Alexandra Rome

Assistant Marketing Professor| Consumer Culture Theorist| Sexuality and Relationship Researcher| avid #netflix fan

[Daniel Kokotajlo @DKokotajlo]

In AI 2027, we predicted that AI would take over the world or irreversibly concentrate power.

In AI 2040: Plan A, we've laid out our positive vision for what should happen instead.

[Harlan Stewart @HumanHarlan]

There’s a good chance that in 2028, we’re alive, much of this didn’t happen, and people say “See? It was all nonsense, there’s no danger.”

But that’s not how it works. If a weather forecast is wrong about the hurricane hitting at exactly 5pm, don't assume things are safe at 5:15

[Alexandra Rome @Romealex]

I desperately wish more political conversations centered around #AGI and the risks of human extinction. https://t.co/FLBCIG3ERj

[Alexandra Rome @Romealex]

Very proud of @JoeBiden for making such a difficult and selfless decision. Record speaks for itself. Excited for @KamalaHarris

[Joe Biden @JoeBiden]

My fellow Democrats, I have decided not to accept the nomination and to focus all my energies on my duties as President for the remainder of my term. My very first decision as the party nominee in 2020 was to pick Kamala Harris as my Vice President. And it’s been the best decision I’ve made. Today I want to offer my full support and endorsement for Kamala to be the nominee of our party this year. Democrats — it’s time to come together and beat Trump. Let’s do this.

[Alexandra Rome @Romealex]

I don’t think what’s happening at @OpenAI is funny, but this is. 😂

[Naomi @nassaraf]

@sama *Elon pops up in Ilya’s DM’s like*

[Eliezer Yudkowsky ⏹️ @ESYudkowsky]

Right, I forgot Disanalogy Three: Humanity is currently too small and weak to be using all of the planetary resources. We haven't yet intercepted all the sunlight, we don't use all the carbon atoms. https://t.co/UwjcLrXBOg

[Eliezer Yudkowsky ⏹️ @ESYudkowsky]

I wish folks on my-side-ish would stop using arguments that go, "A superintelligence would be to humans like we are to insects." I think the relations ASI::humans and humans::insects are just not very similar; I am surprised if people who aren't already convinced, find this argument convincing.

Problem One: Some humans care about insects. Even I care about insects; I think it's sad when any living thing has to die. I kill flies for my own convenience, but I'm not happy when I do. I don't think that flies I kill are conscious and unhappy, to be clear; it's just unpleasant, unaesthetic. My old email headers, if you dug back up emails I was sending at age 13, had the Organization line -- which was a thing back then -- set to "Mitochondrial Liberation Front".

So reason one I don't like the insect analogy, is that some humans care about insects, or even mitochondria. The whole entire worry is that we do not presently have the competence at alignment to make machine superintelligence care at all, not even a little tiny little bit. If somebody thinks that's improbable, then you may need to sit down and have a long conversation about the is/ought gap and how historically-causally humans ended up caring about things. Using the analogy of superintelligences being to humans, as humans are to ants, will lead some good and honest people to reply, "But I do, actually, find it sad to kill ants; and if I were a god I would never have to kill ants again."

Problem Two, and this one will take a little longer:

When humans face down tigers, insects, bacteria, or viruses, what we're facing is not really the cognitive neurons in the tiger or the insect, or the nonexistent brains of the virus or bacterium. We're facing the cumulative optimization power of natural selection. Humanity is not yet so powerful that millions of years of blind mutation and myopic selection can't put up a fight against our centuries of engineering.

We can build skyscrapers taller than any tree; and nuclear reactor cores with higher energy density than any stomach; we have gone to space where no animal or plant ever trod. But we have not built a micron-diameter self-replicating solar-powered factory, like an algae cell.

That's not because the algae cell is infused with essence of vital force that no cold metal can ever posesss. It's because natural selection has invented a bunch of particular detailed biomechanical gears and wheels -- literally, in the case of ATP synthase and the bacterial flagellum. Those are, in fact, 2 out of a total of 3 known cases in which biology has designed a freely rotating wheel.

Humanity uses a lot more freely rotating wheel designs than 3 total. Wheels are genuinely useful, and they're easier complications for intelligence to design than for natural selection to evolve. But we do not have any of our own molecularly precise gear designs with known synthesis pathways. Humanity is just now getting to the point where AlphaFold 3 can predict those sorts of molecular interactions, literally as of a couple of weeks ago as I write this. We'll probably have to wait for AlphaFold 4 on designing them.

The fights between humanity and insects are not too lopsided. Both sides can put up a fight, with neither side having an instant win. In humanity's corner, cumulative centuries of engineering; on the insect's side, millions of generations of random mutation and myopic selection. The combined power of humanity could win a fight against any single insect, if we knew the battle was coming and we were given time to prepare. But on a larger scale, humankind does not utterly dominate insectkind; individual humans still struggle to keep ants out of their house.

Humanity still keeps cows -- maybe not under awesome living conditions, note, but we have not dispensed with cows. Cows can do a better job of turning grass into meat than the current set of artificial-meat technologies still under development, only in the last few months matured to the point where Florida state law felt a need to ban them.

Again, that's not because cows are infused with irreducible vital force that humanity's cold steel can never duplicate. It's because Deepmind hasn't yet debuted AlphaFold 5, and we still need to do all the biodesign stuff via a very slow, painstaking path. Which, to be clear, still goes much faster than natural selection's required timescales! But we have only been at the lab-grown meat business for a few decades.

Would humanity, given time to do science and engineering at current rates of progress for 10,000 years, still have a need for cows? I doubt it. Biochemistry doesn't take that long. It takes decades, not years; but we have no reason to imagine that it would take millennia, even if AlphaFold 5 never did get invented. I think it would be more like centuries, not millennia, before humanity restricted to AlphaFold 3 levels of biochemistry prediction had dispensed not just with cows, but with the conventional biomachinery of DNA-based life. The current crop of lab-grown meat, to be clear, is still built by cells, they're just genetically engineered cells; lab meat still uses ribosomes and proteins, rather than some entirely new framework of neo-biochemistry. But I doubt that would take more than a couple of centuries to change, at current rates of progress; and with AlphaFold 4 it might be decades instead of centuries.

And THEN we'd have no need of cows, because we would have designed our own biomachinery to do what cows do for us now, but faster and cheaper.

The concern about artificial superintelligence, among other things, is that It has access to Its own in-brain equivalent of AlphaFold 7, and so It doesn't ever lose any fights to insects or bacteria or viruses; because It actually can deploy machinery on (at least) the same level that biochemistry builds using a few million years of random mutation and nearsighted selection.

So Problem Two with "ASI::humans ~ humans::insects" is that humankind and insect evolution are fighting a somewhat even battle. The concern about artificial superintelligence is that It is far enough past human engineering to fully beat millions of years of blind mutation and nearsighted selection. It doesn't need to keep some humans in barns to use them as food, It has lab-grown meat. It does not have occasional even fights with human civilization, the way that individual humans fight with the evolved powers of ants or bacteria. It does not trade with us, because It can put together the atoms in our body into more economically productive configurations.

Between Problem One and Problem Two put together, I think "ASI::humans ~ humans::insects" is a bad and invalid analogy which doesn't even yield the argued conclusion. Humans care about insects, fight roughly even battles with evolution-of-insects, and have uses for insects-that-evolution-built which humans cannot readily replace.

The concern about artificial superintelligence is that It beats out not just the cognitive powers of individuals humans, but the cognitive powers of humanity; and to a degree where ASI is also no longer fighting real battles with evolved biology. It has no use for keeping humans in vats to make use of our meat, our ideas, or our electrical generation.

And as for keeping us alive and in good health and free cheer because It cared about us enough to want to do that for its own sake, to spend resources whose only reward is humanity being happy and alive and free -- the notion that this kind of caring is in any way a small target to hit, or a difficult target to hit, is the topic of alignment theory. And if you leave out that question, it doesn't actually follow validly that superintelligence would kill us; because it would only take a tiny bit of caring to keep us alive, and, yes, make us as gods. We're just not on course to get even that tiny bit of caring. It doesn't happen automatically; it's information that needs to be in a particular place, and it's not significantly easier to get that information weakly rather than strongly. It's like asking for a book that only needs to be a very faint copy of Shakespeare's lost folios; letting the ink be weaker doesn't make the informational problem any easier. You can either solve it or you can't.

Some humans would never kill an ant, if they didn't have to; and if they were gods, they'd never have to. The problem is if building an artificial superintelligence with that same kindness is the sort of problem that takes a few years and a lot of retries, and involves some difficulties that don't show up clearly in non-superintelligences.

[Alexandra Rome @Romealex]

When #virality loses its meaning @washingtonpost https://t.co/7wHc5DKv2m

[Alexandra Rome @Romealex]

The silver-lining to embracing this season of #motherhood… #psychoanalysis #newyearresolutions

[George Ferns @ThisGeorgeFerns]

Does your work and career make you anxious? Take a lesson from #CarlJung and embrace the #feminine. @aliette_lambert @ConversationUK @BathSofM @BathCBOS
https://t.co/vrANffgD7w

[Alexandra Rome @Romealex]

Excited to share this @MktgTheory_news commentary written with @aliette_lambert and Francesca Fornari, where we ‘write differently' as we reflect on our #digital dependency to #digitaltechnology and #socialmedia https://t.co/IEPf2B0Dvx

[Eliezer Yudkowsky ⏹️ @ESYudkowsky]

@webmasterdave Yes if there are states of matter or being that they prefer to paperclips.

[Alexandra Rome @Romealex]

Working on a new project.... Sex In The #Metaverse https://t.co/tvOiuWshyD

[Alexandra Rome @Romealex]

“With her worm outfit it was as if, whether consciously or not, she was helping to underscore the endless heavy lifting that a certain kind of femininity requires” #abjectification https://t.co/AP5VOXBSoY via @NewYorker

[Alexandra Rome @Romealex]

@runcarly Envious!! Happy birthday 🥰

[Alexandra Rome @Romealex]

Refusing to offer #flexiblework and/or part-time work is #maternaldiscrimination and #sexdiscrimination. And this is in the so-called “progressive” EU… #academicmama #AcademicTwitter @PregnantScrewed @EverydaySexism

[Bill Maher @billmaher]

There’s a disturbing trend going on in America these days with rewriting science to fit ideology. We’ve gone from fat acceptance to fat celebration.

[Alexandra Rome @Romealex]

I’m a big @elonmusk fan but I feel like this whole #Twitterdeal has real consequences for real people (employees, shareholders, etc.). Maybe privileged people hurting privileged people, but nonetheless nags at me…

[Alexandra Rome @Romealex]

@timminglab Happened to me last week!!!

[Alexandra Rome @Romealex]

Math for mamas @NewYorker https://t.co/E6Ivjgy7jD

[Alexandra Rome @Romealex]

Ditto all of this 💪🏼