1

Translation of BlueDot Impact's AI alignment curriculum into Portuguese

ActiveGrant
$3,300raised
$7,000funding goal

Project summary

Translating BlueDot Impact's AI Alignment curriculum into Brazilian Portuguese.

What are this project's goals and how will you achieve them?

This project's high-level goal is adding to the scarce amount of high-quality, up-to-date resources available to Brazilian people who are interested in AI safety as a career path, or might get interested in it if exposed to the matter.

As to the more implementational side of it, the EA community in Brazil intends to use the material in (1) university study groups and (2) online courses. Last semester I (Luan Marques) translated the BlueDot Impact's Governance curriculum, with a collaborator who runs the Brazilian 80.000 Horas website, Fernando Moreno, and it was used by a UGAP-supported study group at the University of São Paulo in Ribeirão Preto. First, we intend to launch an online governance course modeled on BlueDot's. Later, with the AI alignment translation in hand, we intend to launch an AI alignment online course.

Of course, we also intend to make the AI alignment translation available to university study groups, and the other one, at Getúlio Vargas Foundation in Rio, is worth mentioning here. In Rio, they opted for the alignment course in English anyway, helped by machine translation. So that shows some interest. But the language barrier is still a problem. For example, out of the 12 people that regularly participate in their meetings, 4 say their understanding would improve by 50% with translated material, and 4 by 10%.

The translation will be done by machine translation post-editing, by myself. I must say that I've gotten permission to translate from BlueDot Impact, as well as at least one author of almost every piece of content in the curriculum: 57 out of 64 answered me positively [or were appropriately licensed] while only 8 haven't answered me yet. Also, we intend to make the curriculum available for free on the 80000horas.com.br website, just like the Governance curriculum is: https://80000horas.com.br/fundamentos-de-seguranca-de-ia/governanca-da-ia-um-curso-introdutorio/

How will this funding be used?

This funding will be used as income for the translator's (Luan Marques) work, which is expected to take 2-3 months.

Here's the breakdown: The curriculum’s total word-count is around 176,920. Language professionals’ website ProZ shows a rate for EN-PT translations varying from 0.09 USD (standard) to 0.07 USD (minimum) per word. I see many experienced Brazilians charging a typical rate of 0.06 USD, so let’s put that as the minimum, instead. Assuming machine translation post-editing as 70% of that, it ranges from 0.06 USD to 0.04 USD per word. That totals 10.615 USD to 7.076 USD.

Who is on your team and what's your track record on similar projects?

I'm involved in EA Brazil community, and collaborate with its main actors, Fernando Moreno (head at the 80.000 Horas website: fernando.olivi@gmail.com), Leo Arruda (Head of community at the Condor Initiative: leo@condorinitiative.org) and Juana Martinez (Community Builder: juanamart030@gmail.com).

I have translated a fair amount of EA-realated content. The utilitarianism.net website (unfortunately, not out yet), co-authored by Richard Chappell, Darius Meissner, and William MacAskill. Parts of the Precipice, and of the EA handbook. And as said above, I’ve mostly translated the whole of BlueDot’s governance curriculum.

As to giving courses, the EA Brazil community has some experience, for example, giving an EA Intro course on an ongoing basis, run by the Condor Initiative.

What are the most likely causes and outcomes if this project fails? (premortem)

I said above, about the study group in Rio, that “out of the 12 people that regularly participate in their meetings, 4 say their understanding would improve by 50% with translated material, and 4 by 10%.” You might conclude that, at least for young Brazilians at undergraduate age, who must have an understanding of English better than the vast majority of Brazil, a supplementation is still necessary, at least for content that is so technically demanding as this.

But thinking about an online course, that fact might not generalize to the rest of the population in Brazil that is interested in these matters, so that a translation might not have much of an impact.

Also, much of the curriculum might be updated too regularly, so that this effort might be superfluous. 

What other funding are you or your project getting?

I've applied for EA Funds: Long-Term Future Fund, and got turned down (and they suggested I apply here!). But I'm still waiting for their response for another fund, the Effective Altruism Infrastructure Fund.

Luan-Rafael avatar
Progress update

What progress have you made since your last update?

At the beginning of June, I finished the translation of the Alignment Curriculum. It is published open access on the Brazilian website 80000horas.com.br:

https://80000horas.com.br/alinhamento-da-ia-um-curso-introdutorio/

We also have a translation of the Governance Curriculum:

https://80000horas.com.br/fundamentos-de-seguranca-de-ia/governanca-da-ia-um-curso-introdutorio/

Unfortunately, we don’t have many great prospects for putting this material in good use at least in the next semester, although we’ve had some hopeful developments:

-Our project of an AI Safety Fundamentals virtual programme in Portuguese hasn’t got off the ground yet.

-One study group (at the University of São Paulo - Ribeirão Preto) seems to be currently inactive, and another (at Getúlio Vargas Foundation - Rio) is currently facing some uncertainties about how to proceed for the next semester. 

-Fortunately, a formerly inactive group at the University of São Paulo - São Paulo is coming back to life and they expressed interest in the material.

-At the end of June, we organized a small, experimental local EA conference here in Brazil, and it was quite successful (around 200 people showed up). Among its main subjects was AI Safety, and that might get some people interested in forming more groups in future.

https://forum.effectivealtruism.org/posts/J5miiEvm72mPCr8uk/announcing-ea-brazil-summit

donated $300
🐯

Loppukilpailija

6 months ago

This translation project seems worthwhile to me; I gave a small retrospective donation as a way of saying thank you.

Tangential, but I'm curious about how you proceeded with the translation. Are current machine translators (such as DeepL) good enough to be useful as a first draft, or did you do it completely via human work?

Luan-Rafael avatar

@Loppukilpailija, I really appreciate your contribution. :) (If you're interested, the translation is available here (I'm just in the process of finishing the final chapters' images): https://80000horas.com.br/alinhamento-da-ia-um-curso-introdutorio/)

I did it mostly by machine translation post-editing (the subtitling work, for example, is hard to do by MT post-edting when the original subtitles are not well put together, and they mostly weren't, so they were mostly done by human). I did it with a mix of Google, Deep-L and Chat-GPT, mostly the latter. You can really be impressed by how useful they can be, but also by how and where they need editing. But some errors are quite understandable: standadization of terms, idioms and sayings, complicated English grammar and anglicisms that even natives might make mistakes about, context, etc.

Again, thank you so much

🧡
Jason avatar

Jason

8 months ago

Has there been any recent progress on this, or a decision from EAIF or other possible funders?

Luan-Rafael avatar

Hi there @Jason, Thanks for the interest. Unfortunately, I haven't received any followups from the EAIF. And the only other one I tried for was this one, so...

So, I went on working with the funding I got, and i'm currently ~85% through.

donated $3,000
RenanAraujo avatar

Renan Araujo

11 months ago

Why I’m excited about this project

  • Translations are a robust, low-risk, and relatively low-cost way to spread useful information to untapped audiences and regions.

  • Open Phil  previously funded several translation grants but moved away from funding part-time and/or individual translators to fund more easily scalable, larger efforts (more here). Luan was one of the translators who didn’t get such funding. 

  • I haven’t read Luan’s translations, but he seems generally high-context, smart, and knowledgeable in English. Since he already did similar translation work before, I trust his ability to deliver well in this project.

  • Since Luan reached out to me, it was almost zero cost for me to find this well-positioned translator. And, different from OP, I only have a small regranting pot anyway.

  • Luan’s translations seem unusually useful since they already have a pretty well-defined target audience: newly-founded AI safety study groups in top Brazilian universities.

Challenges and concerns

  • Maybe it’d be more cost-effective to fund a translator for Spanish or some other more popular language than Portuguese. I’d be keen to hear from folks doing that kind of work and for Manifund to support them, especially if OP isn’t.

    • But I expect the most popular languages to be correlated with OP support. I think I’d have expected even Portuguese to receive that kind of support, but maybe the lack of a large, professional translation service for EA work in Portuguese is the bottleneck here. 

  • I’m not sure translating the AGISF curriculum is the best option, considering I’d expect most of the target audience for that to read English relatively well and because I think there’s some nontrivial chance the curriculum will be updated relatively often (caveat: I haven’t checked this with Bluedot). 

    • But I updated after seeing the information Luan collected on people finding reading in Portuguese better for leaning, and I can’t think of an alternative resource that is as good for an introductory-level audience.