Chatbots, online translators, text generators, and grammar checkers – these are a few of many modern applications of artificial intelligence language models. The most popular and advanced models like GPT-3 or BERT are claimed to “understand” human language. Unfortunately, these solutions are not open to the public. Why? Because mainly big corporations like Google, Microsoft or Huawei can afford to access enormous computing resources and have researchers to develop them.
?According to the official paper of GPT-3, it would take 355 years to train this AI model on a computer equipped with one of the most advanced graphics cards.
Open-source solution
To tackle the problem of the centralization of hyper-advanced AI models, hundreds of researchers from all around the world have gathered to bring an open-source solution to the public. Independent research and industrial volunteers formed a collaborative project called BigScience and created an enormous language AI which will complete its training in 4 months – anyone can track the progress in model training here: click!. The model is predicted to be excellent in performing various language-related tasks in 46 languages ?.
Implications of open-source
Currently, most advanced AI owned by hi-tech companies can be accessed by individuals, but one who wants to use them either needs to match specific conditions or pay for the access. One of the reasons is that the organizations that own AI solutions like GPT-3 still don’t know all of the potential misuses that could harm people and thus, want to control the purposes their solutions are used for. However, the new AI model created by BigScience collaborative community is most probably going to be open to anyone. It means that its use-cases will not be controlled.
Governments should join the collaboration
Hyper-advanced AI models can solve many problems and bring enormous value to society. In my opinion, collaborative projects like BigScience show that sooner or later, hyper-advanced AI solutions will be available to anyone. That’s the reason why I believe that governments should look into the problem of harmful applications of AI and establish a law to prevent such cases. Hopefully, that would encourage at least some hi-tech corporations to share their resources to help in democratizing advanced AI safely!
Sources:
https://huggingface.co/bigscience/tr11-176B-ml-logs
https://openai.com/blog/better-language-models/
https://ied.eu/wp-content/uploads/2018/05/sharing-economy.png
https://www.reddit.com/r/GPT3/comments/p1xf10/how_many_days_did_it_take_to_train_gpt3_is/
I think it’s great that there are projects like BigScience that are working to make advanced AI accessible to everyone. I think it’s important for governments to get involved in regulating AI, so that it can be used safely and for the benefit of society as a whole.