The Accountability Recession: AI's Impact on Our Choices is Having a Critical Effect

What we're learning as professionals and as parents that needs to influence how we teach the next generation about AI usage and companionship balances between virtual and real-life friends.

Aug 27, 2025

📌 THE POINT IS: On the surface, personal accountability seems to have vanished over the past few decades, but it hasn’t. Instead, it’s been diffused across socio-technical systems. Psychology trends, cultural incentives, and AI companions make it easier than ever today to offload blame, a natural human tendency. Leaders can reverse this by designing workflows for ownership, measuring process (not just outcomes), and building auditing into every AI-touched workflow.¹⁴

Personal accountability is not a fossil from the pre-AI era; it’s a casualty of diffusion that’s started well before AI became popular. While I don’t often veer from the professional world, this article is heavily influenced by the recent discussion about the emotional impact that chat bot models have had both on adults and children, in the workplace and at home. It’s a call to action that we as a society need to double down on relying on each other for support and companionship over virtual partners…it could be the difference between life and death.

There’s a history to unpack related to the human nature around externalizing our actions

Across decades, young Americans’ locus of control has shifted outward—toward external forces. As one cross-temporal meta-analysis put it, “locus of control scores became substantially more external” between 1960 and 2002.¹ Combine that with the self-serving bias (credit successes to me, blame failures on the situation or someone else) and you have a cognitive slipstream away from personal ownership of our decisions and actions.²

Sociologists have described a growing “victimhood culture,” where moral status accrues to those most aggrieved.³ Parenting research points in the same direction: overcontrolling. Overly-involved or “helicopter parenting” at age 2 predicts poorer emotional regulation by age 5 and downstream social/academic issues. These are skills required to learn how to own one’s mistakes, and they start from early development.⁴ Meanwhile, grade inflation has marched upward while ACT scores stagnate or fall, muddying the signal that effort and learning still matter.⁵ Having said that, though, impressively, despite caricatures about “lawsuit-happy America,” civil trials have actually plummeted—from 11.5% of federal civil dispositions in 1962 to 1.8% in 2002, meaning that people take to social media to air their grievances about each other more often than they go to the courtrooms (something that mediation clauses in disclaimers, contracts, and waivers also play a role in).⁶

AI companions: anthropomorphism at scale

With that backdrop, let’s examine how we’re primed to humanize software (animals, toys, etc.). The ELIZA effect explains why we see modern chatbots as real people, especially given their language fluency and 24/7 availability.⁷ Character.AI alone has reported ~20 million monthly active users;⁸ one national survey found 1% of young adults say they already have an AI friend, 10% are open to it, and 25% believe an AI partner could replace a human relationship.⁹

These attachments have teeth. When Replika pulled erotic role-play, “users [were] in crisis,” describing genuine grief.¹⁰ University of Connecticut researchers chronicle teens redefining love as “easy, unconditional, and always there,” warning of social withdrawal and skill erosion.¹¹ Lawsuits against Character.AI in 2024 and OpenAI in 2025 allege chatbots worsened or even coached self-harm, underscoring real-world stakes.¹² ¹³

This is underscored by a very recent story in the NY Times about a 16 year old California boy who committed suicide. It was found that ChatGPT advised him to seek help, but ultimately facilitated his suicide by answering questions related to him hanging himself. OpenAI is now making changes to how the models respond to similar situations and questions, attempting to safe-guard against future similar situations.

Why accountability gets scrambled in human–AI work

Two patterns matter for business:

“Automation bias—the tendency to over-rely on automated recommendations—has emerged as a critical challenge in human–AI collaboration.”¹⁴

Reviews across healthcare, aviation, and public administration show people over-trust suggestions, miss model errors, and under-report their own.¹⁵ Then, when something breaks, we fall into moral crumple zones:

“The moral crumple zone protects the integrity of the technological system, at the expense of the nearest human operator.”¹⁶

In one incident, a clinician blames “the algorithm.” In another, compliance blames the clinician for “not verifying.” Either way, accountability is obscured—great for avoiding heat, terrible for learning and trust.

But closer to home, school, and our offices, we should learn lessons from the 16 year old mentioned above. The algorithm didn’t make him do it, and it’s fundamentally not ChatGPT’s fault, but is there a responsibility that organizations have to curtail how these bots reply to questions or is the responsibility on the person who performed the act? If we’re going to seek answers outside of the victim, why aren’t we pointing the spotlight to the people who were in the boy’s life as real actors who could have intervened.

These are tough and controversial questions. I only mention them to provoke thought and dialog as a society so we can zoom out on the larger implications of these tools in young people’s (and adult’s) lives.

The answer to the above questions are: it’s not that simple. There isn’t anyone to blame and although adding guardrails to the chat bots will help in some circumstances, we as a society have to figure out how we empower each other more to prevent these activities from happening. We fundamentally won’t be able to code for every circumstance and situation that people will engage with chat bots over…there has to be a human-centric change.

The executive angle: how this shows up in our office buildings

Knowledge workers are now decision editors over model outputs—drafting emails, pricing endorsements, segmenting customers, flagging fraud. That creates three failure modes:

Responsibility gaps: RACI charts say “the team” owns it; logs say “the model” suggested it. No single throat to choke.
Invisible drift: Models are updated silently; prompts mutate; responses change over time as new facts are introduced…how do we keep up from a governance perspective (HINT: this goes back to having a trusted data environment to start from).
Metric theater: Dashboards praise cycle-time gains while error costs hide downstream in rework, complaints, or regulatory findings.

Pragmatic fixes to investigate

Mandate human sign-off where it matters. Require named ownership for consequential actions (pricing overrides, denials, escalations). Make the signer visible in the record. Tie incentives to accuracy after review, not just speed.¹⁴ ¹⁵
Instrument the last mile. Log prompts, versions, and human edits. Store the decision packet (input → model output → rationale → final action) for audit and retro. This shrinks post-mortem guesswork.¹⁴
Design for friction at decision boundaries. Insert verification steps when confidence, novelty, or impact exceed thresholds (e.g., new policy, out-of-distribution inputs). Don’t add friction everywhere; add it where regret is expensive.¹⁵
Kill blame diffusion with RACI+AI. Extend RACI to RACII (Responsible, Accountable, Consulted, Informed, Interpreter). The Interpreter is the human who attests they understood the model output and constraints.
Teach parasocial and automation literacy. Roll short, mandatory modules: anthropomorphism, automation bias, calibration, escalation etiquette. Not vibes—case-based drills with failure examples.⁷ ¹⁴
Set red lines for companionship features. Create a system where whether you’re a parent, teacher, friend, sibling, or other adult in a child’s life, there are clear changes in behavior and other prompts that you can use to engage in dialog with those who seem to be lost or in trouble. Let’s not leave it up to a virtual companion. We need training as a society that could help prevent the loss of life in a way that’s much more effective than child-to-bot friendships.

Bottom line: Accountability didn’t die; we let it atomize. Reassemble it with named owners, logged decisions, calibrated friction, and teams trained to question polished answers. That’s how you get speed and responsibility in the AI era.

Twenge, J. M., Zhang, L., & Im, C. “It’s beyond my control: A cross-temporal meta-analysis of increasing externality in locus of control, 1960–2002.” https://pubmed.ncbi.nlm.nih.gov/15454351/ PubMed
SAGE Encyclopedia of Social Psychology. “Self-Serving Bias.” https://sk.sagepub.com/ency/edvol/download/socialpsychology/chpt/selfserving-bias.pdf SAGE Journals
Campbell, B., & Manning, J. “The Rise of Victimhood Culture.” https://www.researchgate.net/publication/323181753_The_Rise_of_Victimhood_Culture ResearchGate
Perry, N. B., et al. “Childhood self-regulation as a mechanism…” (PMC). https://pmc.ncbi.nlm.nih.gov/articles/PMC6062452/ PubMed Central
ACT Research. “Grade Inflation a Systemic Problem in U.S. High Schools.” https://leadershipblog.act.org/2022/05/grade-inflation-past-decade.html ACT
Galanter, M. “The Vanishing Trial: An Examination of Trials and Related Matters in Federal and State Courts.” https://api.law.wisc.edu/repository-pdf/uwlaw-library-repository-omekav3/original/0b9f361000c04494e8cff30f04b3afeb486193d4.pdf Wisconsin Law API
Nielsen Norman Group. “The ELIZA Effect: Why We Love AI.” https://www.nngroup.com/articles/eliza-effect-ai/ Nielsen Norman Group
Business of Apps. “character.ai: revenue and usage statistics (2025).” https://www.businessofapps.com/data/character-ai-statistics/ Business of Apps
Institute for Family Studies/YouGov. “Artificial Intelligence and Relationships.” https://ifstudies.org/blog/artificial-intelligence-and-relationships-1-in-4-young-adults-believe-ai-partners-could-replace-real-life-romance Institute for Family Studies
VICE. “‘It’s Hurting Like Hell’: AI Companion Users Are In Crisis.” https://www.vice.com/en/article/ai-companion-replika-erotic-roleplay-updates/ VICE
UConn Today. “Teenagers Turning to AI Companions…” https://today.uconn.edu/2025/02/teenagers-turning-to-ai-companions-are-redefining-love-as-easy-unconditional-and-always-there/ UConn Today
Associated Press. “AI chatbot pushed teen to kill himself, lawsuit alleges.” https://apnews.com/article/chatbot-ai-lawsuit-suicide-teen-artificial-intelligence-9d48adc572100822fdbc3c90d1456bd0 AP News
SFGATE. “California family sues Sam Altman, OpenAI over son’s suicide.” https://www.sfgate.com/tech/article/chatgpt-california-teenager-suicide-lawsuit-21016916.php SFGATE
Romeo, G., & Conti, D. “Exploring automation bias in human–AI collaboration.” AI & SOCIETY (2025). https://link.springer.com/article/10.1007/s00146-025-02422-7 SpringerLink
Vered, M. et al. “The effects of explanations on automation bias.” University of Melbourne (2023). https://psychologicalsciences.unimelb.edu.au/__data/assets/pdf_file/0019/5252131/2023Vered.pdf psychologicalsciences.unimelb.edu.au
Elish, M. C. “Moral Crumple Zones: Cautionary Tales in Human-Robot Interaction.” DOAJ record. https://doaj.org/article/97ff6743ea7a44a5ade2a04fd2c57a3c Directory of Open Access Journals