advertisement


AI and human communication

Tim Jones said:
If your definition of "programmer" is stupidly wide enough, it kind-of is. Most machine learning is back-propagation algorithms isn't it?
Click to expand...

My point was that it's not someone sitting down and writing a program, not even a declarative one. It's a more a complicated process or system driven by very large amounts of data and learning over an extremely high level of repetitions.

And of course to note my disdain for the it's not intelligence because *waves hands* metaphysics, the possibility of Cav coming with a more nuanced explanation notwithstanding.
 
matthewr said:
My point was that it's not someone sitting down and writing a program, not even a declarative one. It's a more a complicated process or system driven by very large amounts of data and learning over an extremely high level of repetitions.

And of course to note my disdain for the it's not intelligence because *waves hands* metaphysics, the possibility of Cav coming with a more nuanced explanation notwithstanding.
Click to expand...
AI did not occur spontaneously so is obviously a product of programming. The programme allows for learning. On that basis, it could be described as intelligent, if that is important to you.
 
matthewr said:
My point was that it's not someone sitting down and writing a program, not even a declarative one. It's a more a complicated process or system driven by very large amounts of data and learning over an extremely high level of repetitions.

And of course to note my disdain for the it's not intelligence because *waves hands* metaphysics, the possibility of Cav coming with a more nuanced explanation notwithstanding.
Click to expand...
I think the point is, someone must have explicity created or coded such systems or procesess. They didn't just spawn out of thin air. So in that sense AI and ML are most definitely the result of programming. Basically someone (people) sat down and coded something that when run had the ablity to "absorb all the data it's given and learn from it".
 
Cav said:
AI did not occur spontaneously so is obviously a product of programming. The programme allows for learning. On that basis, it could be described as intelligent, if that is important to you.
Click to expand...
gez said:
I think the point is, someone must have explicity created or coded such systems or procesess. They didn't just spawn out of thin air. So in that sense AI and ML are most definitely the result of programming. Basically someone (people) sat down and coded something that when run had the ablity to "absorb all the data it's given and learn from it".
Click to expand...

Well ok they are not magic and were indeed built by people using computers. But the thing they actually do (the AI bit) is most definitely not the result of programming but rather something much closer to statistics. I.e. the programming makes a system that when used makes an AI much like how your computer is made in a factory using engineering but nobody would say your web browser was built using engineering.

FWIW I have been a programmer for 40 years and worked in AI for nearly a decade so I think I can reasonably claim to know what I am talking about here.
 
I did say "stupidly" wide. The back propagation thing was probably invented by Fourier or someone....

The only function of programming here is to emulate something that no traditional programmer would ever do.
 
Wouldn’t we need to have a far greater understanding of how the human brain works than we do in order to know whether we had an artificial version of it?
 
Whaleblue said:
Wouldn’t we need to have a far greater understanding of how the human brain works than we do in order to know whether we had an artificial version of it?
Click to expand...

One issue here is everyone's favorite; the Turing Test. I think it probably fails quite badly with LLMs like ChatGPT (whose evidence base is everything said on the internet) because....that's all an LLM is. It has no connection to the world of things at all - except insofar as it can convince actually intelligent agents to do something in that world.

This means that these models are actually very good at some things - up to and including mental health consultations, for example (although that raises some other issues). but they can't do stuff that really simple serial computing can do, like fly a plane, etc. Machine learning is less immediately impressive, but probably much more useful. Late presentations of cancer in primary care, for example, is massively appropriate for the right kind of systems that can associate inputs and outputs.

There is a real flaw in the description of these modern systems as being "AI". Part of it is hype, and part of it (IIRC) is based on an academic confusion over semantics. They are signficantly closer to what people (usually philosophers of a certain bent) talked about 40 or 50 years ago, but still a world away from those hypotheses in some ways.
 
matthewr said:
Well ok they are not magic and were indeed built by people using computers. But the thing they actually do (the AI bit) is most definitely not the result of programming but rather something much closer to statistics. I.e. the programming makes a system that when used makes an AI much like how your computer is made in a factory using engineering but nobody would say your web browser was built using engineering.

FWIW I have been a programmer for 40 years and worked in AI for nearly a decade so I think I can reasonably claim to know what I am talking about here.
Click to expand...
Claiming expert knowledge or experience don't impress me much...
 
What do those scientists with their education, knowledge and experience know. I have my opinions and they can never take them away from me.
 
At the end of the day, regardless of if it is a ChatGPT style LLM, or Alpha Go, these are just huge statistical models. There is no AI here just machine learning.

The algorithms used, and the training methods vary depending on the use case, i.e. here is the rules of Go, and your objective is to win go play. Or here is the internet, go read everything and build a model that predicts the most likely next word when responding to a question.

The complexity of the models some of the LLMs build is huge, but it's still just a big bunch of probabilities/statistics. What makes LLMs frequently "look" inteligent is that nearly all of the models have built in "fuzzyness" which allow some leeway with the returned results, and also leads to the well reported hallucinations. This can be mitigated by various techniques if needed.

The businesses that are succesfully using LLMs have done a lot of fine tuning and re-training on their own data to deliver acceptable results, wether it's for internal use or customer use, effectively making custom models and not using the foundation ones such as GPT4.
 
The terminology is confusing.

We already have AI and we've had it for a long time ... AI is a very wide, and fuzzy, term that means ANY level of artificial intelligence implemented in ANY way. Covers the mundane, to the proposed AGI human-equivalent ("general") intelligence, to ASI "super-intelligence".

When it comes to implementations, we mostly use computers. There are algorithmic/symbolic/knowledge representation approaches - I don't think they've had the success that was first hoped. Then there is Machine Learning that uses stats/maths to "learn" desired behaviour. One type of ML is Deep Learning (deep neural networks). One type of DNN is an LLM, these have had recent great success.
 
gintonic said:
ML is an application of AI
Click to expand...

Indeed. And you can use AI to improve ML both as a model (i.e. without a programmer or programming) and in terms of hardware. This self-referential, recursive improvement is why I think progress will continue to accelerate and the "ZOMG look what they can do now!?" effects will continue.
 
matthewr said:
Well ok they are not magic and were indeed built by people using computers. But the thing they actually do (the AI bit) is most definitely not the result of programming but rather something much closer to statistics. I.e. the programming makes a system that when used makes an AI much like how your computer is made in a factory using engineering but nobody would say your web browser was built using engineering.

FWIW I have been a programmer for 40 years and worked in AI for nearly a decade so I think I can reasonably claim to know what I am talking about here.
Click to expand...
I'm sure using your own definitions of all the words you chose, you are correct. But none of that refutes or changes the fact: someone wrote some code, the result of that code is an AI or ML system. QED. People didn't paint AI or brew it. Simply put, you're being a pedant. In layman terms AI and ML are the result of programming.
 
Tim Jones said:
I did say "stupidly" wide. The back propagation thing was probably invented by Fourier or someone....

The only function of programming here is to emulate something that no traditional programmer would ever do.
Click to expand...
As may be, but nothing else did it. In exactly the same way a video game, is just a game that is the result of some code. Games existed for millenia before computers, nobody argues that computer games aren't the result/output of the work of a software engineer. Software is and alway has been a means to an end, in and of itself it produces nothing. So it really is a redundant argument to try and claim AI and ML systems as we know them today aren't the result of code/programming.
 
gintonic said:
ML is an application of AI
Click to expand...
From the standpoint of the naming conventions that we as humans have decided, then yes ML is a subset of AI. The problem is that there is no Inteligence in AI, at least not in the terms that most people think of inteligence. So what we have now is misnamed, as the only inteligence is the human written algorithims that build the models.
 
gez said:
I'm sure using your own definitions of all the words you chose, you are correct. But none of that refutes or changes the fact: someone wrote some code, the result of that code is an AI or ML system. QED. People didn't paint AI or brew it. Simply put, you're being a pedant. In layman terms AI and ML are the result of programming.
Click to expand...

It's not pedantic at all. An AI is fundamentally not a computer program in that they do not do what they do by executing a computer program. If you want to be pedantic, we could say their behaviour is a higher order function unrelated to the computer programming involved.
 
You must log in or register to reply here.


advertisement


Back
Top