The Failure Of Big Data

The assumption that "more" data will produce "more" human like language is wrong. Totally.

ELT Pros Linkedin | Videos | Blog | Printables | ELT News | TpTs | Youtube

This post is a continuation of my previous thoughts about ChatGPT and large language models, generative text technology.

I think it is time many of us get honest about all the “buzz” surrounding ChatGPT 3 and AI in general. It’s like the first week of dating is over and it’s time to really evaluate, get your head out of the clouds and look deeply and honestly into her/its eyes.

ChatGPT when you undress it - is really just an information retrieval system but with a prettier front end than a search engine.

The video above is an excerpt from a discussion on AI and why it is a kind of “lie”. In this piece, Gary Marcus calls out ChatGPT for what it is, a fraud. He basically takes the line and argument of all “Man vs Machine” books and movies. That machines don’t have soul, innate, pre-conditioned humanity and thus, no matter how much information and data, they will never achieve “intelligence” and human like communication and language.

For a long time, I was convinced that consciousness, the “ghost in our machine”, our soul, what makes us alive - I thought consciousness was a function of information. That perhaps with enough data, enough processing of information, consciousness would arise out of the mist. Thus, even a robot or a hard drive could theoretically become “human” and have conscious thoughts, a conscious, aware mind.

I don’t think so any longer. I’ve long been influenced by the work of Rupert Sheldrake and his belief that we aren’t “filled” with information but rather, we tap into information out there in the wide universe. See his fascinating talk about “scientism” and how the modern foundations of science are too mechanical, too gross and fantastical. Myths. So too, our thought that machines will soon, with enough data, be able to sound, act, write, think like a human being.

As the talk alludes to - language is very much “conventional” driven. So, it isn’t too hard for machines with some data, examples and training to pick up on the bland, regular, recursive, formulaic part of language and run with it. But, and here’s the rub, it is that 25% that is poetic, random, creative, uncopyiable that is so important to language.

There are a lot of concerns about ChatGPT and AI in particular. You probably have heard or thought of many of them. Destroying jobs, plagarism, the veracity of information/knowledge (fakery), the erosion of authenticity and ultimately machines becoming human-like - us, not being able to tell the difference.

I think many of these fears are unfounded. Sure, like the talk suggests, we’ll get 75% there but that isn’t anywhere near enough for these problems to be overwhelming.

In my former company, we were always sprouting how our technology was “big data” and had trained on “millions of utterances” etc … But it was all a lie - a lie, we spoke as a truth. We’d go to conferences, publish papers and people were tricked that the tech was doing a great job, when in fact, 75% of the time, it failed. But like our President said, it’s not about the product, it’s about the bow. And that’s how it is with most of AI, it’s 75% there and this seems good enough to fool and wow much of the users and world. It shouldn’t fool us. When you throw back the curtain, there is no Oz, Dorthy.

There is a big “lie” surrounding ChatGPT and I think we need to expose it. It’s not thinking. It is 90% artificial and 10% intelligence. It makes no multilayered metaphorical connections - what we associate learning with. However, the science/tech hype machine keeps marching forward like stalward stormtroppers shouting “the future is now”. It’s all marketing, sorry to say. Yes, some great applications in many domains but as concerns the replacement of humans on all levels, pure hype.

And what of literature, my domain? What will happen to “the word”, in a world of improved ChatGPT?

I’m reminded of a Borges essay about all books originating from one author. Linguistic pantheism. All writers are tapping into the same stream of data, information. Shelly expressed the same opinion that all poems are episodes, fragments of a single, infinite poem, written by all the poets of earth. I kind of like this idea. But it is scary too because ChatGPT ultimately could be that kind of universal writer-god. Accessing all the combinations of words possible and endlessly, effortlessly (except for the heat generated from all the 1,000s of server farms - those quiet, modern libraries), ChatGPT could write anything, everything in an infinite number of forms.

I raise this point to bring up my major concern. It is that we learn to use ChatGPT and technology and not have it use us. It should not be the first option in many cases but rather a tool to serve a purpose. What purposes? Well, we are still figuring that out but its important we set limits and also set those limits with the knowledge that ChatGPT isn’t “intelligent” however you may define the word. It is “artificially” intelligent - it doesn’t and never will pass go and bleed and cry and sing and dance crazily, madly.

Yesterday, a teacher replied on social media to a post of mine about ChatGPT. He wrote,

I used it in one of my classes. A student suddenly asked me to have a hotel dialogue. A teacher could say, let's do it, next class. However, having this tool eased the process.

I wonder why he didn’t just roleplay right there with the student? Why the need of an artificial text generated by a machine, first?

And that’s the danger. For every little thing, we need to run it through the IRS (not that one - I mean what ChatGPT really is, naked with only its underwear on, an Information Retrieval System). And doing that more and more and more will make our world more vanilla, bland, without flavor or soul or humanity.

That’s my concern. Yes, great tool. But like anything learn when and how to use it. You don’t chop onions with a chainsaw …

From Where Comes The Essence Of Anything?

What beauty in the brilliant sun!
Or is that beauty in my eye?
Or perhaps somewhere between,
Drifting with indifferent sky.

How sweet the sound of Orpheus’ lyre!
Or is that sweetness in my ear?
Or perhaps somewhere between,
Hidden by the wind unseen.

What evil oozes from the witch!
Or is that evil in my mind?
Or perhaps somewhere between,
Where witch and I are of the same kind.

From where comes the essence of anything?
Does it come from the self as the doctors say?
Does it come from the thing as the natives pray?
From where comes the essence of anything?

David Deubelbeiss