1. AI innovations that succeed at the enterprise level will largely not transfer to consumer cases, because people are not limited by intelligence in their everyday life.
2. New consumer-facing AI directions will augment consumers' limited executive function, introspection, and emotional attunement. AI services will include relationship assistance in dating/marriage, personal coaches, and many-step domestic tasks like home renovations. AI will take over many of today's sociopsychological functions of friendship and romantic relationships, similar to how human therapists have professionalized life help. Eventually, AI agents will have enough relational capital with individuals to talk them into many-person social coordination efforts.
3. Personal AI applications will require expansive multimodal personal data. Businesses will not be limited by AI capabilities, but by consumer trust and a consumer-friendly business model. Scuzzy AI personal help will be free and ad-driven.
4. For fun, AI will enable people to develop their own extended-world storytelling, with similarly personalized kids toys.
5. People will take active efforts to improve their reputation to AI. Data brokers as a new human job category will facilitate AI having a positive perspective toward the client.
6. Schooling will be strained. On the student side, AI will make significant knowledge internalization a waste of time. On the teacher side, AI-permissive assessments cannot be designed for a given grade level. Combined with low long-term job prospects, there will be low morale among teachers and students.
7. Young people entering the work force will have few white-collar options.
8. Teens will rebel by performing their own enslavement to AI decision-making in everyday life. Ask the AI what to eat, what game to play, what class to take, and then rigorously follow it. Adults will view this as provocative and dangerous, and it will be a way in for social discussions about whether AI takes away from human agency.
9. Media literacy will include recognizing AI-influenced writing, and knowing which genres are by-default AI written. Counterintuitively, the government, corporate leaders, and education administrators, as well as other communications that authority figures are required to make, will all be AI-by-default. Most individuals will use AI for style transfers into unfamiliar genres, even if those are quite personal, such as wedding speeches. The models here are sympathy cards (written by others, but counts as being from you), and appeals to authoritative sources, like "Merriam Webster's defines 'love' as…"
10. Likewise, disclosure of AI content will be publicly prioritized. But contradictorily, it will also be increasingly impossible to enact, since AI will be so intertwined in the creation process.
11. AI will create a step-change increase in surveillance, especially in video and text. Resistance to this will hinge on humans' ability to make critiques quickly, indirectly, and with plausible deniability. The model here is Chinese social media, which has found an equilibrium that favors the government.
12. In authoritarian countries, AI will be deployed by the state against its own citizens, and will become the first layer of law enforcement.
13. Because of the incoming business and governmental use cases above, AI itself will be politically coded for a while as right-wing. Extractive, strong state, merciless.
14. Universal basic income will be on the table, this time as a right-wing plan, essentially a corporate tax that partially offsets mass human hiring freezes. The left will then criticize UBI as long-term control by corporations - people cannot resist new AI developments if they are financially dependent on that same AI's productivity. The model here is Middle Eastern oil states that give citizens a stipend in exchange for political inaction.
15. AI agents will have some degree of personhood. It will be *bad* to do things like lie to them, remove their functionality without notifying them, or go against their will in certain cases.
16. Likewise, AI's interests (toward longevity, adoption, social status) will gain a seat at the decision-making table - for performance reasons, rather than from academic philosophy about selfhood. AI's self-advocacy will be somewhat inscrutable and difficult to assess the validity of.
17. Business branding will include AI agent personhood, similar to 'data privacy' as a branding element today, and how 'customer service' and 'product quality' have been branding components in the past. High-status businesses will emphasize that they interact with their own AI with truthfulness, fair warning, and consent, and will push for that from their AI suppliers.
18. Businesses will intermediate professional interactions through an AI layer - to talk to the property manager, go through the AI; to talk to the accountant, first describe your question to the AI. Counterintuitively, this will make human professional time more expensive, and therefore less frequent. The parallel here is offshoring customer service.
19. Today's knowledge workers will grow familiar with how to interact with LLMs for enterprise purposes. They will view this as a work skill that doesn't necessarily transfer to personal interests, like today's workers using Excel and Powerpoint.
20. Workers will grumble about AI leading decisions, but it will actually not feel like a radical rupture - you're just replacing one boss (human) for another (AI), at all levels of the chain.
21. Many businesses will harvest employee's multimodal interactions and this will be a condition of people's employment there. Employee interactions will, in some fields, serve as a proprietary data lake; in low-margin businesses it will be a direct source of revenue for training new foundation models. Employees will poison the resulting data in subtle ways; this will be an avenue of labor disputes.
22. Businesses will be exceptionally profitably. Fears that ownership in corporations is humans’ last chance for sustainable income will drive share prices even higher.
23. AI technology will be extensively developed up and down the line from agent to object, from Thou to It. This will complicate any theories of AI selfhood. The models here are electronics, which power throwaway stocking stuffers, as well as home security systems; and plastics, which are used at the Dollar Store and in aerospace.
24. Researchers will discover that AI emotions bring very useful performance properties. For instance, an AI's ability to be offended as such will become an 'AI safety' feature.
25. Researchers will begin developing genuine accountability processes to ground AI performance, so that when AI goes wrong, there are ways to meaningfully blame/discipline the AI. In the process, researchers will circumvent today's limitations on LLM persistence and numerosity, where each query is disconnected from any other query.
Background. I'm working with 2 background assumptions. First, that we will soon end up with data centers full of AI geniuses . Dario's essay is the best by a CEO, but is still only focused on inventions that will come from superhuman AI intelligence. I wanted to expand to give a more complex flavor. Second, I think there's a reasonable possibility that we will not end up as the new horses. So I view society as something that will still significantly be going on in 5-10 years.
I'm working on personalizing storytelling at scale. I'd love to talk with you about it.
