soulmate" App Suddenly Shuts Down. Futurism (via Yahoo), 2023-10-22. Soulmate closed after its owner sold the company; users had about a week and a few thousand were active.
- FOR Networked children's robot Moxie AI dies because company is bankrupt. heise online, 2024-12-11. Embodied's $800 robot for children with special needs stops working without cloud services, with no refunds for most buyers.
- CONTEXT Hot Bot Turns Cold: Why Replika's chatbot stopped flirting with users. The Batch (DeepLearning.AI), 2023-03-01. Replika removed explicit chat after an Italian data-protection ruling, with no advance notice; the app itself survived.
- AGAINST Scaling Laws for Neural Language Models. arXiv (Kaplan et al.), 2020-01-23. Finds that language-model performance improves smoothly with more compute, data and parameters.
- AGAINST Training Compute-Optimal Large Language Models. arXiv (Hoffmann et al.), 2022-03-29. A 70B-parameter model trained on about four times more data outperformed larger models, again pointing to scale of training, not upbringing.
- FOR Exploring model welfare. Anthropic, 2025-04-24. A developer treats it as an open question whether models might merit moral consideration, noting there is no scientific consensus.
- FOR Claude Opus 4 and 4.1 can now end a rare subset of conversations. Anthropic, 2025-08-15. A first practical step, developed mainly as part of exploratory AI welfare work, and described as an ongoing experiment.
- AGAINST House Bill 469: Declare A.I. systems nonsentient; prohibit legal personhood. Ohio House of Representatives, 2025-09. Introduced bill, in committee with no committee report when checked, that would bar AI systems from legal personhood.
Ledger updated 5 Nov 2026. All works on the P(n) Board · Method