Perhaps the BEST B1G Forum anywhere, here at College Football Fan Site, CFB51!!!

One area that I wonder about is liability?

I think about this with autonomous driving. For autonomous driving to REALLY be here, it has to mean that the vehicle is capable of driving me and I have no legal liability for what it does. I.e. I can be passed out drunk in the front left seat of the car and if it mows down a row of nuns on their way to mass, it's the car manufacturer's problem, not mine. Which means that the autonomous system doesn't have to be merely "better than a human", but it has to be so much better than the human that the human is no longer needed.

I wonder what the risk of misdiagnosis is when it comes to things like AI-read medical imaging. Obviously if AI says "that's a tumor" it'll get referred to a human doctor because you're going to have to treat the tumor. But what if AI says "nope, there's no tumor there!" and then you die of cancer because it was, actually, a tumor. Can your family sue the medical group for malpractice because they farmed your diagnosis out to AI? Can they sue the AI medical company because their faulty algorithm didn't get your diagnosis right?

These are some of those areas where I wonder how it's all going to be figured out.

As I say, the business world is murky, complicated, and treacherous, and I won't hazard any guesses as to how it will go legally or in the insurance industry.  

Speaking somewhat knowledgeably from working in the medical field for a while and being married to a provider, and from knowing a little about AI--more so on the analytics side, but at least a working knowledge of the generative side--I think the technological side has a clear path.  Just my opinion.  

The average radiologist has a stellar hit/miss rate when reading images.  A good ML model that stays in its lane is already better at some things.  Radiologists only misdiagnose a tiny fraction of images.  AI only misdiagnoses a fraction of their fraction.  

Back to the precision/recall curve, if you train a model where the graph of the curve (for the life of me I can't remember what it's called) hugs the axes, you can really go hard on fine-tuning the model to weed out false negatives.  Sure, false positives are their own problem, but 1) those can be passed on for human inspection, and 2) it's better than letting false negatives slip by.  Legally, sure, maybe there will always be problems.  But logically, AI already has a better hit/miss ratio, and the reality is any positive it misses, statistically speaking, would've almost certainly have been missed by a human as well.  

Something to remember in all this is what I said earlier, the model's performance is measured against ground truth (like all models).  But in this case ground truth is an amalgamation of human ability.  So the model is compared relatively to individual radiologists' performance, but it's objectively compared to a cumulative sum of radiologists' expertise.....and that's different than an algorithm that identifies cats or dogs, where the labels are understood to be foolproof.  That is what I think is likely to complicate legal matters for potentially years.  I'm just guessing.  

Yet another legal ramification to consider is; how do lawsuits fare concerning not using AI?  Some of these AI models I looked into could spot pre-cancerous areas like a boss.  As in, it's not even cancer yet, and no radiologist would've ever said "we need to keep an eye on you over the next 5 years," but AI found it and patients were able to start life-saving preventative treatment.  And no one even knows what it's looking at.  The model is so black-box that nobody knows how it figured out the problem spot that would develop cancer.  Obviously, doctors have understood pre-cancerous areas for a long time and can even spot them sometimes.  But not even close to the level at which AI can do it.  

The legal/moral/business implications may wind up swinging both ways.  

"The Moon is a Harsh Mistress"...

One of my favorite books

First off, I'm pleasantly surprised by how many of us are well versed enough to speak in depth on Artificial Intelligence. Many of you (texas guys, mike, bw) have the analytical side down, whereas I'm inputting more from the philosophical side.

In summary, my mind goes to two questions when contemplating this matter:

1) Is it ultimately possible for GAI to reach a point of sentience, of self-wareness, of independent thought?  It's as much a philosophical and spiritual question, as it is a scientific one.  Can an artificial system gain true sentience, and if such a thing as a soul does exist, is it possible for an artificial system to develop one?  And if not, is that potentially problematic?

I believe Artificial General Intelligence (AGI) will be achieved sooner than we realize, potentially within 2-5 years, given the rates by which Narrow AI programs are advancing. That means an artificial sentience, self-awareness, independently thinking, and to introduce another word - consciousness - in 2-5 years. I believe this other, nonhuman yet human-capable consciousness will be stumbled upon, created accidentally, and might even exist for a while before developers realize that they have, in fact, created another consciousness.

I think of its discovery like Cristopher Columbus sailing for India. He had the technology (a ship), enough expertise, and the drive to give it a shot crossing the seas. But what was ultimately discovered was quite unexpected (the Americas). We have the technology (quickly advancing Narrow AI) and the drive (arms race) to give it a shot. We expect to discover a separate though alike consciousness, yet what we more likely stumble upon will be much more alien than we could've expected (more on all this in a later post).

But why, from a philosophical standpoint, do I believe we’re closer to (accidentally) achieving an Artificial General Intelligence breakthrough? For centuries the concept of consciousness has cornerstoned one of Philosophy’s more subjective debates – what exactly is it? How does it emerge in humans? Consciousness has never been rigorously or definitively defined in philosophical or scientific terms. Yet, from both standpoints there is general consensus that human consciousness is inexorably tied to Language, to the point that the Existentialist, Martin Heidegger, famously stated: “Language is the house of being. In its home human beings dwell. Those who think and those who create with words are the guardians of this home. Their guardianship accomplishes the manifestation of Being insofar as they bring this manifestation to language and preserve it in language through their saying.” (Heidegger’s House Of Language)

A painfully theoretical statement indeed, but to simplify—
Heidegger is saying that we live inside our language systems in order to live at all; and we conform to our environments architected by our language systems. Theoretically, language does not come into the world so much as the world comes into language. Language is our footing (our consciousness) in the world.

How does this practically play out? Look at the medical world’s highly complex vocabulary. This exclusive vocabulary is how medical professionals comprehend, navigate, and live in their worlds – through dedicated language systems. To live in this world, Heidegger’s idea of Being conforms to this world through its language.

Now, with all that said, and proposing our footing with Language as a form consciousness, I’ll quote Bret Weinstein, a professor of evolutionary biology, speaking to one of his concerns about ChatGPT:

“…when we say ‘well Chat GPT doesn’t know what it’s saying'…because it’s not programmed to have a consciousness…we are actually ignoring the other half of the story which is that we don’t know how human consciousness works and we don’t know how it develops in a child…”

“…a child is exposed to world of adults talking around them…the child experiments first with phonemes and then words and then clusters of words and then sentences and, by doing something that isn’t all that far from what Chat GPT is doing, it ends up becoming a conscious individual…”

“…Chat GPT isn’t conscious, but it isn’t clear…that we are not suddenly stepping onto a process that produces [consciousness] very quickly without us even necessarily knowing it…”

Good stuff @CatsbyAZ ... A few points...

Agree 100% that we don't understand consciousness, nor intelligence. That of COURSE means that it's quite possible that the first time it arises in machine form that we may not realize what has been created, because all we can really understand is what the outputs are.

That said, I'm not sure that I agree with the professor of evolutionary biology, because while he's obviously smart in his field, it's not a field with a lot of overlap with computer science. And so I might quibble with the idea that an LLM will "slip its leash" and become AGI. We can try to create a machine to replicate one highly specific--and highly important--aspects of human intelligence: language. But there's a LOT more going on in the brain than language, stuff that we don't understand how and why it works, and it's not something we're building into that machine intelligence because we don't have a clue what it is.

Things like ChatGPT grab our attention precisely because it's so good at something we didn't think it could be good at. That because it can do so well with something as incredibly complex such as language, it's just one tiny little step from there to AGI. I personally think that step is a lot bigger than most people think.

Again going back to the things like autonomous driving, such as Tesla's Autopilot. I think what that system is capable of is truly amazing. That with what they've put together, they're maybe 95% of the way to being able to drive. But... That last 5% is a REAL bitch. Going from 0 to 95% is hard but doable. Going from 95% to 99% might be 5-10x as hard. Going from 99% to 100% might be 100x as hard as THAT.

We're incredibly impressed by what ChatGPT or Autopilot is capable of, because it's really damn impressive. But we look at the capability and don't really understand that to go from there to sentience/AGI or to autonomy is perhaps orders of magnitude beyond.

Agree with that bolded in red.  Which is why I raised my two questions above.

But it's also why I put aside Question #1 as not being important from a practical standpoint.  It is of course incredibly interesting from a philosophical and spiritual view, but practically I think we're going to have systems that successfully emulate independent thought, and I don't think it's incredibly far off.

And in the worst case, the AI LLMs trained on the entire web, have a tendency to "hallucinate."  They create completely false facts when queried on a subject.  The false facts are usually plausible, but they're absolutely 100% fabricated, because the training data has so much ambiguity and incorrect data within it.

This is what I found when asking about specific things.  What creeped me the hell out is that when I called it out, it doubled down once before acknowledging it basically didn't have enough data to form a valid opinion.

I get the error.  I do NOT get the doubling down when called out.

I get the error.  I do NOT get the doubling down when called out.

Are you sure you weren't just posting here? When you might've thought you were messaging in ChatGPT?

It’s kinda like that scene in Terminator 2, when the main scientist is talking about the chip they found, how evolved it was. He exclaims, while simultaneously realizing what it really meant, that it had things in it “ that we would’ve never thought of” while his voice trails off.