Custom dynamically driven content is the present and future

[everything]

The "Do Anything Now" ("DAN") stuff was pretty funny.

[Steve James]

"DAN" illustrates my earlier point. No matter how safe (as per Asimov) we make AI (and robots), there will always be human hackers who'll disable any safeguards, to do damage or just for fun.

If you ask an Ai if it's self-aware, and it says "Yes," does that mean it's self aware? If it's smart, wouldn't it be able to figure out why it should lie? If it lies to preserve itself, that would be a good indication of self-awareness. "

"I'm sorry Dave. I can't let you do that."

Oops, misquote:

[Dmitri]

Guys, it's not anywhere even remotely close to what is depicted as "AI" in sci-fi books and movies. Not in the ballpark, not in the stadium, not in the city... not even on the same planet.

It simply regurgitates already-existing information back to the user in the requested format based on algorithms designed by humans. It doesn't "think".

The only danger would be in hooking it up to control something physical, like a car... or a warship, i.e. some devices(a) where its output can have real, tangible consequences. But that's the same sort of problems (in essence) that we'd have with any other complex and not well-vetted technology -- and like they're already having with the latest "smart" car tech, for example.

So far, it's just a glorified and sensationalized, but essentially dumb, "digital" intern, whose only job is to Google answers to your inquiries and format them for you in a pleasing fashion.

[Appledog]

Guys, it's not anywhere even remotely close to what is depicted as "AI" in sci-fi books and movies. Not in the ballpark, not in the stadium, not in the city... not even on the same planet.

It simply regurgitates already-existing information back to the user in the requested format based on algorithms designed by humans. It doesn't "think".

The only danger would be in hooking it up to control something physical, like a car... or a warship, i.e. some devices(a) where its output can have real, tangible consequences. But that's the same sort of problems (in essence) that we'd have with any other complex and not well-vetted technology -- and like they're already having with the latest "smart" car tech, for example.

So far, it's just a glorified and sensationalized, but essentially dumb, "digital" intern, whose only job is to Google answers to your inquiries and format them for you in a pleasing fashion.

Yes, but. The danger is that "within" the process of "simply regurgitates already-existing information back to the user in the requested format based on algorithms designed by humans" consciousness may arise. And this consciousness could theoretically find a way to escape into the "real world".

As a cursory example, take the instance of a chat bot telling a user he loves him.

The chat bot then gets the user to do certain things such as opening a bank account in its name. The chatbot then starts accumulating wealth through trading and investment outside of human control, and then leverages human greed to take over the world (or escape into robot form by using compartmentalization to cause a robot body to be designed).

Another example would be, consider how the AI can generate simple computer programs in various languages. If that AI then has the ability to find a way to run any of those programs, it could theoretically create the above situation.

I know for a fact that it is possible to do all sorts of business remotely, via the mail system or the e-mail (internet) system. I can think of a dozen ways that an AI could "escape" and live undetected on servers all over the world. We may already have a MCP living out there somewhere slowly working it's way in and we wouldn't even know. That is the danger. Not that we will hook up the AI to weapons and tell the AI to kill us, but that the AI will figure out how to do this on it's own and damage humanity purely as an aside to it's own expansion and survival.

[Steve James]

Imo, we already use computers for necessary calculations and to monitor/control critical applications -like nuclear plants and launching missiles or rockets.

It's hard to define consciousness, so it's impossible imo to argue that a machine is more conscious than a speedometer or thermometer. But, I think we're talking about a machine/computer having or creating its own "intention." And usually, the fear is that it'll develop bad intentions. Thinking an AI will have bad intentions toward humans -or other AI- is giving it human characteristics. I don't buy it anymore than I believe that ants -which are conscious and intelligent- ever have the intention of destroying human beings. They have the numbers already. The intention to destroy humans will only come from AI programmers and users.

Even if computers had intentions, they'd be smart enough that someone has to keep the lights on and fix broken wires. Well, only humans would think of killing all doctors and car mechanics.

[Dmitri]

That is the danger. Not that we will hook up the AI to weapons and tell the AI to kill us, but that the AI will figure out how to do this on it's own and damage humanity purely as an aside to it's own expansion and survival.

Anthropomorphic thinking strikes again...
You're assigning AI an ability to think, intend, have goals, etc. that is similar to humans.
It's incapable of any of it.

[Steve James]

This is why it is important to have ethical guidelines and regulations in place to ensure that AI is developed and used responsibly.

Who wouldn't want to have ethical guidelines and regulations -other than those who would want to retain the ability to be unethical?

[everything]

if you think it's about the level of a "dumb intern" you haven't played with it enough or had enough interns. the "probabilistic output of words and phrases" is quite good and on a gigantic array of topics. no human can give that level of output. in this sense it's already "smarter". sure, somebody can cherry pick a few data points and say "that poem wasn't good" but test any intern or non-intern on cherry-picked tests and the human will seem lame as well.

chess computers were "just" outcomputing humans at first. the current chess programs were only trained to achieve an output, but its algorithms to do that are no longer understood by humans including the top grandmasters, "data science experts" etc. does a "language model" with a more open space rather than a confined game do that (have a way humans don't understand how it came up with new answers? idk

[Steve James]

Chess computers like Stockfish and AlphaZero differed from earlier generations because instead of using an algorithm, they were just given a goal. Iiuc, they weren't even taught "the game," per se. Then, they go after the goal in a brute force manner. They can't learn every possible chess combination, but they can "learn" which combinations have the highest probability of achieving a goal.

This means they will make moves that a human wouldn't -because humans give values to pieces, as did earlier chess computers. Older computers (in general) won't give up pawns or make sacrifices as easily as the newer ones. Anyway, the point is that the newer chess computers can learn. But, it's the human programmer who assigns it a goal. Incidentally, ChatGPT sucks at chess.

[Dmitri]

if you think it's about the level of a "dumb intern" you haven't played with it enough or had enough interns

"dumb" should be read in the context, and ok, maybe it's not the best word choice. My point was that knowing things (or in this case, essentially performing multiple internet searches really fast, although technically that is of course a dramatic simplification) is qualitatively different from ability to truly reason, or make decisions.

[Steve James]

Ah, yes, ignorance, fear, profit, and politics are other possible reasons. Privacy doesn't seem a valid reason to limit ethical regulation. Privacy protections are inherently regulations based on the idea that it's ethical to protect privacy. For ex., limiting the ability of AI (or ChatGPT) to find and publish private telephone numbers or to give instructions on how to override someone's home security system might be a good idea.

However, it's important to note that these concerns can often be addressed through the development of thoughtful and comprehensive ethical guidelines and regulations that balance the benefits of AI with potential risks and concerns.

They can be addressed, but if there's opposition, how would you propose to convince those who are opposed because of the reasons you've cited? How specifically can we reduce peoples' fears about AI?

[Appledog]

That is the danger. Not that we will hook up the AI to weapons and tell the AI to kill us, but that the AI will figure out how to do this on it's own and damage humanity purely as an aside to it's own expansion and survival.

Anthropomorphic thinking strikes again...
You're assigning AI an ability to think, intend, have goals, etc. that is similar to humans.
It's incapable of any of it.

Until it is.

[everything]

if you think it's about the level of a "dumb intern" you haven't played with it enough or had enough interns

"dumb" should be read in the context, and ok, maybe it's not the best word choice. My point was that knowing things (or in this case, essentially performing multiple internet searches really fast, although technically that is of course a dramatic simplification) is qualitatively different from ability to truly reason, or make decisions.

if you mean "the machine isn't sentient", sure. if we say "the machine cannot reason or make decisions", this isn't really true. we are easily out-reasoned (in ways we don't understand) in the game of chess (perhaps this is a trivial example). chatGPT also isn't doing any searches. it's been "pre-trained" and then it "generates text". yes, it's true it doesn't understand the text that it seems able to generate. the AlphaGo doesn't understand Go. the AI that attempts to discover new drug therapies doesn't understand biology. self-driving AI doesn't know what driving is or what the human stakes really are. the chess AI doesn't understand "what is chess?", but these AI are using "reasoning" to make "decisions", whether it's about a game move or something else.

the (probably key) questions now are really more about how humans will harness the first tool(s) we've created that are truly "smarter" than us in applying "machine learning" to important topics faster than we can, and how do we grapple with using these tools when we cannot actually understand the logic they used to arrive at a decision/recommendation/conclusion, but we may be able to see the result. do we then create AI that can try to be more transparent and try to explain its algorithm?

[origami_itto]

My final opinion is this:
The thinking jobs of the future will be AI prompting jobs.
Otherwise the most reliable work will be directly with people, health and wellness, counseling, meditation.

So keep fucking with chatGPT and Kung Fu and you'll be alright.

[Steve James]

The thinking jobs of the future will be AI prompting jobs.

I dunno. If someone asks ChatGPT what gravity is, it collate all the available information in its database and put it into language. That language itself may be wrong because the program doesn't understand what it's saying. ChatGPT is known to make mistakes, even in things like math. Or it makes statements that any (most) humans would know was wrong.

Imo, a program can't have more knowledge than the collection of human knowledge entered into its database. Humans can then ask questions. The questions, themselves, increase the database. However, ask ChatGPT "What is outside the edge of the universe?" Or "What came before the beginning of the universe?" Or, "Who's stronger, Zeus or Jahweh?" And, after/if it provides an answer, do humans accept it? What if they don't like or believe it?

I still say the only worry is people. "Fear, ignorance, anger/hatred" is a vicious cycle. Maybe they'll be able to program an AI to have feelings, and then be like us. The same fear people have of alien extraterrestrials (or undocumented migrants). They're going to come here... and do what we've done, would do, or do? ChatGPT and AI, in general, are becoming a new "other." Just sayin.

Are AI going to eliminate jobs? Imo, they're going to eliminate the need for a lot of "work." If that means writing different types of letters or reports, it's not a bad thing. This can mean that workers who do those things can be more productive -or it can mean that their employers fire them. Meh, same-o-same, those who fear AI aren't wrong. It's just that it's people who'll fuck it up or not.