-artificial intelligence is software that performs actions broadly defined as “intelligent”
-artificial general intelligence is software that is intelligent in every capacity that any given human is, and is said to require consciousness and human psychology to count
-artificial super intelligence is software that is beyond human intelligence in all respects, it’s said to mark the technological singularity and to have exponentially streamlined understanding through self modification that’s more than a team of human geniuses could ever comprehend
no such software under the last two terms exists, language models are filled with data, but none of them are rational like we are. they’re just very inefficient expert systems
It’s nice to neatly categorise these concepts, but it’s not necessarily true. Artificial intelligence has been used to mean both the practical concept of a machine/algorithm that could do what would ordinarily require some human intelligence and the sci-fi concept of an artificial entity capable of comprehension and learning, for around the three decades before LLMs. Moreover, the first use of AGI corresponds to what we’d now call ASI, that’s how I understand it at least. Some also seem to have distinguished weak vs strong AI, corresponding respectively to the current use of AI and AGI.
I knew people working on the precursors of the current models done years back (pre-2020) and they called it machine learning and neural networking. They only ever used “AI” as a hype thing. I’m not saying nobody in the academia called it AI but it’s quite clear that today’s stochastic models are called AI as marketing.
they called it machine learning and neural networking
absolutely, i remember that being the vogue c. 2015 when they were touted as breaking the “ai winter” and to be honest i’d still call these llms under those terms if pressed but i don’t keep up with computer science any more
to clarify some terms:
-artificial intelligence is software that performs actions broadly defined as “intelligent”
-artificial general intelligence is software that is intelligent in every capacity that any given human is, and is said to require consciousness and human psychology to count
-artificial super intelligence is software that is beyond human intelligence in all respects, it’s said to mark the technological singularity and to have exponentially streamlined understanding through self modification that’s more than a team of human geniuses could ever comprehend
no such software under the last two terms exists, language models are filled with data, but none of them are rational like we are. they’re just very inefficient expert systems
It’s nice to neatly categorise these concepts, but it’s not necessarily true. Artificial intelligence has been used to mean both the practical concept of a machine/algorithm that could do what would ordinarily require some human intelligence and the sci-fi concept of an artificial entity capable of comprehension and learning, for around the three decades before LLMs. Moreover, the first use of AGI corresponds to what we’d now call ASI, that’s how I understand it at least. Some also seem to have distinguished weak vs strong AI, corresponding respectively to the current use of AI and AGI.
I knew people working on the precursors of the current models done years back (pre-2020) and they called it machine learning and neural networking. They only ever used “AI” as a hype thing. I’m not saying nobody in the academia called it AI but it’s quite clear that today’s stochastic models are called AI as marketing.
absolutely, i remember that being the vogue c. 2015 when they were touted as breaking the “ai winter” and to be honest i’d still call these llms under those terms if pressed but i don’t keep up with computer science any more