As the New Year 2024 dawns, AI has become a part of our lives and minds. At times it is frightening to read about all the changes it is going to bring. Just search for ‘future of AI’ and the material that comes up is mind boggling. Everything that touches our everyday lives is going to change exponentially. While it changes our lives, the technology of and research on AI is transforming too.

A neural network is the internal engine of all artificial intelligence technologies. The neural network is said to be based on the way the human brain functions. However, currently, the human brain remains a much superior form of intelligence. Human brains apart from being intelligent have awareness and imagination, which has ensured their rise to predominance amongst all species on the planet!

The present AI neural networks are trained on billions of parameters, which require months on powerful servers with hundreds of thousands of processors. The size of processors, the large amount of data and the application possibility of an AI model are the areas that are likely to see research in the next few years.

As per analysis, the deluxe version of ChatGPT needed more than 16000 specialized chips and took many weeks to train, at a cost of nearly $100 million. Computers also started big, but now fit in the palm of human beings. That is the curve that AI is also aiming for. The effort now is for smaller and faster models. Google’s Chinchilla is quarter the size of ChatGPT, faster and trained on four times the data. The aim is to make models that could run on a laptop or a phone.

Large of amounts of data are needed to train AI models. So, the research is now shifting from the quantity of data to the quality of data that a model is trained on. As per analysis the stock of new high-quality text may dry up in the next few years! AI models are being increasingly trained on mix of data types – text, videos, images and computer codes, for better capabilities.

The current rage is learning to use proper ‘prompts’ to get accurate results from AI. Now models are being fine-tuned to respond to focused data sets, like using AI to respond to mental health issues. Combining AI with some software, which first searches relevant data bases, thereafter, supplies the AI with the material to answer, is also gaining ground.

As AI models improve, the research in this field is aiming to make AI function like the human brain. Called ‘neuromorphic’ processing, the circuits would run simultaneously, not sequentially like they do currently, increasing their cognitive capacity.

As all these developments race ahead, the need for AI regulation through bodies competent has become a matter of discussion around the world. New technologies like aviation and nuclear energy saw the emergence of an entire new set of regulation. The new way of life looming in our collective futures, driven by AI, would also need a new comprehensive regulatory framework.

Published in Dainik Bhaskar on 9 January 2024.