The Artificial Intelligence Thread

[Asbo Zaprudder]

Good job then that I'm not a mathematician. It sounds a tedious task.

The Riemann series theorem states you can derive e, pi, or any arbitrary real number by rearranging the terms of any infinite series of real numbers that is conditionally convergent to some other real value such as ln(2) = 1 - 1/2 + 1/3 - 1/4 + 1/5... You can also make the same series diverge if you wish - again by straightforward permutation of the order of the terms. It's weird but it is what it is.

https://en.m.wikipedia.org/wiki/Riemann_series_theorem

I don't know if the AI software you mention uses such a technique. I might have a look later today.

ETA: Having read the article, it seems that the AI can suggest but doesn't prove new formulae. I guess it's a tool that might be useful under some circumstances and a distraction under others. I'd be more impressed if it could derive proofs. Mathematics is just symbol manipulation under a given set of rules FFS.

 

[CorporalCaptain]

Mathematics is just symbol manipulation under a given set of rules FFS.

Gödel's incompleteness theorems imply that there's more to it than that, particularly when it comes to any attempt to limit mathematics to "a given set of rules."

AI is now a Math Whiz conjecture creator.

The Ramanujan Machine

It creates new methods of deriving fundamental constants like pi or e and it is up to mathematicians to prove or disprove the conjecture that the new methods can successfully recreate those constants.

@CorporalCaptain
@Asbo Zaprudder

That's interesting, but Zeilberger's claim of humanity's eventual obsolescence is premature, to say the least.

 

[Asbo Zaprudder]

Kurt Gödel's incompleteness theorems state that in any consistent formal system within which non-trivial types of arithmetic can be performed, there are statements of the language of that formal system that can neither be proved nor disproved and that such a formal system cannot prove that the system itself is consistent. There's also Alfred Tarski's indefinability of truth theorem, Alonzo Church's proof of the insolvability of David Hilbert's Entscheidungsproblem*, and Alan Turing's theorem that there is no algorithm to solve the halting problem. You can vary the symbols and the rules but you can't get round the limitations of such formal systems in mathematics. Maybe there's something beyond mathematics that we've yet to invent or discover, depending on your philosophy.

I think that claims for the current state of AI are mostly hype. A lot of it currently seems like pattern matching to me, particularly the neural network-based stuff. Nudge me when it can do meta^Nth-level inference and actually explain how it has come to a particular conclusion.

Anyway, I feel I'm not competent and knowledgeable enough to comment meaningfully on this stuff so I'll bow out now.

* The Entscheidungsproblem seeks an algorithm that can determine whether a formulaic statement is provably true or false for a given set of axioms in a formal system.

 

[rahullak]

That's interesting, but Zeilberger's claim of humanity's eventual obsolescence is premature, to say the least.

Yes, I agree. Most AI systems today are still hand-designed to a high degree, and until they get so advanced as to require little to no human intervention/supervision, these claims will remain premature. Not even getting into whether it is even possible for a compute system to achieve all levels of human cognition in a generally intelligent way. And then too there's the question of where creativity comes from and whether we ourselves are "lights and clockwork" albeit of a different kind, or is there an ineffable, irreplaceable quality to us humans.

I think that claims for the current state of AI are mostly hype. A lot of it currently seems like pattern matching to me, particularly the neural network-based stuff. Nudge me when it can do meta^Nth-level inference and actually explain how it has come to a particular conclusion.

Judea Pearl's* work on causality might prove to be a promising direction. The Book of Why, which I am currently reading, explains how traditional statistics is simply about correlation and pattern matching, with humans hand-designing the experiments to remain casually appropriate (rather than gibberish such as: the rooster crows and the sun rises approximately every day. Therefore the rooster causes the sun to rise). What is required is a causal framework with the do operator which can be used programatically to tease out causality not only from static data also by conducting experiments that make sense.

Another aspect of AI is explainability. I don't think it has reached too far, but I'm not very well informed as to the results of these endeavours yet.

There's a lot for me to learn in this field along with the Data Science field in general. It's interesting having these conversations though.


*side note: Judea Pearl is the father of Daniel Pearl, who you may be hearing of in the news lately, as his terrorist decapitators in Pakistan are being let off easily.

 

[CorporalCaptain]

Kurt Gödel's incompleteness theorems state that in any consistent formal system within which non-trivial types of arithmetic can be performed, there are statements of the language of that formal system that can neither be proved nor disproved and that such a formal system cannot prove that the system itself is consistent. There's also Alfred Tarski's indefinability of truth theorem, Alonzo Church's proof of the insolvability of David Hilbert's Entscheidungsproblem*, and Alan Turing's theorem that there is no algorithm to solve the halting problem. You can vary the symbols and the rules but you can't get round the limitations of such formal systems in mathematics.

Exactly. Formal systems are too weak for any one of them to capture all mathematical truths.

Maybe there's something beyond mathematics that we've yet to invent or discover, depending on your philosophy.

However, this is a non sequitur.

Mathematics is just symbol manipulation under a given set of rules FFS.

What you're saying here is untrue. I can only suppose that you are led to consider the question of something beyond mathematics in the context of Gödel's incompleteness theorems in light of your assumption or belief that mathematics is just symbol manipulation under a given set of rules.

But what it is that mathematicians contemplate cannot be fully represented as a formal system, so the question is really more one of what there is to mathematics beyond symbol manipulation under a given set of rules. Mathematics was done prior to peoples' efforts to reduce mathematics simply to symbol manipulation under a given set of rules, it was being done while people were engaged in those efforts, and it is still being done after it has been shown that those efforts are incapable of completely succeeding.

My thoughts on the question of what mathematics is beyond formal systems involve the conclusion that mathematics is a science.

However, referring to the Wiki on formal science [https://en.wikipedia.org/wiki/Formal_science#Differences_from_other_forms_of_science]:

One reason why mathematics enjoys special esteem, above all other sciences, is that its laws are absolutely certain and indisputable, while those of other sciences are to some extent debatable and in constant danger of being overthrown by newly discovered facts.

— Albert Einstein​

As opposed to empirical sciences (natural and social), the formal sciences do not involve empirical procedures.​

My thoughts on this are that it is all wrong. Mathematicians engage in empirical procedures when they attempt to decide whether a statement is true or false relative to a given set of axioms. The experiment is the search for proof. In practice, a mathematician does not a priori know which side of a question will be proven correct, or even whether a proof will be discovered. Further, in the background of every search for proof is the possibility of uncovering a proof of the inconsistency of the whole set of axioms under consideration; that this possibility exists is a consequence of Gödel's second incompleteness theorem.

What Einstein said there is most naïve. Again, see Gödel.

 

[Asbo Zaprudder]

Exactly. Formal systems are too weak for any one of them to capture all mathematical truths.

However, this is a non sequitur.


What you're saying here is untrue. I can only suppose that you are led to consider the question of something beyond mathematics in the context of Gödel's incompleteness theorems in light of your assumption or belief that mathematics is just symbol manipulation under a given set of rules.

But what it is that mathematicians contemplate cannot be fully represented as a formal system, so the question is really more one of what there is to mathematics beyond symbol manipulation under a given set of rules. Mathematics was done prior to peoples' efforts to reduce mathematics simply to symbol manipulation under a given set of rules, it was being done while people were engaged in those efforts, and it is still being done after it has been shown that those efforts are incapable of completely succeeding.

My thoughts on the question of what mathematics is beyond formal systems involve the conclusion that mathematics is a science.

However, referring to the Wiki on formal science [https://en.wikipedia.org/wiki/Formal_science#Differences_from_other_forms_of_science]:

One reason why mathematics enjoys special esteem, above all other sciences, is that its laws are absolutely certain and indisputable, while those of other sciences are to some extent debatable and in constant danger of being overthrown by newly discovered facts.

— Albert Einstein​

As opposed to empirical sciences (natural and social), the formal sciences do not involve empirical procedures.​

My thoughts on this are that it is all wrong. Mathematicians engage in empirical procedures when they attempt to decide whether a statement is true or false relative to a given set of axioms. The experiment is the search for proof. In practice, a mathematician does not a priori know which side of a question will be proven correct, or even whether a proof will be discovered. Further, in the background of every search for proof is the possibility of uncovering a proof of the inconsistency of the whole set of axioms under consideration; that this possibility exists is a consequence of Gödel's second incompleteness theorem.

What Einstein said there is most naïve. Again, see Gödel.

So where do you place yourself on the spectrum between Platonic Realist and Absolute Nominalist - or do you take a view similar to Roger Penrose's Platonic circular, non-hierarchical trinity of mental, physical, and mathematical realms? I favour a Nominalist viewpoint and it sounds like you do as well. I think we only disagree on what constitutes mathematics and I concede my definition is open to dispute. I think we might yet invent some other system for describing reality beyond current mathematical systems but I have no idea when and how that might happen. Perhaps we'll need a superbright AI to do so and our brains might not understand what it tells us.

 

[Santaman]

Oeh! I recognize a name.. Penrose! yes! his tribar/triangle will probably be my next tattoo..
As for math.. zero talent for it..

 

[CorporalCaptain]

So where do you place yourself on the spectrum between Platonic Realist and Absolute Nominalist - or do you take a view similar to Roger Penrose's Platonic circular, non-hierarchical trinity of mental, physical, and mathematical realms? I favour a Nominalist viewpoint and it sounds like you do as well. I think we only disagree on what constitutes mathematics and I concede my definition is open to dispute. I think we might yet invent some other system for describing reality beyond current mathematical systems but I have no idea when and how that might happen. Perhaps we'll need a superbright AI to do so and our brains might not understand what it tells us.

Oh, goodness, I don't think I could briefly state my philosophical beliefs. But I can perhaps outline some points that might be important. I doubt that my thoughts align with any particular school of thought entirely.

The success of our scientific theories itself empirically supports the idea that our abstract concepts have reality, at least to the degree that they really occupy our minds during the act of theorizing.

I've gone back and forth on the question of whether abstractions exist apart from our minds. There's a passage of Einstein's in The Meaning of Relativity, boldfaced below, that I think is quite accurate [http://www.gutenberg.org/files/36276/36276-pdf.pdf]:

The only justification for our concepts and system of concepts is that they serve to represent the complex of our experiences; beyond this they have no legitimacy. I am convinced that the philosophers have had a harmful effect upon the progress of scientific thinking in removing certain fundamental concepts from the domain of empiricism, where they are under our control, to the intangible heights of the a priori. For even if it should appear that the universe of ideas cannot be deduced from experience by logical means, but is, in a sense, a creation of the human mind, without which no science is possible, nevertheless this universe of ideas is just as little independent of the nature of our experiences as clothes are of the form of the human body. This is particularly true of our concepts of time and space, which physicists have been obliged by the facts to bring down from the Olympus of the a priori in order to adjust them and put them in a serviceable condition.​

By the boldfaced point, it is not relevant whether mental abstractions per se exist apart from our minds, at least if not especially in the case of those that are abstracted from experience, and at least in that they mirror the reality that does exist apart from our minds. In other words, I think that there's something out there that is reflected in our minds. Again, the success of our scientific theories provides support for the idea that this is so.

Moreover, I don't think there is any reason to suppose a primary distinction between mind and body. Nothing that I'm aware of that refutes the proposition that all mental activity has a physical manifestation (i.e., most likely but not yet assuredly all of which is in the nervous system). It therefore follows that mental abstractions exist physically in a way that corresponds to their mental qualities and attributes. By their very nature as physical phenomena, they obey relations that are not all strictly mental in nature (as the mind itself is therefore not an isolated system, and the mind is in physical contact with other systems, not all of which are mental), and so concepts themselves transcend the mind. Additionally, cognition does not occur without relation to bodily activity.

As one example regarding the question of the existence of universals, color is an aspect of perception that exists as a potential response to stimulus even without stimulus. Not entirely dissimilarly, an electrically charged particle may potentially be present even where it does not exist at a particular time. If we restricted ourselves to accepting as real only that which is, and discounted what theoretically might be, then we would not be able to theorize in any way that remotely resembles the way that we do.

 

[rahullak]

Allow me to posit:

There exists a shared "mind realm" through which our mental abstractions. mental designs and imagination can be conveyed. The truthiness or degree of physical existence of what is conveyed lies on a spectrum (may be multidimensional). So, for instance if a person sees a red shirt and conveys to another person "That is a red shirt", if there is an exact match with the other person then a degree of truth is created both as a conveyed mental abstraction as well as its association with the physical object. If however, the second person is color blind or has photoreceptors that make the red shirt appear slightly different than that conceived by the first person, the degree of truthiness is less so, inasmuch as between the 2 people in question. The first person may continue to strongly believe his/her own conception of the "red shirt" is absolute, or may leave room for doubt and modification. If however both are scientists, well-versed and memorised in the wavelengths of light, they may be able to better convey what they mean by "red", then the degree of truthiness is all the stronger, because the assertion lies on a foundation of reality based on measurements of the world, which can be measured universally (by which it is meant that no matter who conducts the experiment on finding the wavelengths of light emanating from the shirt will draw the same conclusion as to which wavelengths are missing, thus rendering the red shirt "red" in a more objectively measurable way)

In this way, not only mental abstractions associated with physical objects can be conveyed and have truthiness, but also abstractions that have no physical object, whether the abstraction is a design to be worked upon to create a physical object (such as a bridge of a software system), or remains merely an abstraction (for example the theoretical sciences).

 

[Asbo Zaprudder]

A Platonist and a Nominalist walk into a bar. The Platonist says "Ouch, that's an iron bar." The Nominalist sits down and orders a drink. Roger Penrose observes all this and writes another book.

I have no evidence to decide in favour of either philosophical viewpoint. I can't state "Eppur si muove" nor will kicking a rock and exclaiming "I refute it thus" help resolve the issue.

In reference to qualia, there is some experimental evidence in humans that having a name for something strengthens the synaptic networks for identifying that thing - for example, a colour such as "pink". Perhaps ascribing a label allows one to recognise a feature of the world more easily. Evolution might have favoured this mechanism in the hominid brain if it increased the chance of surviving long enough to procreate. The ancient Greeks had no word for "blue" - Homer described the sea as "wine-dark". Maybe they could only perceive the colour when it was given a name.

However, if one thinks hard about "octarine" - the eighth colour of the rainbow on Terry Pratchett's Discworld - will one start seeing it? I already see a colour beyond violet but I don't expect anyone to believe me as I have no proof. It's not the greeny-black that Pratchett describes. It's more an electric bluey-pink. I haven't given it a name. How about "bowie"?

 

[Gingerbread Demon]

Someone did do ratbot. Rat brain in a bot.

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

 

[Captain_Nick]

Hello everyone!

A.I is now a fast-moving field and there are so many advancements being made big and small. Almost every year we hear of something cool. Whether it is in game-playing systems like AlphaZero, Watson etc. or new research advancements such as Deep Learning, GANs, NLP etc., or cool new applications like automated driving, text generation / chat bots, life-like robotics, medical diagnosis etc., the topic is vast and ripe for discussion.

This is also a very interdisciplinary field. A.I influences and is influenced by numerous fields such as neuroscience, linguistics, sociology, philosophy of mind, data science, general business etc. Not to mention that any field that requires data-crunching is today bound to use machine learning in some form or other - whether that is astronomy or genetic engineering or really any data-rich field.

So creating this general thread to share and discuss anything and everything broad and deep about A.I. and have a place to post all of the happenings around this field. And yes that includes books, movies, entertainment in all its forms as well!

To start off, here's an interesting article someone had linked to on another website:

Can a machine have empathy?

While the article was created to discuss the implications for marketing, the points raised are thought-provoking in a more general sense.

Is empathy something deeper than reading the surface reactions of people? If not, then can A.I ultimately be able to mimic that and therefore be said to have empathy? What does it mean to say that we understand something, whereas a silicon-based machine cannot?

Max Tegmark's view is that we humans are a configuration of physical particles (or waves), configured in a way that allows such things as awareness, understanding and empathy. Why should it not be possible that a similar configuration cannot be constructed artificially?

I think that the ideas that we think of know that we have of consciousness, understanding, soul etc. goes beyond the sum of our quarks, but that it is not clear that it is impossible for us to artificially create it either using silicon or other biologically inspired material.

What are your thoughts and comments? Have you found cool new A.I science, applications or advancements today?

None of this is really true though, is it. Most of the hype around AI seems to be centred on simple pattern matchers and statistical models. There haven't been any significant developments in the field in decades. Artificial intelligence is as far away today as it was in the 60s when they were telling us humanoid robots would be doing our housework.

 

[rahullak]

None of this is really true though, is it. Most of the hype around AI seems to be centred on simple pattern matchers and statistical models. There haven't been any significant developments in the field in decades. Artificial intelligence is as far away today as it was in the 60s when they were telling us humanoid robots would be doing our housework.

It's not true that there haven't seen significant developments in decades. Just in the past decade, with the advent of Deep Learning, we have improved our models by leaps and bounds. Whether it is in Image/Object/Face Recognition, Video Processing, Natural Language Processing, Natural Language Generation etc. Auto-driven cars have gone from a nice geeky hobby to production vehicles that run thousands of miles without operator input, to name one specific example. And these won't be termed "simple pattern matches" by people with knowledge of it. Now, almost in every endeavour we can think of, business or science, ML/AI based data science/analytics is being adopted because the technology has been proven to be useful. These are all Artificial Narrow Intelligence ie. the use of AI techniques applied to specific use cases and designed by humans.

To realise that significant advancements have been made, compare the state of AI technology today with 15 or 20 years ago. Now, is there an element of hype? Granted, there is. Especially since once the media gets hold of a story, they like to sensationalise. But the fact that something is being hyped doesn't automatically mean it is all humbug.

Now what you seem to be indicating is Artificial General Intelligence: like a Data or a Sutra. Yes, this is far away likely a few/some decades away by my reckoning. But there's steady progress being made by DeepMind and OpenAI towards AGI and tech companies and entrepreneurs are pouring many millions into these companies and projects.

 

[Asbo Zaprudder]

The human brain contains approximately 85 billion neurons operating in parallel. I expect we need to get to that order of complexity to realise generalised AI. If Roger Penrose is to be believed we might also need quantum computing. He's a skeptic that AI is possible but in my view his reasoning (a product of real intelligence or RI) is fundamentally flawed. Evolutionary selection, possibly coupled with multiverse selection, managed to come up with RI at least once. I don't see why RI can't emulate that in a much shorter time frame once the principles are understood.

 

[publiusr]

Here is something I’d like to ask about the use of AI in terms of model kit building. I am told that most CGI ship models don’t lend themselves to 3D prints because the meshes themselves aren’t encoded tightly enough.

It would be a beast to try to knit all the lose ends through manual coding. But the motion smoothing I hate so much because it makes all the film Twilight Zones look like videotape—maybe that can play a role.

Imagine a virtual 3D printer, where you could do a number of runs inside such a filter...something like Unilever did with natural selection and the making of the soap nozzles. The result would be something now knitted together that you could input it into a physical 3D printer.

Would that work?

 

[Gingerbread Demon]

The human brain contains approximately 85 billion neurons operating in parallel. I expect we need to get to that order of complexity to realise generalised AI. If Roger Penrose is to be believed we might also need quantum computing. He's a skeptic that AI is possible but in my view his reasoning (a product of real intelligence or RI) is fundamentally flawed. Evolutionary selection, possibly coupled with multiverse selection, managed to come up with RI at least once. I don't see why RI can't emulate that in a much shorter time frame once the principles are understood.

How does Penrose explain human intelligence and how the brain functions?
Multiverse selection you lost me with that line, that sounds iffy.

 

[rahullak]

I think he means that in the vast multiverse intelligent life has evolved at least once. If it's possible through natural evolution there's no reason why it shouldn't be possible through artificial constructs that operate on the same principles.

 

[Gingerbread Demon]

I think he means that in the vast multiverse intelligent life has evolved at least once. If it's possible through natural evolution there's no reason why it shouldn't be possible through artificial constructs that operate on the same principles.

OK that makes more sense.. I don't see why that couldn't be possible.

 

[Asbo Zaprudder]

Been a while since I wrote that but I think if we're just material beings in a material world albeit in a multiverse, both RI and AI might be rare but not impossible. As I understand it, Penrose doesn't believe in the multiverse as he thinks the quantum wave function collapses due to gravitational self-attraction of matter at the scale of the Planck mass (about 22 microgrammes) and that human consciousness and hence RI arise from quantum effects below this level. Why that would prevent us developing AI beats me but I admit my RI is nowhere near as high as Penrose's.

Personally, I don't believe emulation of consciousness is required for AI. The metaphor oft quoted is that a submarine can travel underwater just fine without being a fish. An AI might be able to simulate consciousness just fine by employing a suitable internally generated schema to model its reality (this might be how our consciousness works as well but it's perhaps not testable as a hypothesis) but it might not need to anyway. I also believe we might use evolutionary principles to create AI without exactly understanding how we did it. We'd just need sufficient computational complexity and have it evolve through variation and our selection by breeding (like Darwin's examples of pigeons). It might even arise accidentally.

 

[rahullak]

Here is something I’d like to ask about the use of AI in terms of model kit building. I am told that most CGI ship models don’t lend themselves to 3D prints because the meshes themselves aren’t encoded tightly enough.

It would be a beast to try to knit all the lose ends through manual coding. But the motion smoothing I hate so much because it makes all the film Twilight Zones look like videotape—maybe that can play a role.

Imagine a virtual 3D printer, where you could do a number of runs inside such a filter...something like Unilever did with natural selection and the making of the soap nozzles. The result would be something now knitted together that you could input it into a physical 3D printer.

Would that work?

If you're talking about filling in the gaps so that the meshes become tighter, I think it should be doable and may not even require any Machine Learning. Maybe there's some straightforward algorithm that fills it in. Or am I understanding this wrong?

Sorry for the late response.