# It was a spontaneous trivia night at Upside Cafe.

By Arya Garg

Canonical source: https://aryagarg23.com/writing/upside-cafe

Published: 2026-10-08

Register: Speculative

Tags: THOUGHT_EXERCISE

Section: [Futures](https://aryagarg23.com/futures)

Field notes from its future: https://aryagarg23.com/futures/upside-cafe

## Main

### The world in my head

We do futures research at a Lab. My job is to work out how different changes could affect each other, then describe possible futures in enough detail that companies can plan for them. Spend all day doing that and you start expecting those futures in your own life. Weirdly, I might relate to method actors. The regular world feels bizarre.

I initially wanted to hate on the current trajectory of automation, following Max Kemats's post about Muse[^upside-cafe-main-1] on the Foresight Lab's Substack. But I veered off course because of whiplash; whiplash between what I study and how people around me are actually living.

What is whiplash........?

I end up living in a pseudo-real world, majorly influenced by my work. In that world, inhumane automation is assumed. Perfectly optimized statistical modeling has distilled human joy out of daily life, and everyone is racing to optimize every bit of friction they feel. Asking an LLM is faster than working something out yourself, so eventually you stop trying. I spend so much time studying that possibility that I catch myself treating it as inevitable.

Employment among 22-to-25-year-olds in highly AI-exposed jobs was about 19% below where it would have been if it had kept pace with less-exposed jobs through June 2026. That's a gap in payroll data, not a measure of how many jobs AI caused to disappear (Stanford[^upside-cafe-main-2]). More than fourteen thousand people have opted into the lawsuit alleging age discrimination in Workday's hiring tools (Mobley v. Workday[^upside-cafe-main-3]). California and other states have enacted chatbot-specific laws, though they cover different things and many haven't taken effect yet (Fenwick[^upside-cafe-main-4]). Kentucky and Florida have sued chatbot companies, alleging harms to children (Kentucky[^upside-cafe-main-5], Florida[^upside-cafe-main-6]). In April, the Senate Judiciary Committee voted 22 to 0 to advance the GUARD Act's restrictions on companion chatbots for minors (Roll Call[^upside-cafe-main-7]).

The U.S. price index for computers, peripherals and smart home assistants rose 8.4% in the year to August 2026 (BLS[^upside-cafe-main-8]). Conventional DRAM contract prices nearly doubled in the first quarter of 2026, up 93–98%. That's the price manufacturers negotiate, not necessarily what you pay for a stick of RAM (TrendForce[^upside-cafe-main-9]). Data centers are also competing for electricity. PJM's auction for 2027–28 fell short of its overall reliability requirement for the first time, with data centers accounting for most of the increase in forecast demand. That is a shortfall against its reserve target, not a prediction of certain blackouts (PJM[^upside-cafe-main-10]). AEP Ohio now requires binding contracts and collateral from new data-center customers to cover infrastructure commitments (AEP Ohio[^upside-cafe-main-11]).

Cloudflare found that humans accounted for 47% of requests for HTML pages on its network on December 2, 2025. Bots made up the rest (Cloudflare[^upside-cafe-main-12]). I built [Lighthouse](https://aryagarg23.com/lighthouse) so you could watch bots visit this site. And in the Fed's 2025 survey, 45% of adults aged 18–29 said they would cover a $400 emergency expense with cash or its equivalent, down from 51% in 2022 (Fed[^upside-cafe-main-13]).

Then the environment. I sat in John Sterman's En-ROADS workshop at MIT last week. The baseline in that session was 3.3°C of warming by 2100. A carbon price moved the number more than anything else we tried. Fusion didn't change the result in that exercise. An hour earlier, a panel on protecting the coast had a tone of inevitability; one panelist said we should not discount having to move away from coastal areas and evacuate them. Seawalls bend the water round the rich part of town and the flood hits the poor part harder. Florida's state-run insurer held 266,231 policies at the end of August 2026 (Citizens[^upside-cafe-main-14]), and Congress passed another temporary extension of the federal flood insurance program in September (P.L. 119-103, section 139[^upside-cafe-main-15]).

Companies use this research to plan. But I find it impossible to disconnect from these worlds even after work. As an engineer, I want to find the root cause, solve it and find the next thing to tackle. But I can't fix all of this. I just keep finding more signals to believe that the world is getting comprehensively fucked. At the En-ROADS workshop, the scientist running it said 3.3°C of warming by 2100 "would be nothing short of catastrophe", and then: "I'm a scientist. Scientists don't use language like catastrophe very often."

### A spontaneous trivia night

What pulled me out of it was a spontaneous trivia night at Upside Cafe. Someone got a question wrong and everyone was laughing. For a couple of hours I wasn't thinking about work.

Part of me feels delusional. Part of me wants to cherish the last bits of joy I'll probably get, because I forecast a really shit world. Part of me wants to stand on the table and scream at everyone to get ready for the revolution.

That's the whiplash.

### Who owns automation?

What makes me angry about automation is who owns it. Most people don't have the hardware or knowledge to run their own models. When you use a model through a company's service, you can't inspect how it works or see what it was trained on. I think about Linus Torvalds, or the women at Raytheon who wove the Apollo computer's memory by hand, and wonder how we ended up here. So much of the internet, the data and now AI is controlled by a few companies.

I kept blaming some VC-backed 19-year-old who sees the benefit of what they are building from their own perspective and ignore what it might do to everyone else.

I ran a little test with 5-6 agents trying to impersonate a young founder. All of them surfaced the same problem: someone else in their position would probably build the same thing. Blaming them doesn't explain why so many people keep making similar choices.

### A map and a Claude glove

Two things from my personal experience that show both sides of this spectrum, from different perspectives.

I wrote about [making a map for Karen on the flight to Boston](https://aryagarg23.com/writing/the-wibbles).

The other thing is one of my co-ops, and how depressing that job was. All day I would get 60-page AI-generated PDFs from my boss, with 70% of the content being repetitive slop. I could have just talked to the model myself. You know? And I used to wonder: if this guy is my boss, and is worth a $20 subscription, why am I being paid $18 an hour to build entire platforms and internal tools for this company? I digress.

Karen used the map to explore what she could see from the plane. My boss was a Claude glove.

### Hard work and twenty agents

We've always had disadvantages based on class. But I used to believe that being smart and working hard would give me a decent chance. It's getting harder for me to believe that.

I can choose to use cash, take public transport or walk. Those choices can make life less convenient, but I can still get by. AI feels different. If someone competing for the same job has 20 AI agents working for them, they can produce far more than I can on my own. I can choose not to use those tools, but working harder won't necessarily let me keep up. When we're judged on how much we produce, understanding the work matters less. You can throw enough compute at it and keep whatever sticks.

For someone growing up in a slum, hard work and intelligence never guaranteed a way out. People with more advantages could tell themselves that was an exception. Now more of us are finding that education and hard work don't guarantee a job or a decent income either.

I can date when this started bothering me. February 2024, a hackathon, a memory journal for people with dementia called Memento. You ask it where you put your glasses and it goes back through your day. Helping someone remember made sense. But building it was the first time I thought: frictionless would be bad if we applied it to everything.

Then we built WhiteBox in September 2024. It answered questions and showed the sources it used. That made more capable AI feel real to me, and it kept me up sometimes. I didn't have the technical experience to work out everything it could become. But I could see how a model might get the context it needed and let us trace where an answer came from. Those had seemed like major obstacles. Models were also getting more capable as they got bigger. That was a big part of why I moved into foresight work.

### The whiplash of normalcy

Picture Larry Page. Picture Jensen Huang. A whole career at the edge of tech. Now imagine your friends all work in tech too. You never leave that hyper-tech world, and then you never feel the whiplash of normalcy.

I think I do a version of this too. Have 50 happy days and one sad day, and the sad day hits hard. And vice versa. You get used to how your life feels. The term is hedonic adaptation[^upside-cafe-main-16], and research suggests variety and surprise can slow it down. That doesn't prove my point about social circles, but it gives me a way to think about it. Put anyone, me included, in a room where everyone sees the world the same way for long enough and they stop being able to feel how strange their own world is compared with everyone else's.

Maybe trivia felt so good because it was so different from how I spend the rest of my time.

### Do we need more friction?

I used to think the fix was friction. Make the technology push back. Then someone asked me: do you want friction in producing food? No. Friction and novelty aren't objectively good for society. Nobody deserves to spend their day sitting in a cube comparing one record against another, which is the job I automated away at my first co-op, and I felt accomplished doing it.

[Interactive figure: Same destination. Different reasons]

What I want is mixing. I want those encounters that make your own world feel strange for a second. That's the whiplash. I think there's learning in it, and I want more people to feel it without having to study futures for a living.

That won't always go well. A meta-analysis of 515 studies[^upside-cafe-main-17] found that contact between groups typically reduces prejudice. But research on negative contact[^upside-cafe-main-18] found that bad encounters can be associated with more prejudice. Meeting someone different isn't automatically enough.

Our feeds keep showing us people who think like us. Spend enough time there and people outside our group start to seem stupid or impossible to talk to. But they know things we don't, because they've lived different lives and done different work. Someone might understand a problem well and still never get asked about it, because they don't have the right job title or connections. The people making decisions keep asking the same small group for advice, so everyone else has very little say.

### Automating my own field

Here is what I did not expect. Futures work, when you boil it down, is finding all these different bubbles people are living in and building towards. Who's building which world? How do those worlds intersect? What causes a bubble to expand into mainstream? Then comes worldbuilding: taking a world that would be shocking, or would seem stupid, and helping people relate to it. But if people just talked to each other, both halves become useless. The bubbles would already know about each other, and no world would seem stupid, because you'd have a friend from it. Criticism would be natural and we futurists would not be needed. Everyone would self-correct.

So I started on why automation is bad, boiled everything down to why futurists are needed, and found exactly how to make the world not need futurists anymore. Computer scientists are plagued with automation. Leave me alone long enough and I'll automate my own field.

I'm fine with that. Everything I've made unintentionally adds or preserves some aspects of friction. WhiteBox asks you to look at the sources. Lighthouse shows you the bots. Gyrus asks you the question back instead of always answering it. Karen used the map to identify what she was seeing from the plane. None of these tools get people talking to each other. They still depend on the person reading, asking questions and using what they know. I never planned that. The exception is Memento, which was for people with dementia and was supposed to remember for you.

### Learning or ignoring it?

Would meeting different people still have that effect if you did it every day? Or would you get used to that too?

There have been attempts to get people to pause before using technology automatically. Twitter tested a read-before-retweet prompt in 2020[^upside-cafe-main-19]. During the election it also prompted people to add a comment before retweeting. Twitter reported that retweets and quote-tweets combined fell by 20%, and removed that change in December (Twitter[^upside-cafe-main-20]). In a September 2026 preprint[^upside-cafe-main-21], a preregistered experiment with 704 people found that feedback about their use of an AI assistant roughly halved the odds of asking it for a complete answer while practising fractions. It measured one session. I want to know what happens after people have used something like that for weeks.

After weeks of using it, have you learned something, or have you just started ignoring the prompts? I don't know how to tell those apart yet. Every whiplash is supposed to be learning, so if it fades because you learned, this hypothesis could hold true. If it fades because you've started ignoring it, it hasn't helped.

I'm experimenting with this on my own site. [Other Minds](https://aryagarg23.com/other-minds) compares how two people think and tries to make their differences easier to understand.

### One more subject

The report cards below show what I'd want schools to teach. The future version adds one subject: holding more than one view. I'd want that taught and graded as seriously as maths or science.

[Interactive figure: Same card, one more subject]

The report card borrows from Robert Kegan's In Over Our Heads[^upside-cafe-main-22], especially his fifth order of mind: being able to examine the limits of your own way of thinking. The school subject is my proposal, not a curriculum Kegan tested. How could schools give students more practice at that?

I don't know what a post-automation world is. But the next time I hang out again, I'm going to keep trying my best not to stand on the table.

[^upside-cafe-main-1]: [Max Kemats's post about Muse](https://theforesightlab.substack.com/p/we-just-hit-the-next-phase-of-widespread)

[^upside-cafe-main-2]: [Stanford](https://digitaleconomy.stanford.edu/news/canariesaug26/)

[^upside-cafe-main-3]: [Mobley v. Workday](https://storage.courtlistener.com/recap/gov.uscourts.cand.408645/gov.uscourts.cand.408645.296.0.pdf)

[^upside-cafe-main-4]: [Fenwick](https://www.fenwick.com/insights/publications/chatbot-compliance-wave-californias-new-law-not-only-one-watch)

[^upside-cafe-main-5]: [Kentucky](https://www.kentucky.gov/Pages/Activity-stream.aspx?n=AttorneyGeneral&prId=1857)

[^upside-cafe-main-6]: [Florida](https://www.myfloridalegal.com/newsrelease/attorney-general-james-uthmeier-files-first-nation-state-led-lawsuit-against-openai-ceo)

[^upside-cafe-main-7]: [Roll Call](https://rollcall.com/2026/04/30/ban-on-kids-companion-chatbots-advanced-by-senate-committee/)

[^upside-cafe-main-8]: [BLS](https://www.bls.gov/news.release/archives/cpi_09112026.htm)

[^upside-cafe-main-9]: [TrendForce](https://www.trendforce.com/presscenter/news/20260601-13070.html)

[^upside-cafe-main-10]: [PJM](https://insidelines.pjm.com/pjm-auction-procures-134479-mw-of-generation-resources/)

[^upside-cafe-main-11]: [AEP Ohio](https://www.aepohio.com/company/news/view?releaseID=10753)

[^upside-cafe-main-12]: [Cloudflare](https://blog.cloudflare.com/radar-2025-year-in-review/)

[^upside-cafe-main-13]: [Fed](https://www.federalreserve.gov/consumerscommunities/sheddataviz/unexpectedexpenses-table.html)

[^upside-cafe-main-14]: [Citizens](https://www.citizensfla.com/-/20260831-policies-in-force)

[^upside-cafe-main-15]: [P.L. 119-103, section 139](https://www.govinfo.gov/content/pkg/PLAW-119publ103/html/PLAW-119publ103.htm)

[^upside-cafe-main-16]: [hedonic adaptation](https://sonjalyubomirsky.com/wp-content/uploads/2024/03/Lyubomirsky-2011.pdf)

[^upside-cafe-main-17]: [meta-analysis of 515 studies](https://doi.org/10.1037/0022-3514.90.5.751)

[^upside-cafe-main-18]: [research on negative contact](https://pubmed.ncbi.nlm.nih.gov/22941796/)

[^upside-cafe-main-19]: [Twitter tested a read-before-retweet prompt in 2020](https://techcrunch.com/2020/09/24/twitter-read-before-retweet/)

[^upside-cafe-main-20]: [Twitter](https://blog.x.com/en_us/topics/company/2020/2020-election-update)

[^upside-cafe-main-21]: [September 2026 preprint](https://arxiv.org/abs/2609.20143)

[^upside-cafe-main-22]: [In Over Our Heads](https://books.google.com/books/about/In_Over_Our_Heads.html?id=gOssEAAAQBAJ)