Last week when Meta released Galactica, a category of large language models (LLM) that had been trained on “106 billion tokens of open-access scientific text and data. This includes papers, textbooks, scientific websites, encyclopedias, reference material, knowledge bases, and more”( Statement from Meta) , everyone was amazed but very soon we realised that it was more of a shock than a pleasant surprise based on the twitter posts from the community received in the next few hours.
We are now generating art with stable diffusion models, in the area of computer vision this is more of an aspirational pursuit to digitally create art form, similar to the work of some of the famous artists. Of course there was a lot of hue and cry about this from the art community, where artists claimed that years of their effort is being replicated in a few seconds of running code. They now have an option to opt their names from being used as an input prompt into these generative models.
But generating scientific material which looks factual – that’s another ball game altogether. Over the past 2 years I have been in the deep of my research journey, during the course of the Phd program which I am currently pursuing, I got an opportunity to think very deeply about many interdisciplinary areas of management science. I could appreciate the nuances of the field where an elegantly written scientific paper stands out from one that looks like a collection of incoherent facts, past literature citings along with a multitude of statistical methods forced onto the research dataset. These elegant papers are what we refer to as “seminal” works and these might be written decades ago but are so solid in their foundation that their influence will never cease to exist. Whenever a student of the specific field will get to know the area , he/she will benefit from them. I have to give credit to the existence of this scientific elegance as the big reason why I have been enjoying my research journey so much.
Fathom a ML model that can ingest millions of pervious papers with the help of massive computational investments and algorithmic advancements in NLP and then generate scientific material that gives a semblance of genuine authority in the writing form. Unlike digital art form, this area of work cannot be replicated so long as the scientific papers still represent the most complex depths of human thinking ( watch the movie "A Beautiful Mind" to get a gist) or until the time when the machine will become far more superior to humans in all aspects of intelligence far beyond the tasks of performing massive computations and data processing. So while we can have fun playing around with these tools given the resources to do so, we need to be more responsible before unleashing a potential source of chaos and confusion. Imagine a scholar who gets misguided by seemingly real work which is not fully accurate. Enter the world of “Responsible AI”, what better example to see the significance of making it a top priority for not just the CEOs and Top Management Teams but everyone who is a part of the world of AI as a creator or a user. While we must give due credit to the efforts and progress made by the scientists at Meta we must not hurry to reap the benefits of AI without complete understanding of the intended and unintended consequences of the technology.
Comments