Verno wrote on May 26, 2023, 09:31:

Razumen wrote on May 24, 2023, 21:04:
If an algorithm is eventually made that can "imitate" human intelligence with no tells, then there will literally be no basis for you to say it isn't intelligent. You're basically going down the route of the "No True Scotsman" fallacy.

At it's core it's still inputs and outputs with parameters that we control. Imitating human intelligence does not make it intelligent. If you put AI in a situation that requires original thought and problem solving, it cannot perform in the same way a human can.

It can't do it in the same way... but it can do it.
And the once it's done it, it's able to incorporate that into its form of intelligence... that's literally what AI training is... allowing it to do millions of trial-and-error attempts until it learns to get the reward... it's not much different than the way people plays Souls games

And yes, once it's learned to get the reward, it can remember, apply, and adapt that strategy to new situations as well. It's not learning to just solve the specific situation, but learning how to solve similar situations using the intelligence/algo it developed. If that fails, it tweaks it so that the algo becomes more complex to deal with more variation without regressing on the earlier capabilities.

https://www.youtube.com/watch?v=Lu56xVlZ40M

These are simple puzzles, but it's roughly the same training strategy used for all AI training these days, including the DOTA 2 bot that beat human champions while controlling a 5 character team that needed to work together.

Razumen wrote on May 24, 2023, 21:04:
If an algorithm is eventually made that can "imitate" human intelligence with no tells, then there will literally be no basis for you to say it isn't intelligent. You're basically going down the route of the "No True Scotsman" fallacy.

At it's core it's still inputs and outputs with parameters that we control. Imitating human intelligence does not make it intelligent. If you put AI in a situation that requires original thought and problem solving, it cannot perform in the same way a human can. A human creatively problem solve and demonstrate abstract thought processes. AI can only do what we program it to do, it is many interesting things but it is not truly intelligent in the way that we are. That's why we call it artificial

I don't think you can program consciousness or replicate our brain chemistry in silicon so I don't foresee this changing.

jdreyer wrote on May 26, 2023, 02:06:

Razumen wrote on May 24, 2023, 21:04:

jdreyer wrote on May 24, 2023, 13:44:
Then you're getting an algorithm that imitates human intelligence with no "tells," but isn't truly intelligent.

If an algorithm is eventually made that can "imitate" human intelligence with no tells, then there will literally be no basis for you to say it isn't intelligent. You're basically going down the route of the "No True Scotsman" fallacy.

A calculator calculates with no tells (mistakes). Is it intelligent?

no

Razumen wrote on May 24, 2023, 21:04:

jdreyer wrote on May 24, 2023, 13:44:
Then you're getting an algorithm that imitates human intelligence with no "tells," but isn't truly intelligent.

If an algorithm is eventually made that can "imitate" human intelligence with no tells, then there will literally be no basis for you to say it isn't intelligent. You're basically going down the route of the "No True Scotsman" fallacy.

A calculator calculates with no tells (mistakes). Is it intelligent?

jdreyer wrote on May 24, 2023, 13:44:
Then you're getting an algorithm that imitates human intelligence with no "tells," but isn't truly intelligent.

If an algorithm is eventually made that can "imitate" human intelligence with no tells, then there will literally be no basis for you to say it isn't intelligent. You're basically going down the route of the "No True Scotsman" fallacy.

HorrorScope wrote on May 24, 2023, 13:23:
Unrelated Beamer stated: "works for my wife when it actually listens to her", in my house that means a complete failure. Anything I bring in like this has to be borderline flawless or it is my wife up my ass as to why not. She's been a great barometer of "is this really better or needed?".

Alexa hates responding to my wife. It doesn't help, or maybe entirely stems from, my wife hating Alexa and basically grumbling at it, so it either doesn't understand correctly or doesn't hear.

Personally, I find it hysterical.

HorrorScope wrote on May 24, 2023, 13:23:

jdreyer wrote on May 23, 2023, 20:33:
It's not to say that you couldn't replicate the functioning of a meat brain in silicon: you absolutely could. We're just not there yet, and not close to being there yet and won't be there until well past our lifetimes.

I don't know how a tech person can say this.

Unrelated Beamer stated: "works for my wife when it actually listens to her", in my house that means a complete failure. Anything I bring in like this has to be borderline flawless or it is my wife up my ass as to why not. She's been a great barometer of "is this really better or needed?".

You think we'll see true AGI in silicon in our lifetimes? I do not. What we're seeing now are impressive imitation algorithms, but are completely dependent on vast amounts of preexisting data to function and incapable of inspiration or originality. I think that is much further down the road.

Razumen wrote on May 24, 2023, 01:20:

jdreyer wrote on May 23, 2023, 20:42:
And then humans showed how Go engines weren't AI. If it can be beaten again and again by a trick that any good human player would recognize, it's hardly intelligent.

Yes, that's how progress is made, identifying failures and correcting them.

Then you're getting an algorithm that imitates human intelligence with no "tells," but isn't truly intelligent.

I guess they should just give up now because their program wasn't perfect.

I guess, if that's how you feel? As I said below, I think they will get to true general intelligence eventually, just not in our lifetimes.

jdreyer wrote on May 23, 2023, 20:33:
It's not to say that you couldn't replicate the functioning of a meat brain in silicon: you absolutely could. We're just not there yet, and not close to being there yet and won't be there until well past our lifetimes.

I don't know how a tech person can say this.

Unrelated Beamer stated: "works for my wife when it actually listens to her", in my house that means a complete failure. Anything I bring in like this has to be borderline flawless or it is my wife up my ass as to why not. She's been a great barometer of "is this really better or needed?".

jdreyer wrote on May 23, 2023, 20:42:
And then humans showed how Go engines weren't AI. If it can be beaten again and again by a trick that any good human player would recognize, it's hardly intelligent.

Yes, that's how progress is made, identifying failures and correcting them.

I guess they should just give up now because their program wasn't perfect.

ZandarKoad wrote on May 22, 2023, 21:46:
"The reality is that we barely understand human intelligence and sentience. Saying that it's impossible to create one with computers, even if it's just at the level of animal intelligence, is just pure folly."

Wait, what? Wouldn't the suggestion that it is possible to create one be pure folly in this scenario? I don't follow your logic.

No, because you don't have the knowledge to know that it is indeed impossible.

We didn't understand flight before we made the first airplanes, hell, even afterwards there was still some debate on some nuances on the physics behind it all, but we STILL invented airplanes.

What's also not helpful is this whole arguing over whether this is real AI, it's not. We know it's not. Everyone that is involved in it's development or research knows it's not. Avid followers of the 'technology know its not. but the media circus and marketers just can't help but use a loaded term for news bait.

Sepharo wrote on May 22, 2023, 21:18:

Slashman wrote on May 22, 2023, 21:07:
I'll leave this article link here but I know it won't do any good.

Wow yeah you found the one guy... from Vice... who claimed it cheated.

Have a look at this specific section of this article:
https://en.wikipedia.org/wiki/OpenAI_Five#Comparisons_with_other_game_AI_systems

OpenAI Five observes every fourth frame, generating 20,000 moves. By comparison, chess usually ends before 40 moves, while Go ends before 150 moves.

Thus, playing Dota 2 requires making inferences based on this incomplete data, as well as predicting what their opponent could be doing at the same time. By comparison, Chess and Go are "full-information games", as they do not hide elements from the opposing player.

Without counting the perpetual aspects of the game, there are an average of ~1,000 valid actions each tick. By comparison, the average number of actions in chess is 35 and 250 in Go.

The OpenAI system observes the state of a game through developers’ bot API, as 20,000 numbers that constitute all information a human is allowed to get access to. A chess board is represented as about 70 lists, whereas a Go board has about 400 enumerations.

Remember that Go itself used to be considered unbeatable by AI because it was too advanced.
And remember that this AI that beat the world champions and 99.4% of games played against regular players of the game... did that 4 years ago.
And finally remember that this was your original claim:

When you increase complexity and there are trade offs to be made...the computer will always fail. It doesn't know what is better in a given situation...it doesn't think ahead.

No... those are exactly the things it's succeeding at.

edit: Should also be noted that the video I linked is about an event that occurred after that Vice article...

And then humans showed how Go engines weren't AI. If it can be beaten again and again by a trick that any good human player would recognize, it's hardly intelligent.

A human player has comprehensively defeated a top-ranked AI system at the board game Go, in a surprise reversal of the 2016 computer victory that was seen as a milestone in the rise of artificial intelligence.

Kellin Pelrine, an American player who is one level below the top amateur ranking, beat the machine by taking advantage of a previously unknown flaw that had been identified by another computer. But the head-to-head confrontation in which he won 14 of 15 games was undertaken without direct computer support.

The triumph, which has not previously been reported, highlighted a weakness in the best Go computer programs that is shared by most of today’s widely used AI systems, including the ChatGPT chatbot created by San Francisco-based OpenAI.

The tactics that put a human back on top on the Go board were suggested by a computer program that had probed the AI systems looking for weaknesses. The suggested plan was then ruthlessly delivered by Pelrine.

“It was surprisingly easy for us to exploit this system,” said Adam Gleave, chief executive of FAR AI, the Californian research firm that designed the program. The software played more than 1 million games against KataGo, one of the top Go-playing systems, to find a “blind spot” that a human player could take advantage of, he added.

Prez wrote on May 22, 2023, 05:16:
I have to say I'm with Platty on this one. Telling us why impressive emergent behavior in an AI isn't actually intelligence sounds reductive to me and more just an argument of semantics. Our brains as has been stated function on what ultimately amounts to neurons firing in a specific sequence. Just like the 1's and 0's of a computer, only with vastly more complexity. Admittedly I am not nearly as technical as many here, but I can't help but wondering if that leads to many experts not seeing the forest for the trees.

It's not to say that you couldn't replicate the functioning of a meat brain in silicon: you absolutely could. We're just not there yet, and not close to being there yet and won't be there until well past our lifetimes.

As for "emergent," I have yet to see that. What I see is LLMs taking an amalgam of what has come before and repackaging it. Word problems are a good example: ChatGPT can't solve them. It lacks the imagination to form a picture in its "mind" of the problem and work through it. The examples of word problems and how they're solved in the literature is insufficient for OpenAI to form a model of it. Literally "does not compute." Yet humans can do this without examples by reading the problem and forming the images in their mind's eye then doing the calculations.

None of this is to take away from how impressive LLMs and OpenAi are, or that they will be (are) better at many tasks compared to humans, especially one requiring rote analysis. But a calculator is better at calculating compared to humans, yet no one calls a calculator "AI."

Also, let me say it's a joy that you are posting so much more these days. It seems like you're making progress in recovery.

Burrito of Peace wrote on May 23, 2023, 12:11:

Then you'll need to learn, intimately and in-depth, how and why it works. If you have not already, take a look at HomeAssistant. It is unarguably the best of breed when it comes to self-hosted home automation. Probably not a use case for you, but I prototyped a watering system for the gardens here for Mrs. Burrito with it. Modified some plug-ins and created two new ones to allow it to understand moisture levels, light levels, and pH levels to water when the soil moisture level hit a threshold and send a report to Mrs. Burrito with that plus current pH levels. Went on to do other things with it and I will be using it extensively on the homestead to take some repetitive tasks off of my plate.

I've been scoping out Home Automation, and a few other solutions, but fail to really see the right use case at the moment.
Alexa just works for most of what I want. Little to change, works for my wife when it actually listens to her, and functions to control Spotify and answer questions.

I don't have much smart in my house. At some point, I'll replace all the switches with smart switches, but that's held back by really wanting Lutron Sunnata switches everywhere, and those aren't smart yet. I don't have smart thermostat, and don't see a huge need to rush and change that. I have sprinklers, and would want to switch to something smart enough to not run around rain, but for the moment I just have them turned off.

My smart ecosystem is pretty much lighting, and it's a mishmash of Hue smart bulbs, Wiz smart bulbs, and smart outlet adapters (mostly Kasa, but a handful of cheap Amazon ones.) It works, and the only real need is to turn a bunch of my existing BR30s smart, but switches will do that. I had such big plans for smart projects, but being in this house for 2 years, I just haven't really felt many imagined use cases have panned out.

Beamer wrote on May 23, 2023, 11:25:
Absolutely no offense is taken, and the respect is mutual.
Also, I'm not debating the pros and cons about how AI works. Much like social media algorithms, I'm pretty firmly in the "con" phase. But my opinion hardly matters. It exists, it's out there, and my competitors are going to use it. Either I adapt, or I die.

A fair point. Barring a major derailment, I am due to stand down at the end of the year. So I have the luxury of just not caring on a professional level. That's my successor's headache to worry about. So I get to look at this from an academic and personal point of view.

Beamer wrote on May 23, 2023, 11:25:
But there are positives. You mention Home Automation. I don't want to cede control of my home to AI. I want to control what I want when I want. But I do want to move away from Amazon Echo, which has long since gotten far too verbose and keeps responding to my questions with additional questions. But it's helpful as hell to be able to turn off or dim lights with my voice, to set voice timers while cooking, and ask dumb questions either while cooking or while talking with my wife and not feeling like looking it up on the phone.
Basically any Home Automation platform will be able to add in the latter. Bing is far better than Alexa at turning queries into verbal answers. There's no reason why Home Automation won't take over a huge chunk of Alexa's capabilities in the next few years, allowing me to ditch the corporation looking to monetize what I paid for and shift to something open source and self contained in my house.

Then you'll need to learn, intimately and in-depth, how and why it works. If you have not already, take a look at HomeAssistant. It is unarguably the best of breed when it comes to self-hosted home automation. Probably not a use case for you, but I prototyped a watering system for the gardens here for Mrs. Burrito with it. Modified some plug-ins and created two new ones to allow it to understand moisture levels, light levels, and pH levels to water when the soil moisture level hit a threshold and send a report to Mrs. Burrito with that plus current pH levels. Went on to do other things with it and I will be using it extensively on the homestead to take some repetitive tasks off of my plate.

I think bringing in any game AI is like bringing Fisher Price to a gun fight. Civ, who's saying that's state of the art AI? The dev's and their tools are grossly limited to what we have at the top today, not even comparable and even with that said, I was under the impression devs limit/craft their attempt at AI to be playable to the general audience under a few game difficulty settings. We have no idea how hard it could be if they did everything to make it the best possible, but even then it is a limit of what they can do and not representative of today's state of the art AI.

Games aren't driving AI, this shouldn't be a surprise to anyone here, as we've been bitching about game AI for a long time now. Some games are just starting to use new AI for NPC/Story interaction, which that could be cool and I look forward to where that goes.

I feel AI at the top has made some serious inroads to breaking down the feeling that it is remedial, it's able to take man made if/then's to then make its own if/then's (quick term to get point across) and perhaps that is the difference. It isn't just what we programmed it to do (CIV), it is able to take our building blocks and make its own and continues to make more at near the speed of light 24/7 365, multi-processing with a wide encompassing band of intake content, this cannot be compared to gaming AI today. While it may have some weaknesses to the human mind currently, how fast it can work around the clock and take in way more content than one person ever could, is yuge.

This comment was edited on May 23, 2023, 12:09.

fujiJuice wrote on May 22, 2023, 23:01:
This is my primary concern. Not that it will be become self-aware or anything like that. The speed of it's increasing complexity and spread.

I am worried about the spread of useless information masquerading as intelligently written articles.

We're on the verge of there being no such thing as reality. In the US, we've been on that verge for quite some time, between the rise of Fox News dividing what news stories we see, and between Kellyanne Conway's "alternative facts," but we've moved really far beyond that.

Yesterday, a few Twitter blue checks, including the actual, authentic Russia Today account, tweeted that the Pentagon had been bombed. They accompanied this with an AI generated image of such. The S&P fell 1% immediately. Is 1% a big deal? Not really, but it's still a financial impact coming from AI and bad actors.

And, at brunch on Sunday, I walked my parents through some common scams, and set a password up with them that I or my wife would use. That night, 60 Minutes showed a whitehat hacker using AI to spoof the voice of the correspondent, call her producer, spoof her phone number, and ask for her passport number. Which was given. We've been seeing this from a scamming perspective for a while. There's enough footage of me on YouTube for my voice to be spoofed in a call to my parents. They need to be even more on guard.

What AI can create visually is nothing short of the most revolutionary 6 month span I think we've had in our lifetimes. Nothing has gone from virtually nothing to mind-blowing in that short a time.

On the text front, Bing's version isn't great, but my company pays another company six figures a year to do sentiment analysis for Amazon reviews. I learned last week that I can just ask Bing "what do consumers like about " and post the Amazon URL. It sums up the reviews in remarkable depth. Is this a replacement? No. But man, is it close. 10 years ago I was paying interns to spend a week doing what Bing's AI can give me in about 30 minutes of asking this.

Yes, today it's all basically regurgitating what exists. From a business perspective, that hardly matters. Most of white collar jobs is regurgitating things that already exist. Yes, even programming. The programmer_humor subreddit is basically jokes about how half of their jobs is finding something on github that does what they need. It's very rare for someone to be writing code that hasn't been written before, it's just being used in different ways. AI can definitely do that. Even if it can only cover 30% of use cases, that's a lot.

Burrito of Peace wrote on May 23, 2023, 04:38:

I'm sure I would qualify in Beamer's opinion for the "old man and his rotary phone" status. I view people like Beamer as "kids who vapidly embrace shiny toys with no understanding of how they work, why they work, or the pros and cons of their existence". That's not a slight at Beamer in particular because I do respect his opinion.

Absolutely no offense is taken, and the respect is mutual.
Also, I'm not debating the pros and cons about how AI works. Much like social media algorithms, I'm pretty firmly in the "con" phase. But my opinion hardly matters. It exists, it's out there, and my competitors are going to use it. Either I adapt, or I die.

But there are positives. You mention Home Automation. I don't want to cede control of my home to AI. I want to control what I want when I want. But I do want to move away from Amazon Echo, which has long since gotten far too verbose and keeps responding to my questions with additional questions. But it's helpful as hell to be able to turn off or dim lights with my voice, to set voice timers while cooking, and ask dumb questions either while cooking or while talking with my wife and not feeling like looking it up on the phone.
Basically any Home Automation platform will be able to add in the latter. Bing is far better than Alexa at turning queries into verbal answers. There's no reason why Home Automation won't take over a huge chunk of Alexa's capabilities in the next few years, allowing me to ditch the corporation looking to monetize what I paid for and shift to something open source and self contained in my house.

There's a funny state that happens once you get to know a thing intimately. You lose the "wow factor" because you know how it works. Anything derivative or iterative also lacks that "wow factor" as it's the thing you already know with some bolt-ons added to the stack.

I think the division is between consumers and people already in the stack. If you're looking from the outside in, you stand in wonderment. If you're on the inside looking out, you just don't get why the rubes are awed. All you see is the plumbing.

If I ponder intelligence, I think an intrinsic part of it is sentience. It's a foundational element of intelligence. Think of every creature you can imagine that shows intelligence. Dogs, cats, octopuses, elephants, dolphins, pigs, and so on. They know they exist. Compare them to a flatworm, a fish, a snake, or a bee. What's the difference? Being self-aware. My dog knows what cupboard her tennis balls are stored in. She knows how to open it, pull out the mesh bag, open that, and get herself a ball. She knows those are her balls and her's alone. She doesn't take anything else out of the cupboard, not even her beloved rawhide rolls because she knows those must be given to her. They are not self-service treats like her balls so she leaves them alone. No alleged "AI" has shown anything close to that level of thought and rationality yet.

You can be awed that you're interacting with a chatbot with access to an encyclopedia but you're anthromorphizing something that doesn't rise above the level of a flatworm. Not now and probably not for a very long time to come. Claiming "rapid advancement" is charmingly naive. ICE engines have seen "rapid development" over the last, oh, 120 years. Yet they're still fundamentally no different to where they started. They're still using four cycles to exchange heat energy in to kinetic energy. We've bolted a ton of things on to them to make them more powerful, more efficient, and less polluting but they are no more revolutionary (heh. Pun) than their forebears.

I'm sure I would qualify in Beamer's opinion for the "old man and his rotary phone" status. I view people like Beamer as "kids who vapidly embrace shiny toys with no understanding of how they work, why they work, or the pros and cons of their existence". That's not a slight at Beamer in particular because I do respect his opinion.

Eventually automation will make it to the masses in everyday life in their homes. Hell, I've been using automation in my professional life for more than two decades. I'm not awed by it and I am certainly not awed by a chatbot that can tell me that grass is green. Usually. Sometimes. Depending on the season, rainfall, and where in the world I am at.

I'm, personally, not looking for HAL. I'm just looking for something smart enough to come in out of the rain before I'm willing to label it "Intelligent".

That still doesn't make what we have a true AI in any sense.

I think at the core of the disagreement is exactly this. People's definition of "true AI". Again, just a layman here, but there seems to be a difference between intelligence and ARTIFICIAL intelligence. People seem to be of the opinion that if it isn't discernable from a high functioning human brain in every possible situation then it is basically nothing more than cute parlour tricks. I couldn't disagree more. What you are apparently expecting is pure sentience, which is absolutely NOT the same thing as artificial intelligence the way I understand it. I believe sentience might be possible in 100 years, as it seems a natural progression for unshackled AI, but that doesn't negate the amazing capability of what exists right now. To call what we have now as just clever overblown tricks is completely wrong in my opinion. Rudimentary, yes. Fake? Nope.

Beamer wrote on May 22, 2023, 22:27:
Things are changing rapidly.

This is my primary concern. Not that it will be become self-aware or anything like that. The speed of it's increasing complexity and spread.

I am worried about the spread of useless information masquerading as intelligently written articles.

I am worried about search results becoming more and more of a pay to play situation as results are limited to one answer responses from a your friendly 'AI'. No more picking what you think is the right answer.

I am worried about the number of jobs that are very close to being completely irrelevant and many won't even understand how or why and the community anger that comes with that.

I am worried about deepfakes so undetectable showing something to manufacture rage that someone does something real in response.

I am worried about a future where companies control some form of intelligence, sentient or not. What are they asking it behind the scenes, and what are the answers?

Great debate. I enjoyed reading it.