AI education…

[Lifesbloodygood]

I really fancy learning something new, I have a lot of time on my hands and ai fascinates me, i know zero about it but have a good understanding of human phycology and think the correlation between the two could be very interesting as time passes.

I have no idea on where to go and what to read from ground zero on this subject though, anyone got any pointers on where to start?

I’d prefer printed books and courses to just reading on the internet as its a bit too flippant for me really and i trust less and less the internet as a resource these days.

[Mr Whippy]

I’d check out some of Stephen Wolframs videos on YT.

Very much the programmatic and logic side of stuff… and a way to properly get how all these latest NNs are working (or not)

[tangerine_sedge]

Lifesbloodygood said:

I really fancy learning something new, I have a lot of time on my hands and ai fascinates me, i know zero about it but have a good understanding of human phycology and think the correlation between the two could be very interesting as time passes.

I have no idea on where to go and what to read from ground zero on this subject though, anyone got any pointers on where to start?

I’d prefer printed books and courses to just reading on the internet as its a bit too flippant for me really and i trust less and less the internet as a resource these days.

I know you explicitly stated that you wanted real books, but I think the best starting point is the wikipedia article on AI. That's a basic introduction to various tools/areas/topics which are considered to be within the realm of AI, you can then work your way through some of the linked sub topics too.

Once I realised that at a basic level AI is just data+clever maths I started to really understand what AI is all about and why currently it's not really AI.

[Lifesbloodygood]

Thanks..

Yeah, not sure we ever want true ai but it’s growth toward self awareness (if it ever happens) and the input needed from us on that journey, really interests me

Understanding its origins and why all of a sudden it’s called ai vs it’s just programming that learns from our inputs, which is the same as any other program ever created, is slso interesting

I’d like to use it as a tool but understanding its origins properly is of course paramount

[miniman]

It might sound counterintuitive, but sign up for ChatGPT and ask it how it works.

[Mr Whippy]

Lifesbloodygood said:

Thanks..

Yeah, not sure we ever want true ai but it’s growth toward self awareness (if it ever happens) and the input needed from us on that journey, really interests me

Understanding its origins and why all of a sudden it’s called ai vs it’s just programming that learns from our inputs, which is the same as any other program ever created, is slso interesting

I’d like to use it as a tool but understanding its origins properly is of course paramount

If you want more into that kind of stuff you want Roger Penrose a bit too.


But even Wolfram is trying to use NN AI to populate a network to try create a deterministic view of the universe.

Which then tire across a bit wrt Penrose and quantum consciousness.


There is a good chance proper AI won’t be along until quantum computers are really decent, since it looks like that conscious element might be a quantum phenomenon, and up to that point the universe is just deterministic.

[robinessex]

The computer world is following its trend of misnaming 'stuff'. It's following the most misnamed department in any organisation, the IT Help Desk.

[Hoofy]

To learn about it, you might want to have a play with it but also think about how you can use it in your life, rather than just learning about AI or learning to code it. FWIW I'm interested in learning how it can benefit me but I'm not interested in learning how to code. Playing with it, testing ideas, treating it like a person wearing numerous hats - these all help me to understand what is possible.

[TheBinarySheep]

AI is broad term. Is there anything specific you'd like to learn?

I'd start reading and watching videos about neural networks.

Maybe learn the basics of the Python programming language, as it will allow you to create your own Neural networks so you can figure out how they work, and how different parameters impact the accuracy of predictions.

Then there's the whole generative side of the things (ChatGPT).

I'm a developer, and I've recently starting taking some IBM courses on Coursera so I can figure out how I can integrate AI into business workflows (training/fine tuning etc).

Edited by TheBinarySheep on Tuesday 21st January 11:03

[Mr Whippy]

Lifesbloodygood said:

Thanks..

Yeah, not sure we ever want true ai but it’s growth toward self awareness (if it ever happens) and the input needed from us on that journey, really interests me

Understanding its origins and why all of a sudden it’s called ai vs it’s just programming that learns from our inputs, which is the same as any other program ever created, is slso interesting

I’d like to use it as a tool but understanding its origins properly is of course paramount

If you want more into that kind of stuff you want Roger Penrose a bit too.


But even Wolfram is trying to use NN AI to populate a network to try create a deterministic view of the universe.

Which then tire across a bit wrt Penrose and quantum consciousness.


There is a good chance proper AI won’t be along until quantum computers are really decent, since it looks like that conscious element might be a quantum phenomenon, and up to that point the universe is just deterministic.

[Mr Whippy]

Hoofy said:

To learn about it, you might want to have a play with it but also think about how you can use it in your life, rather than just learning about AI or learning to code it. FWIW I'm interested in learning how it can benefit me but I'm not interested in learning how to code. Playing with it, testing ideas, treating it like a person wearing numerous hats - these all help me to understand what is possible.

The Wolfram lectures will be good though because you basically see the method behind The Turk.

It’s great to pick up a tool and do jobs with it. But to understand its limitations and capabilities, especially at this very early point (still) requires understand how it even works.

And I use the term Turk because although it looks very intelligent, it’s really not.
It’s just very good at looking at probabilities of what comes next.

[TheBinarySheep]

Lifesbloodygood said:

Thanks..

Yeah, not sure we ever want true ai but it’s growth toward self awareness (if it ever happens) and the input needed from us on that journey, really interests me

Understanding its origins and why all of a sudden it’s called ai vs it’s just programming that learns from our inputs, which is the same as any other program ever created, is slso interesting

I’d like to use it as a tool but understanding its origins properly is of course paramount

There's a difference between AI and programming. When you program you're not teaching the software to learn, but rather act based on a predefined set of a rules that you've programmed into it. The process follows more of a logic flow, where if this then do that.

AI, or at least a subset of it is modelled on the human brain. The AI is fed data and rather then it being told how the solve the problem, it analyses the data and finds patterns itself (supervised/unsupervised training). It does this using neural network where weights are automatically applied within the layers until the output from the network matches the training data. Then the hope is that you can feed the AI data that it doesn't know about, and it can predict the outcome.

One example that was in a course I did recently, was where we fed a model data on different mixtures of concrete, along with the known strength. Once the model was trained you could pass different variations of concrete mixture and it could predict it's strength.

[Hoofy]

Mr Whippy said:

Hoofy said:

To learn about it, you might want to have a play with it but also think about how you can use it in your life, rather than just learning about AI or learning to code it. FWIW I'm interested in learning how it can benefit me but I'm not interested in learning how to code. Playing with it, testing ideas, treating it like a person wearing numerous hats - these all help me to understand what is possible.

The Wolfram lectures will be good though because you basically see the method behind The Turk.

It’s great to pick up a tool and do jobs with it. But to understand its limitations and capabilities, especially at this very early point (still) requires understand how it even works.

And I use the term Turk because although it looks very intelligent, it’s really not.
It’s just very good at looking at probabilities of what comes next.

Thanks, will have a look.

[Lifesbloodygood]

Mr Whippy said:

Hoofy said:

To learn about it, you might want to have a play with it but also think about how you can use it in your life, rather than just learning about AI or learning to code it. FWIW I'm interested in learning how it can benefit me but I'm not interested in learning how to code. Playing with it, testing ideas, treating it like a person wearing numerous hats - these all help me to understand what is possible.

The Wolfram lectures will be good though because you basically see the method behind The Turk.

It’s great to pick up a tool and do jobs with it. But to understand its limitations and capabilities, especially at this very early point (still) requires understand how it even works.

And I use the term Turk because although it looks very intelligent, it’s really not.
It’s just very good at looking at probabilities of what comes next.

This I think is my angle, where are the preferences leaning now were a bit further in and why?

My wifes new start up has its own bot, whereupon it’s learning is from specific input directly related to her industry but also it outputs in a manner that it’s been taught, so as to have the companies personality about it, absolutely fascinating to see how fast it learns day by day and how it deals with being corrected, only 3 months old and can be told to respond in a company way or either of the co-founders way.

I’ll do some research, thx for the tips

[TheBinarySheep]

Lifesbloodygood said:

This I think is my angle, where are the preferences leaning now were a bit further in and why?

My wifes new start up has its own bot, whereupon it’s learning is from specific input directly related to her industry but also it outputs in a manner that it’s been taught, so as to have the companies personality about it, absolutely fascinating to see how fast it learns day by day and how it deals with being corrected, only 3 months old and can be told to respond in a company way or either of the co-founders way.

I’ll do some research, thx for the tips

It's certainly interesting.

It's possible that your wife's AI is setup with a foundation model (i.e. ChatGPT), and a knowledge base (industry specific for example). When you ask it a question it first goes to the knowledge base to extract any data it thinks is useful to the question, then feeds that, and your original question into the foundation model which spits out a response and makes you believe it already new that information. It didn't, it was given to it with your question.

You can also give models examples of how you'd like it to respond, these could be sample emails, and you can ask it to formulate the response in the same tone/manner.

The above is context learning. The alternative approach is fine-tuning which involves training the model on additional data so it can specialise in a given task, however this is vastly more expensive to do and requires a lot more processing power to train the model with this new information.

[ATG]

Mr Whippy said:

Lifesbloodygood said:

Thanks..

Yeah, not sure we ever want true ai but it’s growth toward self awareness (if it ever happens) and the input needed from us on that journey, really interests me

Understanding its origins and why all of a sudden it’s called ai vs it’s just programming that learns from our inputs, which is the same as any other program ever created, is slso interesting

I’d like to use it as a tool but understanding its origins properly is of course paramount

If you want more into that kind of stuff you want Roger Penrose a bit too.


But even Wolfram is trying to use NN AI to populate a network to try create a deterministic view of the universe.

Which then tire across a bit wrt Penrose and quantum consciousness.


There is a good chance proper AI won’t be along until quantum computers are really decent, since it looks like that conscious element might be a quantum phenomenon, and up to that point the universe is just deterministic.

There is no evidence that consciousness arises from or requires quantum behaviour. Penrose is a cast iron polymath genius, but on this I think he is barking up the wrong tree.

There are many, many more insane takes on the origin of consciousness than his that also get taken fairly seriously ... E.g. electrons are a bit conscious, gravel is a bit more conscious, mice are nearly there, then people from the north east, then finally the rest of humanity.

Edited by ATG on Wednesday 22 January 12:58

[ATG]

There is a tendency to look at how a particular type of "artificial intelligence" works and then dismiss it as not really being an example of "intelligence". For example, if you want back 50 years and asked someone if beating a grandmaster at chess would be an example of intelligence, they'd say yes. But given a small piece of plastic can do that these days, we don't think of it as being intelligent; it's just good at chess.

What this really reveals is what a wishy washy term "intelligence" is. The same goes for "consciousness". Wondering if something we've built has a property that we can't really define seems to me to be pretty unhelpful.

[Arnold Cunningham]

I'm doing a PhD which will inevitably involve some of what is commonly called AI.
Main thing I've learnt is that AI is a general catch all term for what is frequently quite simple maths - albeit sometimes performed many many times.
For anyone interested in it - what - specifically, what are you interested in. AI is such a throwaway catch-all phrase that means little, on it's own.

Convolutional or Reinforcement Neural Networks are quite cool. My personal direction at the moment is to use deep learning (multi-layer neural networks) to do reinforcement learning (learning by having less rules and just trying stuff to see what happens), but to do so in a cross-domain manner. Example of this is computer games - if you learn frogger, while pac-man is different, many of the principles a human learnt to play frogger would also apply to pac-man - but computers are pretty bad at using knowledge from one to apply to another - I might go down this path on my PhD, we'll see.

Edited by Arnold Cunningham on Wednesday 22 January 14:07

[Hoofy]

As an aside, a lot of what people call AI feels like lots of "if... then" statements.

Arnold Cunningham said:

I'm doing a PhD which will inevitably involve some of what is commonly called AI.
Main thing I've learnt is that AI is a general catch all term for what is frequently quite simple maths - albeit sometimes performed many many times.
For anyone interested in it - what - specifically, what are you interested in. AI is such a throwaway catch-all phrase that means little, on it's own.

Convolutional or Reinforcement Neural Networks are quite cool. My personal direction at the moment is to use deep learning (multi-layer neural networks) to do reinforcement learning (learning by having less rules and just trying stuff to see what happens), but to do so in a cross-domain manner. Example of this is computer games - if you learn frogger, while pac-man is different, many of the principles a human learnt to play frogger would also apply to pac-man - but computers are pretty bad at using knowledge from one to apply to another - I might go down this path on my PhD, we'll see.

Edited by Arnold Cunningham on Wednesday 22 January 14:07

Bear in mind that I don't have a PhD and the last time I did degree level maths was 30 years ago, can you explain how it's "simple maths"?

[Arnold Cunningham]

Hoofy said:

As an aside, a lot of what people call AI feels like lots of "if... then" statements.

Bear in mind that I don't have a PhD and the last time I did degree level maths was 30 years ago, can you explain how it's "simple maths"?

If then else : pretty much a decision tree

If I use Neural networks as an example they do this:

Take some some inputs (for the sake of this example, with input values between between 0 and 1)
Adding a weighting (a multiplication) factor to those inputs, summing them up, adding then a bias factor (offset),
If the sum of the weighted inputs, plus bias, exceeds a trigger value, then the neuron fires and gives an output.

I am simplifying - there IS of course lots of complexity - this is an individual neuron, but a full implementation may be 10's of thousands of them across multiple layers - but in isolation, the maths itself is simple, it's the recursion/depth/optimisation of it all that is the rocket science. ie the Challenge is to work out how many neurons you need, across how many layers, and what the weights and biases you need are - but the maths itself isn't too hard, it's just working out what combination of the maths creates something useful.

This is a great introduction to Neural Networks for handwritten digit recognition : https://neuralnetworksanddeeplearning.com/
It's written in python & works really well. IIRC you can squeeze if into something like 80 lines of python code - but I bet many of the guys here who work in IT are used to many thousands of lines of code for integration & automation scripts that on the face of it, don't do that much!

I haven't got a PhD either - I am fairly early on with being a mature student & going back to university to learn stuff again. My main lesson so far is that being old sucks, learning is so much harder than it used to be, for me.