AI for a reason

When Shaastra magazine was started up four years ago, it was looking to fill a void in Indian publishing. Despite an expanding research ecosystem, India didn't have a science magazine produced by professional journalists – a magazine that could reflect the changes in the country, with a global perspective. At a basic level, Shaastra was intended as a vehicle for communicating the fruits of research to other players in the ecosystem, such as start-up founders, investors, researchers in other fields, and students. At a broader level, Shaastra was also a platform for initiating conversations in the country on how science and technology are shaping society, and how it can continue to do that.

These objectives determine what we cover in every issue, within the limitations of space and time. The primary aim of the magazine is to report, without judgement, on research and development in the country's labs and technology companies. When we pick stories for an issue, we often choose those that have not been covered – or have been inadequately covered – in other publications in the country. As we have said in this space on earlier occasions, the magazine looks to stay away from 'fashionable' topics – which other publications pick up. But we do that only up to a degree.

Artificial intelligence (AI) is a case in point. It is currently one of the most discussed topics in publications around the world, and a key enabler in almost all domains of human activity. There has been plenty of reporting and discussion on the topic in Indian and overseas publications – not to speak of lectures and discussions on YouTube. All of this is being done even as new developments unfold in the field. And yet, no publication can stay away from these momentous developments, even at the risk of repetition. So, despite this deluge, Shaastra has taken up coverage of AI. However, Shaastra distinguishes itself by venturing into the deeper aspects of the AI revolution: the trends that take long to be evident, and the sectoral technologies that AI influences. The deeply philosophical, combined with the intensely utilitarian.

This month's Cover Story deals with a major problem in AI: its lack of ability to reason. While AI can do remarkably well in some areas, like picking patterns in data or defeating the world chess champion, it is poor in forming judgements logically, which human beings do with ease. AI is also a black box at the moment. Engineers do not quite know how it comes up with its results, but it is evidently not through reasoning of the kind human beings perform. Large language models (LLMs), therefore, seem both intelligent and stupid at once, like someone who can solve complex differential equations but not perform simple addition. AI can be tricked into giving wrong answers in problems that high school students easily solve.

LLMs are a new path of development in AI, a big shift from the deep problems that it was being trained to solve until the 1990s. Early models of AI were being developed to solve problems through understanding, just the way humans do. When the internet generated a lot of data in the 2010s, a movement started among researchers to train AI on data and come up with answers using queries in natural language. This was quite unlike the work of human beings, who do not solve problems by reading millions of documents. The drawback of this approach is now evident to researchers. Sweta Akundi's story, from Teaching AI to reason, is about another imminent paradigm shift: of getting AI to solve problems through reason.

Stories on AI are quite different from our regular fare. In all our science stories and most of our technology stories, we describe the mechanisms behind the development that we report. For science, this means explaining the principles behind the new discovery or technology, no matter how complex, at a level that a 12th-grader can understand. This is hard in AI because the mechanisms are not known in many cases. So, in AI, we focus on the reasons that drive the researchers – rather than how the technology works. This Cover Story is, therefore, as much about the motivations of AI researchers as it is about the technology. Human motivation is a strong element of all of the stories that Shaastra takes up.

See also:

Teaching AI to reason
AI models for India
The open-source edge in AI