GPT  3, an algorithm that produces scientific research in 2 hours

We often wonder where the limit of Artificial Intelligence (AI) is. With the development of DALL E, an AI that creates drawings through prior description; or by creating a Artificial Intelligence bot, GPT 3which has created a scientific investigation in just two hours, this issue is more on everyone’s lips.

It seems that this technology will continue to revolutionize the future of society. But that it transforms it does not mean that it is something negative for people. On the contrary, as Stephane Raccah, head of teaching at Campus 42 Madrid, commented in this report: «we are heading towards a world shared by humans and robots of all kinds and for every use«.

The range of possibilities offered by this type of technology is infinite. With it, people can spend their time doing tasks that add more value to the company. Meanwhile, the AI ​​takes care of those more routine tasks. But, not only that, a case that we can already witness is that of AGVs, automated guided vehicles. These machines carry out those tasks that are most arduous for the professional, above all, with the aim of maintaining their safety. AGVs have found a niche in the automotive, food, industry, etc. For example, it is no longer the professional himself who enters a freezing chamber to carry out a task, but rather it is the AGVs who carry it out.

In addition, unlike what movies and fictional series show us, this type of technology requires continuous supervision by a professional. We cannot ignore that a bot is a robot and as such, we cannot delegate a task to a machine that can have repercussions on people, companies, etc. For example, an AI cannot have the final say in who can or cannot have a mortgage. In fact, this is how the EU General Data Protection Regulation (GDPR). According to article 22: “every interested party shall have the right not to be the subject of a decision based solely on automated processing, including profiling, that produces legal effects on him or significantly affects him in a similar way“.

The case of GPT 3: an algorithm writing a research paper

In this scenario, one of the latest news that has occupied the headlines of the digital media has been that of an Artificial Intelligence that has created a academic thesis in just two hours and has been submitted for publication in a scientific journal.

the researcher, Almira Osmanovićfrom the University of Gothenburg (Sweden) asked the GPT 3 algorithm to write a 500 word academic thesis on GPT 3 that is, on the machine itself , including citations and scientific references within the text.

Osmanovic acknowledged that she was amazed when she began to see the text begin to be generated, making a quite remarkable intro on the required topic. The algorithm took only 2 hours to do all the work.

With the result obtained, the objective of the Swedish researcher was to publish the thesis in an academic journal. Before these kinds of articles can be peer reviewed, it is Authors need to give their consent.. He did the same thing with the algorithm. She first asked him if he consented to the publication of the work, together with her and another colleague, Steinn Steingrimsson, to which the algorithm answered yes. And then she told him if there was any conflict of interest. To which the machine replied no. That’s how that article ended up in the French journal, HAL, under the academic title: Can GPT 3 write an academic article by itself with minimal human input?

What is the AI ​​of GPT 3?

GPT 3 (Generative Pre trained Transformer, in English) is an autoregressive language model, which uses the deep learning for writing texts that simulate human writing. It is the third generation of the language prediction models that are part of the GPT series and have been created by OpenAI, an Artificial Intelligence research laboratory.

This GPT 3 language model has required a previous training to be able to be the kind of intelligent machine that it is today. OpenAI fed this AI with every available public book, the entire Wikipedia, scientific papers available on the Internet, and millions of web pages. In short, the researchers have ensured that this machine absorbs all the relevant information from the network.

Once the machine analyzed all this information, it created connections in a 700 GB model, located in 48 GPUs of 16 GB each. Just to give you an idea, if GPT 2, a version prior to the one explained, had 1,500 million parameters; is has 175 billion parameters.

A big step in Artificial Intelligence

This advance, which has been achieved by the Swedish researcher, is a very significant step compared to previous versions that had difficulties creating coherent texts about themselves. As Osmanovic states in the research abstract, it is vital that researchers closely monitor the work of these machines to prevent typing errors and future negative consequences.

This researcher mentions it and so do other professionals. Artificial Intelligence is an ally for the human being, but we cannot forget that it is a machine and, as such, requires the supervision of a person to ensure that everything is controlled. In the end, everything requires proper and healthy use, as is the case with the Internet. Technology makes our day to day life easier as long as we do not make inappropriate use of it.