Futuristic Super Human Large Language Models

The very term ' machine intelligence ' is premised on the limits of human potential since machines are created by human beings. If machines could think for themselves, without having to be pre-programmed by humans, one can safely assume that artificial intelligence is coming of age - that there is the possibility of their having common sense, to make decisions outside of their packaged abilities. But then, large language models (LLMs) such as ChatGPT and other such generative AI platforms have not so far evidenced independent, volitional thought processes or actions. They are as yet mere tools, functioning as per human design patterns and inputs. It follows therefore that for truly creative writers, pre-programmed AI tools like ChatGPT are hindrances rather than enablers as they cramp imagination, limiting the writer's infinite creativity.

If and when the day comes when machines can think for themselves independent of human inputs, only then can we conclude that yes, we have a huge threat or wonderful ally on our hands, depending on the perspective from which you view LLMs, for instance. And depending on the extent to which you can exercise some sort of control so that AI does not go off the rails. It will also open up the debate as to whether machines can have consciousness, for, to be self-aware and ruminate on the nature of the higher Self, so far, one needs to be human with the ability to carry out this kind of thought process.

At best, generative AI could serve the purpose of guidebooks that students resort to on the eve of examinations, with all the risks of rote learning, and inbuilt errors.

Autocorrect, a boon for quick reviews and proof-reading, can also completely change the context and meaning of a sentence or phrase when spell checks are done without common sense. This is how causal is "corrected" to casual and Hare Krishn becomes Hate Krishn! Hence relying on generative AI completely when it is still functioning on human inputs, even if it has some amount of deductive ability, can be a slippery slope.

Can AI be engineered to the extent that it can have its own common sense? In other words, can AI be endowed with the power of discrimination? Can it deal with abstract scenarios? Human cognitive ability, intuition, emotion and imagination are attributes that are currently absent in AI including LLMs. So, AI response is always constrained by facts, figures and cold logic rather than reasoning and common sense. A choreographed, deliberate response or reaction to a situation is not always the optimal way to deal with it; more human qualities like self-awareness, reflection, reasoning, imagination and compassion can play far more crucial roles.

AI tools may not have social skills, so there are several limitations holding back AI from replicating human capabilities. But who knows, soon enough, not only might AI become more human-like; they may even become our pathfinders. As Nature magazine says in its report: "In the long-term, perhaps the biggest contribution of the science of machine common sense will be to allow humans to understand themselves more deeply." Geoffrey Hinton, awarded the 2024 Nobel Prize in Physics for his work on AI, says: "The idea that this stuff (AI) could actually get smarter than people...I thought was a long way off...but I no longer think that."

Authored by: Narayani Ganesh
ganeshnarayani@yahoo.com