@Gargron agreed. It's just LLMs, the professional nonsense generators

@Gargron
Nobody says #ArtificialSmartness.

Just sayin’

@Gargron funny I had this exact same exact conversation in my head yesterday about something in a paper I am writing. I decided to define the method as ML throughout and I will most def argue to keep it as such even if any reviewer/editor will suggest to me otherwise 😌

@Gargron

Maybe it should be consistently written as "AI" instead of as AI.

@Gargron well said. #AI is high throughput data integration, processing and repackaging.

@Gargron I once saw a picture of ai-generated js code with thousands of errors🤣

@Gargron

It's another round of Wall Street hype by anti-democracy billionaires, no different than NFT's or cryptocurrency - just another scam.

@Gargron

You may have seen this, Eugen - I used to have a Buddhist roshi who would say "Anything worth saying once is worth saying a thousand times."

This is the plainest, most sensible explanation of this AI beast I've ever seen in 40 years of working with machine intelligence.

https://youtu.be/eK0md9tQ1KY?si=k0eEy0rO9Q9tOLs2

@Gargron and sadly nobody will become more intelligent by using it

@Gargron It's another huge tech scam, like crypto...

@Gargron still wonders about your main goal of that post?!

@Gargron
"AI" can mean a number of things:
- General AI, à la HAL9000 or The Culture: ChatGPTs are not it and will never be, just like we cannot reanimate dead corpses simply by pumping massive quantities of electricity into them
- The field of AI, an academic discipline that has been around since the 1950s: ChatGPTs are a misguided application of a sub-sub-sub-field of that discipline, based on misappropriated IP.

@Gargron This false sense of skill or fitness for purpose comes from its incomprehensibility. Not because it's complex, although it is that, too. More because it is not designed to be comprehended. Only to be consumed.

The danger of AI is not in that it's not intelligent, it is in that it's unintelligible.

@Gargron Then we can agree that at this stage AI safety is another trick of marketing.

@Gargron As usual, rms is telling the truth and nobody listens ;-)

"I can't foretell the future, but it is important to realize that ChatGPT is not artificial intelligence. It has no intelligence; it doesn't know anything and doesn't understand anything. It plays games with words to make plausible-sounding English text, but any statements made in it are liable to be false. It can't avoid that because it doesn't know what the words _mean_."

@Gargron
The deceit doesn't start there, but in the A. None of this is any more artificial than most of what surrounds us, certainly all of our software. It's automation. It's also, in most cases, inference. So Automated Inference.

The questions are what is being automated, who stands to benefit, who is at risk, and what are the guardrails around.

@Gargron What is intelligence?

@Gargron yep

@Gargron that, the climate impact and the underpaid annotators are the three points I always try to fit in when I talk to people about LLMs and their chatbot interfaces.

@Gargron
I have been calling it an algorithmic tool. I like @emilymbender suggestion to reference it as automation.

#AI

@Gargron this is the same problem we had when expert systems were called "AI" https://en.m.wikipedia.org/wiki/Expert_system

I guess the temptation to think a problem is solved is too high.

At least we're consistent in calling rubbish systems "intelligent" 😂

@Gargron
Isn't that exactly why it's called *artificial*?

@Gargron Artificial Incompetence?

@Gargron Whatever intelligence is. Advanced statistics?

@Gargron Reminds me how it is fashionable to say AI subtitling, AI voice etc. I am old enough to remember when these things were called speech recognition, text-to-speech etc.

@Gargron @darylgibson

A college professor of mine back in 1983 said "'AI' is what we call software we don't know how to write yet." I think this neatly captures the problem we have talking about current "AI". In 2000, nobody knew how to write software that would drive cars, write poetry, play grandmaster-level chess, or summarize text, so those were considered to be examples of what AI might accomplish. Now we know how to write systems that do those things, so they are no longer AI.

@Gargron The people using the phrase "AI" to describe their weak products are the same people who used the word "algorithm" for everything ten years ago.

@Gargron I agree, which is why I usually describe contemporary AI as Statistics on Steroids

@Gargron My concern with AI is also one of category. I think we can think of it it to reductionist a way. It's going into a socio/pysch complex system to be used. It's the use that likely will define it.

@Gargron is this not “gatekeeping” the term AI?

Eg if ChatGPT is not a form of AI, what does good AI look like?

Maybe we are thinking to much about Sci-Fi where we want to have a human-like bot?

@Gargron Like everything in computer science, AI is only as good as the program behind it (programmers are human and make mistakes) and the data it is fed with...

@Gargron It's a hybrid of Symbolic AI and Neural Networks, which is where this was always going. This current weak adaptation is already destroying egos left right and centre, and it's going to improve. More feedback. Deeper abstraction. Inference.

@Gargron yeah. The true test for AI I believe is when they’re able to self learn and improve. So recessions of generations without human input can result in them being “better”.
So far all the “AI” systems being pushed are known to become worse if fed their own output.

@Gargron By now I dream of AI under the influence it may make it more human.. though I am not under influence by myself :)

@Gargron
There is no common accepted definition of intelligence with which you can prove that a human around you is really intelligent and not just acting like?

It's a double standard if a Computer has to give more proof as a human.

EDIT: In reality its the same with humans. Humans dont understand the logic behind grammar, they just use it because they learned it statistically from their parents, not the rules.

@Gargron True. There is no original intelligence but data processed and delivered as information with a really good packaging.

@Gargron It’s all quantum fields.

@Gargron What I consider to be proper "A.I." is implementing coded algorithms to do stuff like make a robot that learns its way round a maze or a chess playing algorithm or the min/max algorithm as examples.

@Gargron I prefer expanding AI as Adaptive Inference

@Gargron cosmic brain: intelligence doesn't exist at all

@Gargron I would remind you of the percentage of the population that still believes that the CIA puts chips in Covid vaccines, that lizard men are not D&D NPCs but are real and have invaded Earth, and believe that Harry Potter is a danger to children because they will learn magic spells and become involved in witchcraft and become damned. For most, including college educated adults and many with graduate degrees, AI really means MMM — Magic Mojo Machine. It’s good gris-gris.

@Gargron Well, as John Haugeland writes in his AI book, this is no different than how we refer to natural occurred, synthetic and artificial e.g diamonds.
Are we looking for the creation of "true", synthetic intelligence? Sure in the long run.
But, artificial is the best way to describe the current state.

@Gargron

And it consumes energy as if there's no tomorrow.

@Gargron I've been trying to say exactly this to people for years. It's basically just autocorrect with a way bigger database and better(ish) context awareness. AI requires sentience, which isn't possible with current hardware, and probably never will be.

The misunderstanding is leading to some companies trying to adopt language models to replace workers, but it doesn't play out well. Eventually it'll settle down to be a toolset but not a replacement for humans. There's little to fear there.

@Gargron I agree with you 100% I've tried to talk sense into acquaintances, but they're buying into the mass hysteria created by either ignorant fools or people who have something at stake.

I think this issue highlights the importance of science education.

@Gargron THANK you

@Gargron That battle has been lost about 40 years ago. It will just be called AI from now on. Those who can‘t accept it will be forgotten in the same corner as the "it‘s not <the internet>, it’s called <the World Wide Web>" guys.

@Gargron Very well put.

@Gargron alternative suggestions:

- glorified autocomplete
- plausible-sounding text generation
- primitive pattern recognition less intelligent than an average five-year-old

@Gargron I hear you. 99% of what we do at work is classic ML using statistical models. The math exists since the 1970s but the availability of data and cloud proccessing have really elevated its usefulness and companies jump on these wherever they can. Marketing does the rest.

But not everything we do right now is just statistics. Looking at advances in ANN for example.

@Gargron I feel the machines can respond more intelligently that most humans, they may not be right but, indeed, after several elections results in countries, where humans seem to vote contrary to their best interest, I believe they will do better in any field than us, over not too long time.

@Gargron very recently seen the suggestion to call it “automation” (don’t remember from whom, maybe @emilymbender ?..)

@Gargron And let me guess...your definition of "intelligence" specifically excludes what people are calling AI. But here's the thing...there is nothing called "intelligence." Find out what the independent-variables are that are relevant to the behavior we _call_ "intelligent" - even in nonhumans. And, actually, after the era of GOFAI ended, they did...sort of. AI people have ignored the natural science of behavior and it has hurt their efforts for AGI. Think "conditioning," people. Sheesh!

@Gargron i have been long thinking about this. Isn't our own brain electric pulses? We are social animals, we learn imitating others, so, what is intelligence? Not trolling, i'd like to read you

@Gargron or as I like to call it “garbage in, garbage out”.

@Gargron agree

@Gargron
It's not "AI" 𝘺𝘦𝘵, that we know of.

@Gargron
true
+an unlimited computing capacity

@Gargron

@Gargron But human can be viewed as a function with several inputs and outputs,when software can simulate this function at some extent, call it intelligence is not improper.

Althouth the Turing test not so proper for today's deep leaning, you cannot say it is not about intelligence because it is largely based on statistics. They are investing on principles people uses daily, but the principles is got by statistics.

@Gargron they will be really intelligent than us

@gargron

Where did people read about the autocorrect metaphor?

When more than one is using the term, it is highly likely they saw it somewhere.

It doesn't give any sense to me. Maybe I understand, when reading the original thoughts.

The picture, I am usually using, is the Rorschach test with prepared pieces of paper.

@Gargron

If silicat based lifeforms come to earth, what if they say: „they cant become sentient. Its just chemistry“?

Don’t underestimate emerging effects. Sure, we are not at this point, yet, however: never say never.

@Gargron Exactly this! Far too many are relying on it for actual answers about everything, as if it actually knows. No it doesn't, it's just statistics, it doesn't "know" shit.

@Gargron I call it “absent intelligence”. (German: KI, “keine Intelligenz”)

@Gargron Those systems behave intelligently even if they are not. They understand complex questions and complex code for example. Even if it is all statistics at end, the results are breathtaking. #AI #ChatGPT #OpenAI

@Gargron I'm glad that those are going through this Ouroboros effect. Internet is flooding of "AI" sh*t and it's so annoying 🫠

@Gargron "AI" is a testament to how we assign a different meaning to the word as we see fit or fancy. Which "AI" is incapable of.

@Gargron Please compare with Geoffrey Hinton's answer in last week's interview with @erictopol : https://mstdn.social/@erictopol/111546278425434431

ET: "Human Intuition vs what a LLM (Large Language Model) can do: are these complementary?"

GH: "Yes & No. [...] I think they work pretty much in the same way as us. Most people who say `they don't work like us' don't actually have a model about how we work. It might interest them that these models were actually introduced in order to understand how our brain works".

@Gargron No. The danger they pose is very real and present, through the massive abuse of $2/hour Kenyan workers, to the wholesale stealing of licensed works, to the fraudulent marketing, to the very real impacts on the poor and minorities.
Read Emily Bender and Timnit Gebru's works, amongst others, to understand the actual harms being done to real people, today.

@Gargron this.

@Gargron I used to do research (20+ years ago) getting statistical models to generate plausible sounding musical performances. We got quite good and learnt a lot about what makes music musical but there was certainly no intelligence there. Lots of musicians worried that work like ours would take away their livelihoods, but actually it was streaming that did that ;-) LLMs are probably a similar red herring.

@Gargron Well put. I'll give "AI" points for being clever, maybe even useful in some limited cases, but "garbage in, garbage out" still applies.

@Gargron as seen elsewhere...

ML is written in Python.

AI is written in PowerPoint.

@Gargron

I call them LLMs

I call general intelligences to be Digital Intelligence because when eventually real they wont be Artificial

AI is a field of research that has existed for decades, and has already resulted in chess computers, expert systems, automatic pattern recognition, image recognition and more.

Of course LLMs are AI too. What they are not, is Strong AI or AGI: human-level general intelligence.

@Gargron

how do we know we're any different? we can't prove it. but I'll agree they need more scrutiny and less hype. they seem designed to sell compute services.

@Gargron that’s like saying it’s not the heat that kills you in an explosion but the change in pressure. It doesn’t change the result and in Marketing the plume of flames is what people will recognise and understand.

@Gargron
As I say in one of my pinned Toots: "Aritificial intelligence" is just digitally amplified natural stupidity.

@Gargron It's tempting to call it AI, but I'd rather refer to it as 'sophisticated pattern recognition software.' Ultimately it's just a fancy algorithm, or a glorified data matcher.

@Gargron what i dont get is the implicit bias they build in the machine decision making that they don’t even see. Its bizarre and dangerous. Hubris.

@Gargron nicely put. Been working on this for several years ... https://berryvilleiml.com/

@Gargron wrote about that a few months ago: https://preslav.me/2023/05/22/i-believe-in-machine-learning-dont-believe-in-artificial-intelligence/

The best we get to call it is Augmented (Human) Intelligence. Like those glasses that overlay things in front of your eyes, current AI is mostly a tool that does stuff for you, and it just happens to do it better than tools before it. New types of problems bring in new tools to solve them. But it’s just brute-forcing an answer in the end, I agree.

@Gargron I try to always put "AI" in quotes, or use LLM or chatbot instead.

@Gargron

The basic concept of AI, although hyped, is to increase frequency in correlation of data to provide new hypothesis.
How that can lead to sentient computers learning will be discussed at the time when language models are massive. 10 to 15 years from now.
That's it IMHO

@Gargron going to start calling it AB

@Gargron Instead of focusing on the possibilities, the description signals the desire to belittle, classic human behavior under stress. We know nothing about future developments.

@Gargron My son gets this lecture every time someone so much as breathes the word. “It’s not AI. It’s just a chat bot with internet access. It’s also inaccurate. AI means intelligence. They have none. They’re just computer programs.”

(Not disparaging anyone who writes aforementioned programs, but they’re not artificial intelligence, sorry. Also our councils and the government don’t use computer programs to detect fraud, but “AI” which just adds to the whole lecture.)

@Gargron Yes and yes, but current views of human language processing suggests it is highly "statistical" in nature insofar as that it seems to track and record conditional probabilities, which are also at the heart of modern AI.

@Gargron It is called "artificial" for reasons. AIs are trying to solve classes of problems which require natural intelligence to be solved by human beings. That doesn't mean the computer solution is doing it the same way.

One problem with current LLMs is that their knowledge base are vast amounts of web pages created by humans which lack natural intelligence. Garbage in, garbage out. The machine didn't original research.

@Gargron

just an example, you are looking for details about the rockets of Hamas terrorists.

Google [DE] automatically asks for you:
How many rockets owns Hamas?

Google [DE] automatically answers for you:
The IDF recently said 7.000

:digitalcourage:

This is complete bullshit, 7.000 is not the count but a number of rockets fired into Israel within a few days.
This is google fueling Jew-hatred by guessing numbers.
A huge danger @EC_Commissioner_Breton

@Gargron No, the biggest danger they pose is that we computerize/network everyfuckingthing. Intelligence is not different from logic circuitry – we just have emo-chemistry to flavor it. AI could pose a real, physical danger because we let it.

@Gargron come now! This overstates our current knowledge of the nature of intelligence. LLMs are adaptive, they have memory, they use language to communicate, and they integrate disparate experiences to solve problems. They have many of the hallmarks of what we call intelligence. They have more such characteristics than, say, dolphins or chimps. Us knowing how they work is not a disqualifier for them being intelligent.