Since last year, there has been a change in the discourse on innovation in the technology sector. If between 2021 and 2022 terms like metaverse, web3 and crypto dominated conversations, in 2023 it was artificial intelligence that established itself as the center of attention.
It makes sense: tools like ChatGPT and Midjourney have very clear uses. The operation of these products can be observed here and now, with applications ranging from Internet search to improved productivity. While the metaverse is a speculative endeavor, AI is already proving useful in practice.
The excitement is therefore justified. And it foments a race in research and development in the field, a search for new potentials to be achieved with artificial intelligence.
Many people participate in this search, but there is one name that has stood out among the rest. This is the organization that, in many ways, is largely responsible for the new wave of enthusiasm for artificial intelligence: OpenAI.
A non-profit start
OpenAI was founded in 2015. In addition to Sam Altman, current CEO, and Ilya Sutskever, chief scientist, the founding group included Elon Musk (who left the company in 2018). Among the project’s financiers were investor Peter Thiel and Amazon Web Services (AWS).
In the beginning, OpenAI was not a traditional company founded for profit, but a not-for-profit organization. And the “open” in the name expressed the aim of operating transparently, collaborating with other players in the sector.
According to inaugural post on the company blogOpenAI’s goal was to promote digital intelligence in a way that it promoted the benefit of “humanity as a whole”, without being tied to the need to generate financial return.
But research in the field of AI is not cheap, and OpenAI needed to attract more investment. For this purpose was created the OpenAI LP, an entity that would operate seeking “capped-profit”. This occurred in 2019. In the same year, Microsoft injected US$ 1 billion into the company, starting a partnership that lasts until today.
Together, the two companies began to shake up the Internet search game, upsetting Google. And therein lies the irony of it. After all, there was Google’s contribution to what OpenAI would become.
Clear search
As a research laboratory, OpenAI was part of a collaborative landscape. Private institutions and universities shared data and models so that other researchers could contribute and refine the results. OpenAI benefited greatly from this.
This is what Diogo Cortiz, a cognitive scientist and professor at PUC-SP, points out in an interview on Tecnocast 282. According to him, this cooperation environment developed organically, with exchanges between different laboratories. Progress in the area, therefore, took place from a science made with collectively.
When we think of the ChatGPT tool itself, it was built based on an architecture called Transformers, a neural network architecture that was proposed by Google in 2017. This was written in a paper, then it was implemented, Google trained the model, made it openly available.
Diogo Cortiz
Meta researchers also worked on top of the same architecture provided by Google, as well as OpenAI. In a way, Google’s open search fostered who would become its competition.
Over time, due to the need for more computational power to pursue research, the private sector ended up gaining prominence in the area, with emphasis on Big Techs. It is in this context of heavy investment that tools such as ChatGPT emerged, throwing artificial intelligence into the public’s lap.
And as that happened, OpenAI lost sight of the emphasis on openness that seemed to guide the company.
the black box
In the announcement of GPT-4, the technology that makes ChatGPT and the new Bing work, OpenAI chose not to reveal information about language model training. We therefore do not know what data the AI was trained with, nor the number of parameters, nor the learning strategy. No opening here.
It is a movement that goes against the idea of transparency and open knowledge that seemed so dear to the research community. For this reason, OpenAI has received criticism in the industry. However, this change appears to be here to stay.
This is what Ilya Sutskever implies. When asked by Verge on the reason for this shift away from the openness of the past, the co-founder of OpenAI replied, curtly: “We were wrong.”
The arrival of artificial intelligences in the mainstream changed the game. Now, due to strong competition in the area – after all, Google also works on similar tools – OpenAI can reserve the right not to open your data.
The decision could have impacts beyond OpenAI. If other companies follow suit, the development of a potentially transformative technology could end up moving in the dark.
Diogo Cortiz summarizes the situation when he says that OpenAI’s new language model is a “black box. Sealed. And hidden at the bottom of the sea.” After benefiting so much from open research, OpenAI doesn’t seem interested in returning the favor.
