Thoughts On A.i

Being a computer scientist this subject has always interested me.

Let's look at the most popular known form of current A.I, "Chat GPT". It's a script - simple as that. It's coded in a language called "Python" and resides on an API (Application Programming Interface), which is connected to servers and threads via information streams known as 'pockets' or 'packets' of data heaps which read the input of a user, then searches the Internet for a solution before compiling together a readable solution and presenting it back to the user. All this is done within seconds due to the speed of the current signal transmission and data transfer.

is it clever? Sure. But it's still just what it is, 'Artificial Intelligence'. It it wasn't the for developers behind it nothing would happen and we would of never heard of it. Chat GPT cannot work without it's buddy system of the Internet to cobble together it's results so as an intermediate A.I it would fail in the real world outside of touch-screen technology and smartphones. Anyone could write a clone of Chat GPT (many have) and they operate just in the same way. It has it's benefits but in my own view I do feel it has made certain sectors of information finding and writing rather lazy. Why put pen to paper when you can get a computer application to write it for you?

Moving away from Chat GPT, if we look at A.I from a much broader concept. should we worry about it taking over our jobs? Well, this is a difficult question to answer. There are positions in some jobs that A.I can and currently does operate in. Look at car factories. Welding is done via huge mechanical arms that are using a very basic for of A.I - call and send. A message is sent to the CPU of the arm which instructs it how to move, where to move and when to stop/start etc. A.I can impact other roles that are taken up by humans as well. Many people who work in medical, transport and software will of course be concerned about A.I taking some of their roles in the futute.

My answer remains the same even to this day. A.I is only as powerful as the team of people behind it. That is why self-aware A.I is almost impossible to develop. You can give a computer list of set commands to carry out, and it will do so without any argument. Developers can modify and maintain said instructions at anytime if they have access to the source code. For something to be made "self aware" it needs to be able to copy itself to carry out a task and not break that cycle. This seems possible, but then you have to consider, again, that these instructions to be "able to write the instructions" must be hard coded into the software before hand.

Because programs can always be modified, there will always be a 'kill switch' in this respect.

Let's look at a very basic example.

Code:

while Res = True loop
    Action := Do_Something();
    exit when Do_Something() = False;
end loop;

The above will carry on executing until the function Do_Something() returns a Boolean value of False. When that happens, the loop iteration will terminate and cease functioning. It's the same for A.I in this respect. It will always have a way to terminate, and because today's software tools will never run a loop forever (eventually it would write to the stack too much and crash) then stopping any sort of A.I always has a loop hole of termination. When you switch on your computer for example, your BIOS (Basic Input Output System) self checks all the RAM sticks, detect their is a working CPU and GPU, check the validity of the drives (HDD and SSD), request that a mouse and keyboard are present and if needed, clear the CMOSS battery.

It does all this even before it posts something to the screen, and if a single one of those above checks fails, it will not "post" and start Windows up. It will generally beep in a certain order to indicate the problem, or if the problem is not critical it will post but boot the PC into 'Safe Mode'. Which allows you to tinker with the BIOS settings to find the fault. So what's the point of all this boring tech waffle? It matters because it proves no software is faultless. For an A.I to be self-aware it would need to be capable of never failing, and let's be honest as a race we have never designed/built any sort of system that has failed at least once.

I have a lot of time for A.I and it's advances. But I think as a race, we should always be on the side of caution, Technology is a wonderful thing, and computing has come along way since Ada Lovelace wrote the first ever executable instructions for Charles Babbage. But it needs to be reigned it at all times. Advances in how we move forward must always be safety critical.

 

That's pretty low-level software. If your looking that far into how a mouse would act, then decide on directions etc. I can only guess it was probably Assembly Language or the bare bones of C as it's as close to the 'metal' as you can get. Instruction sets would of been in a matrix, as of

"Move Up"
"Move Left"

Then it would of had more detailed sub programs such as

"Is this A Wall?"
"Is this the Exit?"

I think those are brilliant exercises to do, and it really makes you try to understand how the mouse would move/think/react. Which is what makes machine learning/CS so much fun.

I would of guessed the issue it had was the numbers. It depends how deeply it understands the difference between metrics of a length, width and area. If you had written the sample sizes you needed in an equation I very much doubt it would of made much of a difference. As you said, Chat GPT is still in it's infancy, and will probably take a few years to be useful outside of deciding what to have for breakfast.

 

Spot on, second Paragraph. Love ya RC x

 

Last sentence is spot on especially.

 

I cannot believe it did not pick up on your spelling ability! It more or less scans all threads it can find then prints a summary.

 

Ok I think this concentrated more on role here than my posts... oh well!

 

It didn't figure out your love for Air Fryers Moog... that's a no from me!

 

I do not personally feel you have much to fear reids. A computer is not 100% correct and has it's faults. It can only produce what it's been told to produce and there is no way that a computer program (which is what AI is) can think of every single possible idea that a human can. Look at chess engines - they are good, well designed but you can always win or find an effective strategy. You make you're role your own - you put into what you feel is required. A computer cannot read a person's thoughts, and A.I is far, far away from doing anything that can change a person's perspective on what they do or feel.

It may seem like your fighting against an unstoppable force, but as long as you present your findings at a high level and show your methods, you will win most battles in that regard. Software developers can be lazy, especially ones who code massive systems. A lot of the code is copy pasted from old legacy code and modified. If every single A.I had to produce it's methods of "how and why" it came to said conclusion it would have to declare why it's methods were better than a humans. It can't do that, and that is why you will still have a role because unlike software, you have a freedom of speech.

 

Close to the metal is a term used for a language that is very low level, or in most cases, the least readable high level code. The only language above pure binary 1s and 0s is Assembly, which is sets of instructions that use some very complex data pooling to get workloads done. NES games were coded in pure Assembly (or a variable of the language that the NES hardware could understand). No human can read and write binary code, it's (almost) impossible to determine which sets of "off" and "on" will do what.

You are correct however in that a compiler merely translates the human written language code into machine language that a computer can understand. When a program is compiled all binary is stored in an object file which is the only file the computer reads. A compiler uses a symbol checker to read through the code line by line in accordance to the language specifications (written into the compiler software) and parses it line by line. When you talk about higher level languages (such as C# and Java) they use a second level of abstraction above the compiler like the MSIL (Microsoft Intermediate Language) which act like a layer above the translated code of say a language written in C. This is translated into byte-code then translated into machine code.

I do however fully commend you if you can write machine code flawlessly, I would of thought it would resemble something like this:



which is a level above pure binary 0010011100 etc.

Fridges have micro processors (most, not all) that calculate the timers. Those are written in assembly (machine code) or an embedded version of C which is specially aimed at a certain platform/hardware specification.

 

Interesting. Sounds like Moonpig have used the Chat GPT formula in a very simple format. Somebody still has to put the data into the A.I before it can produce a result. Therefore I would like to believe that any position where you are being asked "how an A.I can improve how we do our jobs" is a subtle nod to the research and development teams. One of the best ways to think about that is to decide what you can do without computerised aid. How you can (if possible) improve your output and cut down on work loads. The better a company can operate without the need for an A.I the better - it will save jobs and possibly create new ones in the process.

 

I have respect for anyone who can code in C safely, using the memory allocation properly and not abusing function overloading. C is still one of the most used languages in the world due to it's speed. However, one thing (as you rightly pointed out) is that Assembly will not allow you to make a mistake. As it translates to instant binary without the need for a compiler any program written will instantly crash if it has the slightest fault.

This is perfectly "legal" C code according to GNU:

Code:

typedef struct {
    char *str;
    int x, y, z;
} Demo;

What is str an array of? Is it an array of strings? Pointers to strings? Single characters? Who knows what the pointer of the char array is pointing to before it's used. One of the reasons we mainly use Ada at SpaceX is because of the total opposite of the above. You have to declare what everything is. Everything is a type - it has it's ranges, it has exception handling done right. A constraint error won't crash a vital program because you can loop through the data until the correct value is found.

Assembly can do everything Ada and C can, but quicker and type safe.