In my short story AI.GOV, I posit – no intelligence will ever surpass that of its creator. ChatGPT, and other generative AIs, have not been able to change my opinion yet. The hue & cry and pleas to form some legislation to limit or ban further development in AI has more to do with corporate politics and ‘land grabbing’ than to do with perceived threats from it.
These LLMs (Larger Language Models) appear brilliant and, as has been demonstrated, can easily ace various tests, like SAT, and bar exam. But I say, they better perform well on these tests because they have been trained on all this data! It is equivalent to a person studying hard, with the help of one or more tutors, for an exam and then acing it. The irony is that, in spite of being trained on all of the humanity’s content, they still show dismal performance in other similar tests.
Consider the following to understand my position:
- AI is and continues to be trained (i.e., it learns from) on human generated content. As its use grows, the quality of content it learns from and generates will decline rapidly. My prediction is that the degradation will happen faster than we think. This will be true for monolithic and consumer-oriented AI models like ChatGPT
- To have an apple-to-apple comparison between AI and a human, we should compare ‘unit of intelligence’; i.e., AI running on a single machine that is not connected to the cloud computing infrastructure against one human with both exposed to the same set of information (learning dataset)
- We humans are not sitting duck like a deer-in-a-headlight. We have our own trajectory of increasing knowledge and competence
- AI, in fact, will be a catalyst and an inflection point in accelerating personalized education that will unleash human creativity at unforeseen scale
When people state their fear of AI outsmarting all of us, they are comparing one individual pitted against massive compute power with access to all of content available on Internet for the machine to learn from. This is an unfair comparison.
In one of my posts on LinkedIn, I said that until AI can traverse the journey from Information -> Knowledge -> Experience -> Intuition we do not have to worry about it taking over. It is still at the first step of ‘processing information’. Let’s see how quickly it traverses this journey.