We are the world, we are the sensors.

Typical Artificial Intelligence (AI) approaches model the data they are fed with. Data is what we can measure and store. From these two statements, we can say that AI learns from measurable (quantitative) inputs only. Naturally, this will limit its capability to be as intelligent as humans, because we-humans learn from measurable (quantitative, eg temperature, prices, etc.) as well as non-measurable (qualitative, e.g. emotions, culture, etc.) inputs. Thus HI (Human Intelligence) will always be superior to AI and then full-blown AGI (Artificial General Intelligence, mimicking full HI) would be impossible. Well, that's what I thought.

But reflecting more on this, now thinking that this may not be entirely true.

In finance, there is Efficient Market Hypothesis (EMH), which states that price of an asset incorporates all the information available at that point. Meaning, not just quantitative information such as profits-loss, economic indicators etc., but also qualitative information such as crowd sentiment, political environment, etc.

Taking clue from this, can we say that even the data that is fed to AI, also incorporates all the information, ie both, quantitative as well as qualitative? Even when the data appears numeric (quantitative), as seen in EMH, it may have qualitative information incorporated in it. Looking at similar example, if we are building stock price prediction model, the price time series, has all the information in it, even political sentiment, cultural biases, etc, everything!! Another example could be, if you are building tweets sentiment prediction model, the texts has all the information. The ability to extract depends on the language model (embeddings) and AI architectures employed. So its just a matter of time that both of them become so powerful (like GPT *) they can lead us to AGI like state.

So, we can say that, at least some form of data has all the information. Especially the supervised data. Annotations, which typically humans do, will inappropriate both quantitative as well as qualitative information in it. Person's world view gets embedded in the label, when one says that this tweet is Negative, even though none of the individual words indicate so!! Thus, we humans have become sensors, sensing the world around us and funneling it into the data.

As data gets richer and AI models get more powerful, looks like AGI would not be far away!!