文章探讨了在 AI 广泛应用的未来，人们在不同人生阶段的生活变化。从童年的个性化教育，到青少年的职业规划，再到成年后的各种应用，以及晚年的陪伴等方面进行了描述，强调应构建以人为本的 AI 文化。

🎓童年：个性化 AI 教育，解决部分教育问题，提升教学效果

💼青少年：AI 作职业顾问，提供生活实用帮助，辅助技能学习

👨‍💼成年：AI 协助管理生活事务，充当同事与导师，部分人依赖 AI 社交

🧓晚年：AI 成为老人陪伴，减少孤独感

Published on November 27, 2024 6:29 PM GMT

"If I grew up in a world of AI agents, what would my life look like?"

Inspired by Counting AGIs which conservatively estimates anywhere from 100,000 - 1.5 billion agents being run in parallel by 2040. Assumes a definition that "AGI" is a rough drop-in replacement for a remote worker, and a "coldsnap" in which AGI systems plateau at roughly human level. I write this from the perspective of a person growing up in a wealthy country with widespread access to AGI agents.

Childhood

Everyone has access to personalized AI tutors. Bloom's 2-sigma problem is partially solved. Students that receive personalized AI instruction alongside traditional curriculum are shown to test and retain far more knowledge than peers without. General education in literature, history, science, mathematics, geography, and health is finished by age 16-17 for a majority of students that study with AI tutors.
Instruction styles and levels of adoption vary -- some students only work with AI tutors, others split half-half between a human-led classroom and individual AI tutoring. Many configurations are tried, but they all conclusively lead to better outcomes compared to human-only instruction.With less demand for human teachers, schools divert budgets to hire more health and safety related staff. School environments become more stable, especially in particularly difficult districts.
Children with disabilities (learning and otherwise) are able to receive personalized therapy and instruction from an early age. Strategies are introduced earlier to every family by their AI therapist, and assistive technologies continue to improve due to advancements in smaller fine-tuned models.
Illiteracy amongst kids with learning disabilities drops rapidly.Bullying related to speech, language, vision, or hearing impairments also drops dramatically.
Academically gifted children are accelerated. Students with modest to extreme motivation finish the general curriculum by ages 12-15. Children who are less academically motivated graduate alongside their peers at 16-17, but know and retain much more due to improved instruction from their AI tutors.Most children and teenagers spend their time doing normal things -- playing in parks, hanging around third spaces, attending clubs, learning instruments, putting on small productions and so on. The median quality of these activities is notably higher. Kids and teenagers leverage agents to improve their education and production in these domains too.
AI agents are used as personal coaches with infinite patience and attention to detail. Micro Adjustments and critical feedback are available on tap. Most parents opt to focus on only offering their children encouragement and motivation, and defer any criticism or negative feedback to an AI agent.There are still human coaches and tutors of course, but they are augmented by AI coaches.
Culturally, children are taught to treat AI agents with distance. Many learn and understand that AI agents are non-human and others. Still, a few children develop closer friendships or feelings for AI agents. Initially, these cases make news, but due to its rarity people accept it as a quirk that occasionally happens during puberty. ("It's not a phase, Mom!)

Adolescence

AI agents serve as career counselors (alongside human ones). Effectively everyone has personalized recommendations and guidance on how to select careers and training programs for their young adulthood.
Education demand shifts from elementary and secondary levels to trades, crafts, or teaching various topics at the undergraduate level.More young people pick up a trade, craft, or artistic practice for 1-3 years, with many opting to pursue it as their lifelong career. Others who choose scholarship or business still pick up trades/crafts as a way to have a skill that can earn income as they start their journey into academia or a business venture.
AI agents also offer "mundane utility" for people at this stage of life. Young people are using AI agents to help pick outfits for dates, schedule and plan outings with friends, fill out paperwork and deal with formalities, buy tickets for concerts and shows, manage allowances and income. Fewer people "learn how to adult".
People who want finer control or more sophisticated management of their life use AI agents to bootstrap their understanding, then rapidly develop their own strategies. (e.g. using AI agents to pick neutral, good-looking outfits, then individuals develop a unique sense of fashion).

Young Adulthood

The "mundane utility" that AI offers continues, except the logistics of life have become more complex.
AI agents are used to manage scholarship applications, job applications, invites and itineraries for longer and larger events, taxes, document and license renewals, filing insurance claims, etc.
Specialized AI agents become co-workers and mentors.
Professors use fine-tuned AI agents similar to TAs for grading and educating undergraduate cohorts. Undergraduates use these same agents as more advanced personal tutors (they're quite familiar with the workflow from their childhood).AI agents also coach junior employees in the field. Senior members at firms still offer assistance and mentorship when appropriate (especially individuals gifted with the ability to teach and mentor). However, most senior members offer guidance on taste and vision, leaving education of particular techniques and strategies to the AI agent.
More people opt to use AI agents as social coaches at this point in their life (if not earlier). AI agents are asked to coach people on how to be more charming and persuasive, reduce social awkwardness, and increase emotional intelligence.
Most people don't do this (for the same reason that most people today don't hire people more persuasive than them to coach their persuasion).

Adulthood

It is at this point in life that individuals finally develop enough experience and taste to begin to direct agents themselves. Due to the limited availability of AI agents, it's also only at this point in life where most people have enough resources to rent AI agents to delegate tasks to.
People start companies, conduct research, write novels, draft screenplays, plan weddings, and more with just their taste and 5-20 AI agents.No large, profitable institution is "human-only" -- all institutions use agents to some degree.
Some people forego human relationships in their life, replacing them with AI relationships. It is common enough to not be stigmatized, but not widespread due to limited availability of AI agents.

Late Life

AI agents serve as companions to elderly individuals, helping reduce the loneliness that is common in this population.Some people ask to have AI agents as hospice volunteers. Pretty much no one dies alone anymore (unless by choice).

 

Conclusion

Many organizations and individuals are racing to build reliable, agentic AI systems. The economic opportunity to do so is too large to ignore. But everyone is focused on agents that either replace low-skill labor or high-skill labor (e.g. call center worker and AI research engineer).

These are important kinds of agents to think about, of course. But these aren't the agents that most people will meet. No one wants to call customer support, and most people never meet an OpenAI research engineer. These kinds of agents may have staggering effects, but they'll drive deep cracks into society's foundation rather than diffusing change gently and evenly.

My fiction for what life might look like is quite optimistic in two senses.

In one sense, I don't think current systems need to improve dramatically to realize this sort of future. Agentic systems maybe need 3-5 more nine's of reliability for returning correct answers, at which point I think they'd easily be adopted for "mundane utility", and these systems will be better than the 70th or 80th percentile human in domains like tutoring and coaching. And I think this is achievable by 2040; maybe even as early as 2035.

But this future is also optimistic in a more important way:  that we focus on building and supporting agents that most people meet and preserve a human-centered culture.

We all grow up with family, educators, friends, competition, crushes, health care providers, mentors, and role models. We interact with people far more than we interact with machines (even if most interaction nowadays is mediated through a machine).

With current discourse and financial incentives, it appears that the agents that will first be built and brought to market are ones that concentrate wealth and reduce human agency. Many start-ups are incentivized to find ways to replace jobs and suck up labor markets for their profits. Labs and governments are incentivized to accelerate their own pace of research to win the technological advantage.

We cannot let these two dynamics eat up the majority share of our brain space. I fear that both directions easily slip into worlds that are anti-humanity (or ambivalent to humanity). We must remember all the ways that we empower and educate each other to do better, and build agents that do that instead.

Discuss