광고
광고
광고
광고
광고
광고
광고
광고
광고
로고

세계적으로 중요한 인공지능 비영리 단체들. 협력이 용이한 단체들 등

운영자 | 기사입력 2022/01/30 [07:56]

세계적으로 중요한 인공지능 비영리 단체들. 협력이 용이한 단체들 등

운영자 | 입력 : 2022/01/30 [07:56]

 

1. 한국유치가 필요한 중요단체들

 

1) aiforgood.ITU.INT: 유엔의 AI관련 SDG목표 달성를 위해 인공지능을 잘 활용하자는 단체로, 다양한 행사를 진행하면서 인공지능의 역할을 정의하고 유엔의 SDG목표달성을위해 노력하려는 단체이다. 

https://aiforgood.itu.int/

 

2) AGI Society 일반인공지능협회 비영리단체로 인공지능은 Artificial Narrow Intelligence(알파고, 자율주행처럼 한가지밖에 할줄 모르는 좁은AI, 그러나 인간처럼 더 진화한 Artificial General Intelligence가 한꺼번에 인간처럼 여러가지 생각을 하고 다양한 기술을 가진 ChatGPT같은것을 AGI라고 하며, 마지막으로 Artificial Super Intelligence는 인간이 수퍼맨처럼 날아다니고 인간이 상상할 수 없는 힘, 능력, 속도를 가지는 아직다가오지 않은 인공지능을 말한다.) 

https://agi-society.org/

 

3) OpenCog 인공일반지능협회 비영리단체:  AI전문가들이 모인 단체로 OpenAI 일론머스크가 먼저 이 도메인을 가져갔지만 먼저 시작된 운동으로 인류의 복지를 위해 일을 하겠다는 단체

https://opencog.org/

 

4) SingularityNet 인공지능관련 모든 전문가및 기업 협회, 비영리단체: 거대한 인공지능 빅테크와 협력과 경쟁을 하기위해 모였으며 다양한 인공지능의 평화적인 사용 기회를 찾고, 인공지능 전문가들의 협업의 장, 그리고 글로벌 브레인등을 만드는 단체. 특히 인공지능으로 창업하는 스타트업의 모임은 Singularity Studio에 모여있다.

https://singularitynet.io/

 

5) Mondragon Corporation: 스페인의 협동조합으로 자체내에 정부, 국회, 몬드라곤 대학 등이 존재하며, 이곳에서 몬드라곤 AI 팀이 존재한다. 

 

6) IEEE, Artificial Intelligence Standards Committee

https://sagroups.ieee.org/ai-sc/

 

7) The SAE Ground Vehicle Artificial Intelligence (GVAI) Committee 

https://standardsworks.sae.org/standards-committees/artificial-intelligence

 

8) ISO/IEC JTC 1/SC 42, Artificial intelligence

https://www.iso.org/committee/6794475.html

 

 
세계일반인공지능, 오픈코그, 싱귤래리티넷 등의 산하에 존재하는 각종 스타트업들
SingularityNet, We have a number of projects that might be appropriate for Korean funding, and Janet can gather and send you materials on these

-- TrueAGI, productizing OpenCog AGI tech for the enterprise
-- Rejuve Biotech, designing longevity therapies by applying ML to genomic data from long-lived flies and humans
-- Simuli, AGI chips, which you know about
-- NuNet, Internet AI, which has done a token sale but also has a for-profit company associated, which is building an amazing decentralized computing platform (allowing sharing of processing power across many different machines including home computers, phones, server farms, etc.)
--  Hypercycle, customized blockchain infrastructure for AI, being led by Toufi ... designed to run as a sort of underlayer for SingularityNET
-- Awakening Health, robots and (eventually) avatars for healthcare and eldercare, you know about it
-- Musaic, AI for helping with music creation ... like ChatGPT or Dall-E but for music, basically
-- Sophia DAO, Sophiaverse.ai 소피아 인공지능과 메타버스의 협업을 시도하는 단체
-- ChatGPT Korea:  Matt and the team
 
2. ChatGPT Korea:
 
현재 챗GPT는 한글로 질문했을때 만족스러운 답이 잘 나오지 않을 뿐만아니라, 전세계에 한국의 연구조사 결과, 한국의 논문들, 한국에서 생산한 각종 자료들이 입력이 되지 않아 잘 끌려나오지 않는다. 고로 한국의 모든 자료, 각종 학계의 논문및 연구결과, 각 대학의 연구결과 등 뿐만아니라 한국정부의 각종정책, 각 정부기관 산하기관들의 역할및 자료들을 집대성하여 ChatGPT Korea에 넣고 훈련을 시키지 않으면 한국의 홍보가 왜곡되거나 제대로 전달되지 않는다. 이를 위해 ChatGPT msKorea 를 만들어 결국 국위선양, 국가기관 기업, 사회홍보를 해야한다. 이를 위한 국가예산이 필요하고, 우선 예산이전에 관심있는 기업을 이 인공일반지능협회와 연결시켜서 시작을 하는 방안이 필요하다.
One other thing that would be useful would be to find some big company or gov't agency willing and able to spend US$5M or so on server time to train a large language model similar to GPT3, which could then be made open source and widely available, and which we could use to fuel a lot of other AI work.   We could make sure the model was trained on all possible Korean language data and was as brilliant as possible in Korean.   We know how to train a model like this, and have a lot of ideas about how to put it together w/ OpenCog to achieve even greater intelligence for so many applications and for AGI R&D ... but while SNet is doing OK financially we don't have $5M to spare for paying for server time to train a model like this....   Potentially this could happen as part of a partnership including funding of one of the other projects mentioned above.   To gov't agencies, the pitch would be: Let's have the next GPT3-like breakthrough model trained and hosted in Korea...  We can make a system that is like ChatGPT3 but less full of crap and with a better understanding of truth vs. falsehood, by incorporating OpenCog reasoning methods along with the GPT3-like model... but we need to train our own GPT3-like model first so that we can fine-tune the. model freely for our purposes and query it unlimitedly and probe its internal states etc.

 

Matt Ikle' who can help put materials together on this, though probably not by tomorrow.
There are a few key points here I think....
 
1) To train a model comparable to (or better than ChatGPT) will take around $5M-$7M USD  worth of compute time on multi-GPU servers.   It's expensive, which is why only Big Tech companies or very well funded tech startups have tried to train this sort of model.
 
2) We have novel techniques for using LLMs (specifically, models like OpenAI Codex or Salesforce's CodeGEN, which. model computer program code along with natural language) to translate natural language sentences into symbolic-logic  format.  This allows us to do reasoning to distinguish truth from falsehood in a way that LLMs cannot do on their own.   However applying reasoning at the large scale to knowledge from an LLM will also be expensive, probably $2M-$3M of computer time at the large scale needed for models like GPT3/ChatGPT
 
3) Bottom line then, we are looking at something like a 2 year initiative that would cost something like $10M USD for computer time and something like $8M USD for human expenses (roughly)
 
4) The result would be a system similar to ChatGPT but really good at Korean language, and with a much stronger ability to distinguish truth from BS

If a project like the above is in the general scope of interest, then Matt and I can put together a more careful proposal w/ help from our colleagues.
 
I'm sorry the costs for this sort of thing are a bit high but it's easy to check that OpenAI and Google have spent much more than this on training their own models.. and this is the sort of reason that Stability AI (which trained the Stable Diffusion generative image model) raised $100M in early-stage capital, etc.
 

3. 홍보 덱 decks (위의 단체들의 상세 자료)

 
 
PHOTO
1/6
광고
광고
광고
광고
광고
많이 본 기사
유엔미래포럼 많이 본 기사
최신기사