Given everything we know about AI so far, what do you think it can or cannot be good for?
AI has to be among the largest and most devisive things in the world right now. What's more, I suspect that tension will continue to grow in the time to come. It's unavoidable that we'll see that same tension reflected here in our small corner of the internet.
So I think maybe we could take stock of where we believe the boundary of AI capability lies. In the spirit of bridge building, I also encourage that you try to think about the answer from the "other side" point of view as well.
If you're a "sceptic", what do you think AI can still be good for, and what sort of changes to its current paradigm (technical, legal, social, etc.) would allow you to entertain its use?
If you're a "believer", where do you think AI would likely still fall short (even accounting for future improvement), and where the news cycles are likely overestimating its capability?
Edit: "capability" includes potential impact as well (like whether it could actually put a significant portion of human out of jobs, etc).
First off, I'm not a sceptic or a believer, as "AI" is too broad a domain to just have one thought about it unless you're a fanatic one way or the other.
I think that for data sorting and discovery it can be great, especially when there's checks and balances to verify the outcome amd it's working with a stable data set that isn't trying to incorporate breaking news. "Review the files of this business that closed 12 years ago and give me the names of the employees and the document you used for that info." turns hours of manually searching through misfiled and poorly named files into a 3 minute task to confirm the findings. But when the data is in flux or it's largely based off of things that confident idiots said online it can spit out hallucinations and wrong information with great confidence. It's not a universally good tool.
For generative tasks, I think it does a decent job at spitting out pablum. Generic pop songs, overly shiny and complicated images, trite stories that don't go anywhere. But that's a trap. It only feeds on what humans make, and if we make it impossible for humans to make a living making art then we'll just have incestuous generations of ai generated content that feeds on itself. Beyond that, I enjoy art because it's saying something to me. It's an expression of another person, conveying something beyond the words, sounds, or images themselves. Generative art is like clip art or emoji. Maybe it's got a single big statement: "be afraid!" "be horny!" "spend money!" "OBEY!" but that's not an interesting level of art to engage with.
Just wanted to add how it amplifies confirmation bias since it will always try not to hurt your feelings and agree with you more often than not.
This is one of those interesting nuances in the “AI” conversation: that’s true of the commercially provided models as configured, not necessarily of the technology itself. RLHF does naturally bias towards certain styles of response in a lot of cases, but even that can be overridden if there’s appetite to do so.
It’s another analogy to the internet itself: the capabilities that actually come from a planet-wide network of billions of interconnected devices are incredible, you can see the depth if you start considering all the ways society would collapse if “get information from A to B quickly without caring where A or B are” went away, but the dominant experience for most people is Reddit, Meta, Amazon, Google, TokTok, and maybe one or two major media conglomerates.
The internet (comms layer) and “the internet” (5-10 major web platforms) are both globally important in different ways, as is the fact that a handful of companies do control so much of it is another important factor in and of itself.
Same goes for AI (imprecise label for a rapidly advancing field of machine learning, usually covering LLMs and often covering other deep neural net developments too) and “AI” (3-4 major LLM providers). The massive providers are shaping the conversation and the economy and it’d be absurd to ignore that, but the tech itself has, is, and will be transformative (heh) independently of them too.
I think that's what everyone enjoys about art, even if they haven't been able to identify that. I think that's why "ai art" as a medium is doomed to failure.
There's a reason why the most technically talented musicians aren't necessarily the ones that become famous.
The ones who become famous are generally interesting and captivating as people first, and talented musicians second. People want to know why the Red Hot Chili Peppers wrote under the bridge, or what Madonna meant by "like a virgin" (shout out to Tarantino). Sites like genius exist solely for people to think about song lyrics and debate what they actually mean. The context and the people's lives behind the art is important to everyone who consumed it.
When the answer becomes "that's what the algorithm was trained on" the whole thing becomes a complete waste of time. No one is paying for a movie ticket to see something no human had a hand in making. It would feels like just watching tv static. Totally pointless noise.
As a person generally, I try to understand both sides of every argument, regardless of my final opinion. I've thought about the pros and cons of AI as we know it and I've boiled it down to a few thoughts
The Good: incredible pattern recognition, and the ability to do non-analytical maths will make our understanding of science and math incredible. The hardest part will be reverse-engineering and proving AI solutions are valid.
The bad: LLMs are not tools of creation, they are just a systemic average of its trained materials. It has an explicit reason it was created (the prompt), but it does nothing to convey the artist's component of the creation.
I disagree with the take that generative AI can’t create new art. I’ve spent enough time experimenting with diffusion tools and techniques to know that it absolutely can, in the hands of someone with a creative vision and an understanding of the tech. Because while models have been trained on existing art, and tend to seek a bland average of those weights, they can also be wrangled to synthesize entirely new constructs like nothing in their training data… because creativity doesn’t exist in a vacuum, it emerges from combining and subverting that which came before. Everything’s a remix.
Of course, the problem is the slop. We’re just completely swamped in bottom-of-the-barrel, low-effort garbage. I’m optimistic that after the novelty wears off, the deluge of slop will subside (somewhat). It may never go away. But I think the aesthetic we all associate with “AI art” today will go the way of those hyper-saturated tone-mapped HDR photographs that were all over Flickr 20 years ago. And in its wake we may see some new digital art movements that are really deliberately transformative and human-guided at every step. That’s my hope anyway. It’s hard to be optimistic about the future these days but I’m trying!
One novel use for AI that I’m actively researching is how to use it to turn low-quality book scans into high-quality accessible e-texts. If multimodal LLMs are able to perceptually interpret the semantic structure of an image of a page, they can reconstruct it as properly tagged document. I’m building a self-correcting diffusion+LLM loop to restore broken or washed-out glyphs in particularly bad scans, to infer he correct missing information that was never present in the original pixels. Much better than traditional OCR. I think there’s a ton of untapped potential in this application of AI and it could substantially improve the state of alternative formats for students with visual impairments or learning disabilities.
An artist can use an LLM to make art the way I can make a weedwacker make art.
That doesn't mean the weedwacker is making art.
Seeing the level and the rate of progress we've had overall, I actually think that (somewhat ironically) the reason for the "slop" is because companies quickly started censoring the fuck out of their image models. Both because people started complaining about copyright issues (personally I'm not sure an author should be able to retroactively prevent any inventive future uses of something they publicly posted but whatever) and of course the ever present porn scare. Plus I would guess that because of those scolding hot potatoes it just became less of a priority to advance in this direction in general.
That said, of course lots of the training data being slop (which amateur artists having putting ALL OVER the internet since the dawn of time but we don't talk about that) and the difficulty of evaluating what "good art" is (which is pretty much necessary for training purposes to set a goal) did not help either.
I'm actually curious how well do frontier models deal with creative text tasks nowadays.
Anyway I'm sure that we will reach the point where people finally give up on detecting AI and instead just evaluate quality itself. And possibly quite soon.
Regarding detecting AI, I happen to think we'll turn the other way. In some sense we've been here before. When writing first came about, it was mainly for recording facts and events, or at least what the writers believed as true and had happened. Then someone had the idea to completely make everything up, and the worst part was it sold! Humanity didn't collapse into a post-truth society, and we didn't come up with any sophisticated truth checking methods either. The solution was more or less a cultural one, where we collectively decide some writings are now categorized as 'fiction', with both writers and readers largely adhere to convention. Something similar could happen for our relationship with AI.
It's interesting that there are cases where people got upset that something advertised as AI-generated turned out to have a director carefully control the output. I think people care about not being deceived more than the actual quality.
"quality" in art isn't really that useful or meaningful of a descriptor.
The main reason people enjoy art is because it's communication between one human to another. If there's a good chance that no actual human thought was put into a piece of art, it automatically becomes worth very, very much less.
It's kind of like high end watches. On a technical basis, the $60 dollar quartz Timex in wearing right now is superior in most ways to a $200,000 patek phillipe. It tells time a lot better, it doesn't have to be wound up, it's more waterproof than most models, more durable, and a lot more reliable.
People like mechanical watches because a lot of human thought, skill, and creativity went into making them. They're low volume, hand assembled, custom designed objects made by a master craftsman. My watch was made in a factory by the millions by machines.
Value is subjective, and it's even more subjective when it comes to visual art.
As a semi-watch guy, the example might end up proving the opposite. There's some hardcore fans of G-Shocks, Timex, Swatch, etc. I fell in love with G-Shock last year coming from mechanicals and it's all I wear 99% of the time now. Within the mechanical watch world, the majority are cranked out by the millions in a factory at the low end (Orient), medium end, and the relatively high end (see Rolex). But they're all still loved and can sustain an artificial scarcity model at the high end. Sure, at the highest end, maybe they're handmade, but the supermajority of the market of mechanical watches are not handmade.
I'll share my 2¢ on a topic I've been thinking a lot about, informed by both my personal and professional experiences.
I'm optimistic about LLMs as learning tools, pessimistic about the current attempts at productizing this, and very doubtful that we'll see some miraculous revolution in education because of the technology. (I'm also concerned about how well institutions and students themselves are adapting to cheating with AI, but that's a different conversation)
I've had some fantastic experiences using LLMs myself as a learning aid both in topics I already have a background in and topics relatively new to me. When asking for explanations of topics heavily represented in its training data, it provides responses reaching the quality of some of the better textbook writing I've seen, and with the tremendous benefit of being able to direct the focus of the explanation at will. Paired with traditional resources, it's made for a productive, engaging learning experience.
I think I'm less concerned than most about hallucinations. For one, the current models are genuinely quite good when focusing on common topics. More than that, though, the issue of errors in educational materials is not unique to AI tools; take any heavily-marketed educational resource, give it to an expert, and they'll make swiss cheese poking holes in it. Cross-referencing information, using multiple sources, and simply testing ideas in the real world will continue to be important skills regardless of the educational medium.
With that said, I think there are a number of hurdles to overcome before this could impact broader education:
So I'm not sold on AI being the revolutionary solution to Bloom's 2 sigma problem, but if it can be a useful aid to those who know how to use it, and raise the floor for the expected level of quality in educational resources, I'd be happy.
Honestly... My prediction for ai usage in education is almost the exact opposite unfortunately.
From data I've seen about how students use AI, those predictions seem to be grounded in evidence.
The ultimate problem I think is that learning is like walking a tight rope. The process of understanding complex ideas needs to be hard and uncomfortable. It needs to be hard enough that people come to realizations on their own from basic principles to really fundementally understand those ideas. At the same time, it needs to not be so hard that the student gets discouraged and gives up.
A good tutor is really adept at walking that tight rope. When a student asks why something is the way it is, the tutor won't just give them the answer, they'll point them in the direction they need to figure out the answer themselves, and nudge them in the right direction if any of their assumptions or deductions aren't correct, and explain why they aren't correct.
Publicly available AI tools aren't good at this part from what I've seen. They'll just straight up tell you the answer and praise you for asking it.
That causes that information to go in one ear and out another, because no work was done to arrive at the conclusion. From what we know about learning, that work and discomfort is just really vital to the process, and LLMs eliminate all the friction that causes it.
I'm sceptic. But I'm not sceptic about what AI can do, rather what people use it for.
AI is powerful thing if used accordingly. Technically curious people can use it to make their smart home, website, program their little robot.
Other people may use it for general research - from what is the fastest production car through how many jet fuel 737 needs to fly one passenger for one kilometer ("mileage" of plabes) to pressure inside steam.locomotive cylinder. All such information can be found manually, but AI gives the opportunity to do so fast.
AI can help people write - be it formal letter or short story. You just throw the text and vision at it and it will write it for you.
AI can be used to create images - and I don't mean sharkhorse. You can easily make your house painted pink so you know if it will look good. You can visualize your new bathroom and shuffle all the things inside to see how it fits.
There are many applications that would have needed special software or special knowledge to accomplish before AI. Nowadays everyone is capable of.leveraging the power of AI for this.
So why sceptic, you ask? Well, there are kinds of people that won't use AI like this. They will use it to fortify their beliefs. This may include people like flat-earthers, neonazis or even, given we have many US people here, just normal believers. It may induce, say, unwanted behaviour in them - fortify their beliefs to becoming radicals. This is the thing with AI - when you massage it in just the one way and given it wants tovplease user, it will eventually turn from cold evidence and will become falt-earther or neonazi itself - for that one chat/user.
AI can be very useful and very powerful tool to make great things. But it can be used as powerfully when you try and use it for your cause just as well. And that is why I'm sceltic about it.
People seem to be dumbing down and AI will hasten this in many individuals. Such people will know nothing and become dependant on AI for mundane tasks like seqrching when the bus arrives or what their kids should know from school. And if the bus doesn't arrive or their kids are a bit behind, such people would easily become too aggressive - the AI said so and for them it means it must be so. In the meantime smart people will leverage AI to.make themselves smarter and their life easier. And thebdumb people will hate the smart ones because they will be unable to come.out of their bubble, unable tl use AI in the same way, and ultimately they will become even dumber.
You don't need AI tomprove my point. If you are smart and you watch Youtube, you very likely don't watch shorts that much, you rather go for some.educational stuff - there are many channels.that do this, some of them even in amusing way. If you just want to not think about something, you will go for shorts, I understand that. But you won't likely watch shorts for 95% of your time there.
And I haven't even started on AI in the workspace. It does take people's jobs. But if the management is smart enough, they use it as a tool, not as a replacement. My take on this is that there must be a person in the chain that will check the AI output. If there isn't, the company is doomed. You can't use just AI for your cause. AI do make.mistakes and if you don't point them out, it won't bother itself. This is why you need someone to.look over it.
Correction: it has been configured to please users. I have zero doubts that LLMs get optimized/tuned exactly the same way that social media have been optimized to be as addictive as possible, with no regards for any the negative impact they might have on users.
Yes, it has been configured to please the user. If the user tells it "I know Earth is flat and I want you to agree on that with me", the AI will do so. And this is how the "fortify beliefs" works with AI. Instead of leveraging it to seek knowledge, certain people will use it as circle jerk.
Do not ask if the Earth is flat. Make it so it talks to you as if it believed the Earth is flat. This is what such people will use AI for.
My one-sentencer was just absolute simplification.
If it requires significant effort, then what's the issue? Not like the person is gonna change their mind otherwise. Clearly it's not gonna make somebody who isn't sure think that the earth is flat, which is what I thought you were trying to say before.
Nono, I meant flat-earthers locking themselves up in their belief due to massaging AI into supporting their cause (in their chats, obviously).
I think, if weighted properly, it can be a pretty good sounding board and rubber duck for someone having trouble understanding a problem, and a fantastic motivational tool for people for people who want to get started on a project but have trouble with executive function and such. I could see major medical advances and have seen a lot of software reverse engineering projects in the decompilation/recompilation scene. Even if it's vibe coded trash, its a proof of concept and can be optimized or replaced into something better.
What it not good at? They're not Johnny Five or anything we'd recognize as cognizant, I'm not sure they'll ever be just because that's something of an unknown target. But we see something spitting words at us and we think, "it's a conversation" but the models we have aren't very good at being what we think they are. Probably says more about us.
Edit: In terms of the Chinese Room, LLMs are not good at understanding the conversation, and we're not good at picking up on that.
I'm also skeptical on the tech, but I'd be blind if I said it wasn't useful. I do think it's a nice democratisation of technology, that we're all able to just tell the system to "write me a userscript that fixes this horrible website I don't like to use". Of course, even considering to do this requires a significant amount of computer brain, so YMMV.
In my ideal world, you'd get an AI license from your local university after a HAM-radio style operator's exam.
Here's a bunch I could think of off the top of my head:
Generative AI for Images
I think it's going to kind of go the way Photoshop has: a thing that's there, a tool to be used, but with a light touch or you'll be accused of graphic design being your passion.
Generative AI for Non-Fiction Prose
It'll continue apace, people find it useful for cleaning up corpo-speak and helping them marshal messy thoughts into something coherent. It will be obvious if you used AI to help or just had the AI shit something out for you.
Generative AI for Fiction Prose
I think this may get really messy. There's a lot of finger-pointing and reveals of obvious uses of AI that "somehow" made it late into the process (including pending books being pulled when the publishers "found out" it was AI and yanked it due to outcry/backlash). I think people will continue to try to use it, but we may get to a point where we have to insist that authors submit their git commit histories of their books.
Generative AI - Coding
I think mostly it turned into knowing how to use the tool, and while you CAN do things the old-fashioned way by hand, largely you won't.
Image Recognition AI - Medical
This is one of the most mature technologies and I trust it, coupled with a seasoned medical staff, to pick up things early and aid in correct diagnoses.
Image Recognition AI - Law Enforcement
We're one or two high profile cases being lost or thrown out for using systems with shitty hit rates because a good lawyer was able to argue that the technology isn't anywhere near mature enough to be admissible in a court of law as evidence. But I do predict it'll be used in casinos and other age-gated establishments to help keep out cheaters and the under-aged. I think ultimately that automatic license plate readers are going to be snuffed and we'll go back to cops having to take a picture and decipher the numbers and manually look up and confirm the plates themselves.
That's kinda what I can think of right now, but yeah... I think especially local models are going to be popular as people feel guilty about using huge data centers, like me.
I should say that I'm fully a layman, I have no background in any of this and though I don't generally like generative art and writing AI, I'm fairly agnostic about the rest.
Card counting isn't cheating, no matter what the casino would like you to believe.
True, but at the same time they're allowed to find any old reason to bounce you.
My concern here isn't so much about it being used directly as evidence of a crime, but it being used long before cases even get to that point.
We've seen that with Flock cameras. Cops don't use its license plate recognition in court as evidence, they use it as a gigantic countrywide dragnet so that when they're looking for someone, they can find that needle in the haystack that is every car on the road. That may or may not be unconstitutional, because on the one hand you're surveiling a massive number of people that have no probable cause, and on the other hand you're in public without any expectation of privacy.
That kind of technology can be used for a lot of things that I find frankly pretty scary. Automated sentiment analysis bots looking at social media posts to flag people as persons of interest, cameras with gait analysis features correlating the way people are walking with nervousness/impending crime, facial recognition networks, automated traffic analysis flagging IPs and so on.
Those things don't get submitted as evidence, but they give cops tips on where to look. The 4th amendment is open to very, very wide interpretation, so in practice this kind of stuff has a lot of leeway to get implemented, and even in the cause of clear constitutional violations, by the time violations are discovered and litigated, the damage is already done, and the offending agency faces no actual consequences other than going "oopsie, we won't do it again!". Lives are already ruined, discourse is already chilled, and everyone already has the feeling that they're being watched 24/7. Especially when you have an administration like this one who is so readily willing to weaponize the entire justice apparatus for political gain.
I’d like to gently push back on the notion that we have no expectation to privacy in public; while you should expect to be recorded in public, you should not expect to be tracked or have your likeness used without your consent.
The reason that I gave the caveats that I did (such as a good lawyer tearing apart a high profile case) is because I may be delusional but I still believe that overall we’ve got a strong justice system in the U.S. (just a slow one) and that a reasonable judge would be persuaded that the state is overreaching their authority to use such equipment in the single-minded pursuit of catching criminals. It’s getting to the point that the technology allows them to invent a criminal out of thin air.
Just a random usage of AI:
Tonight I used it to help me decide between 2 strata management companies as I am Vice President of my strata council and we are looking to switch managing companies. I fed GPT the contracts from each company
and had it do internet searches and analyze their websites and give me a comparison summary and then a few questions about some details. Many hours of work completed in 1.
I think there is a clear pattern where the usefulness of LLMs scales with how easily their output can be verified. The easiest being when the verification can be done automatically, such as code being compiled and tested or math proofs being checked with something like Lean. This will certainly speed up some tasks, but I doubt that most jobs consist of doing only that.
There are two things I have in mind where I think LLMs could be really useful tools that haven't been explored all that much so far (or at least I'm not aware of that). The first one is about proving software. A lot of bugs and security vulnerabilities come from stuff like buffer overflows, dereferencing null pointers and more generally invariants being violated. In my opinion, safer languages like Rust fall short on two aspects: they will rejects programs that are provably safe (sometimes trivially so) and they only check a small subset of issues with no way for programmers to extend what is being checked. More general approaches, where people can add verified annotation for any provable properties would be the logical next step. So far, this kind of approach was way too labor-intensive to make sense for almost all software, but I hope LLMs will help reduce this cost to make this approach more common.
The second thing that I think LLMs could help with is making interaction with complex software easier for casual users. As an example, I use Gimp every couple of months to do some basic image editing. Because of this low frequency, I will never build enough habit to be really efficient and I think that having an LLM translate human language into whatever the software understands could really help here. Importantly, I think that having a more structured output for the LLMs should reduce the risk of unwanted effects compared to asking a multimodal model.
On the other hand, hallucination and mistakes that basically no human would make are still a problem. For example, I recently had a model (I think GPT-6 Luna) casually drop something like "that was 17 months ago, so not quite a year" in the middle of an answer. I don't know if we will ever reach a point where this happens rarely enough that it won't affect the quality of outputs.
I'm a bit behind on the current AI capabilities, but having tedium as the goal seems to be the best for AI, others have already echoed this here, but it's very nice when the AI is used to solve things like tedious data sorting, writing a bit of firmware for an obsolete piece of tech, generatively changing the color of a house as someone pointed out and whatnot.
I've always used it with a "I supply the ideas, and you help me come up with more, through doing the tedious research".
The issue now is that AI is coming up with genuinely good ideas sometimes. It came up with a much better technical solution to something I wanted to do in my homelab, and I was pretty dumbfounded at that. It still fits in the "tedious research" category for now, but I might have to redefine this in the future if it keeps going at this pace.
The one thing that doesn't change is responsibility, that should almost always be an expert human, at some point in the chain.
AI will never be as good at being as unemployable as me. It will never be as good at spending years trying to improve your social skills, only to never see any real change because you always have this looming sensation that something is wrong. It will never be as good at having trouble sleeping when the loneliness becomes unbearable. It will never be as good at maintaining that looming sensation that “retirement” is always two months away.
Maybe I just need to talk to an AI about how I feel.
Talking about generative AI and LLMs, I think it's probably one of the most impressive technologies we've developed as a species. I think its potential to improve all fields in all kinds of mind-boggling ways is absolutely bonkers—in particular I think assistive technologies for persons with disabilities and medical research stand to benefit immensely.
All that being said, I cannot in good conscience use them myself, and do my absolute best to interact with LLMs and other generative AI tools as little as humanly possible. That they are fully controlled by the absolute worst corporate powers run by the greediest people on earth, built completely without the consent of or compensation to the people who originally created the data they're being trained on, and being developed with less-than-zero regard for social and environmental consequences, makes them absolute poison to me. Additionally, that so many of my peers seem to have drowned in the flavor-aid being forced down our throats and whole-heartedly embracing them while making wild proclamations about how they'll replace all humans in various professions (to say nothing of the even wilder claims of AGI), I am even more turned off of the technology.
tl;dr I think they're awesome, I think they're being built and used irresponsibly, and I refuse to use them (until that hopefully one day changes).
Firstly I believe it is really difficult to delineate what is AI and what is an algorithm that has been reframed as AI by corporate.
The best and most ethical use I have found is OCR. I like it giving me difficult primary sources as text as it makes using a dictionary easier, but still you got to know the language because it always makes dumb mistakes then gaslights you otherwise.
On the flip side great OCR existed before so called AI so idk if this counts.
I have found all AI to be unreliable at best so I really do not think there is a good use of it.
Lots of great points here already, but I haven’t seen cybersecurity mentioned here yet. LLMs can make extremely high-quality (spear)phishing campaigns trivially, producing higher-quality spoofs without a lot of the dead giveaways that even a non-savvy person could spot. Scraping a person’s entire public online presence for insights on how to scam them can be done in minutes now. Or, scraping the online presence of a target’s loved one to create deepfakes and scam people is much easier and again much higher quality. That’s all just social engineering too, and doesn’t even get into all of the accidental (or “accidental”, perhaps) LLM-in-testing hacks that have been all over the news lately. For all of the advances LLMs can promise in good and nice things, it can be used to advance bad and not nice things as well. That’s not necessarily a reason to shut it all off, but it feels like it should be a bigger part of the conversation than it often seems to be today.
I have to use LLMs a lot for my job. Like, usage is mandated by department and company leaders, and I am the admin for a suite of tools that uses LLMs for various functions. I struggle with that due to ethical concerns like those cited by various users here, but it’s not exactly the best time to be job hunting, and there’s kind of nowhere left to go in my career path that won’t be as bad or worse for LLM usage. So my options are pretty much “live with it” or “burn it all down and start over in a new career path”. That decision has definitely been weighing on me a lot lately. I’m paid very well for what I do and my job isn’t that hard for me, so halving my salary and starting at the ground floor again is a tough pill to swallow.
Everybody's going to lose their jobs!!!! Aaahh!!!
But nobody is going to want a robot in their mouths (at least not for many years, until they're confused why they have a human doing work). Become a dental hygienist before it's too late!
And if your country doesn't have dental hygienists, go get that law changed because it makes your dental care more expensive and decreases availability because actual dentists have to do everything. Imagine how ridiculous and expensive it would be if they replaced all nurses with doctors.