Max Tegmark: AI discovers physics

49 views
Skip to first unread message

Philip Thrift

unread,
May 28, 2020, 12:20:49 PM5/28/20
to Everything List


https://www.facebook.com/461616050561921/posts/3107668729289960/


We just posted a new AI paper on how to automatically discover laws of physics from raw video with machine learning. For example, we feed in the video below of a rocket moving in a circles in a magnetic field, seen through a distorting lens, and our code automatically discovers the Lorentz Force Law. It took Silviu and me about a year to get this working, by using ideas inspired by general relativity and the the theory of knots in 5-dimensional space, so we're excited to be done! https://arxiv.org/abs/2005.11212

@philipthrift

Lawrence Crowell

unread,
May 29, 2020, 8:45:58 PM5/29/20
to Everything List
The preprint address is below. I would like to think the big question on quantum gravitation is resolved by basic human thought. Maybe AI systems can verify the theoretical result(s) and give some support.

LC
 

Symbolic Pregression: Discovering Physical Laws from Raw Distorted Video

We present a method for unsupervised learning of equations of motion for objects in raw and optionally distorted unlabeled video. We first train an autoencoder that maps each video frame into a low-dimensional latent space where the laws of motion are as simple as possible, by minimizing a combination of non-linearity, acceleration and prediction error. Differential equations describing the motion are then discovered using Pareto-optimal symbolic regression. We find that our pre-regression ("pregression") step is able to rediscover Cartesian coordinates of unlabeled moving objects even when the video is distorted by a generalized lens. Using intuition from multidimensional knot-theory, we find that the pregression step is facilitated by first adding extra latent space dimensions to avoid topological problems during training and then removing these extra dimensions via principal component analysis.
Comments:12 pages, including 6 figs
Subjects:Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Machine Learning (stat.ML)
Cite as:arXiv:2005.11212 [cs.CV]
 (or arXiv:2005.11212v1 [cs.CV] for this version)

Bibliographic data

Submission history

From: Max Tegmark [view email]
[v1] Tue, 19 May 2020 18:00:52 UTC (6,098 KB)
Current browse context:
cs.CV
< prev   |   next >
new | recent | 2005
Change to browse by:

Bookmark

BibSonomy logo Mendeley logo Reddit logo ScienceWISE logo

Philip Thrift

unread,
May 30, 2020, 3:00:16 AM5/30/20
to Everything List


Of course nature's "theory" could be beyond a human's comprehension.

It is assumed that there all that's needed can be reduced to human (mathematical) language that can be expressed in a few lines of LaTeX Math.

@philipthrift

Bruno Marchal

unread,
May 30, 2020, 6:35:52 AM5/30/20
to everyth...@googlegroups.com

> On 28 May 2020, at 18:20, Philip Thrift <cloud...@gmail.com> wrote:
>
>
>
>
> https://www.facebook.com/461616050561921/posts/3107668729289960/
>
>
> We just posted a new AI paper on how to automatically discover laws of physics from raw video with machine learning. For example, we feed in the video below of a rocket moving in a circles in a magnetic field, seen through a distorting lens, and our code automatically discovers the Lorentz Force Law. It took Silviu and me about a year to get this working, by using ideas inspired by general relativity and the the theory of knots in 5-dimensional space, so we're excited to be done! https://arxiv.org/abs/2005.11212


That is a nice exercise in AI. It is not the discovery of physics, but on some physical laws, when giving to a machine some physical data.

This has no relation with “discovering the physical laws in the head of the universal machine”, which explain where the physical data come from, and why they remain stable and eventually are explained by quantum logic and its many histories semantic.

For his, it is needed to understand the difference between the arithmetical reality/truth, and all human or not, theories we can make on this. Incompleteness explains this, like it explains why there is a mathematical reality, or, put in another way, why conventionalism is false in the philosophy of mathematics (provably so with only the CT part of mechanism, which is YD + CT. No need of thought experiments here. I will may be make a post on this, as you and Bruce defended conventionalism in mathematics some times ago.

(Soon, or a bit later, as May and June are busy for teachers, even more so when doing everything, including oral exams, “at a distance”).

Bruno





>
> @philipthrift
>
> --
> You received this message because you are subscribed to the Google Groups "Everything List" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to everything-li...@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/msgid/everything-list/48579c56-2a59-430c-b3a8-a91e29a6cd2d%40googlegroups.com.

Lawrence Crowell

unread,
May 30, 2020, 1:24:10 PM5/30/20
to Everything List
I wrote a paper recently for publication on how the unital set of QM is a Cantor/fractal set that is fundamentally incomputable. This is a measure of nonlinearity a quantum system is forced into, say with gravitation or with einselection into classicality. To compute it requires a single algorithmic system for computing all p-adic sets, where the theorem by Matyiaesovich is a form of Gödel's theorem that illustrates this does not exist. It corresponds to the unobservability of hidden variables, or that they are nonlocal, and establishes entanglement symmetries as topological indices or obstructions. This might mean we are saved by the Bell, here Bell's theorem in a sense, from the invasion of the robots. It will be some time I think before AI systems can work through self-referential inference. 

LC

ronaldheld

unread,
May 30, 2020, 3:51:32 PM5/30/20
to Everything List
Is the AI discovering some Physics or just fitting data which produces equations that look like physical laws?
     Ronald

Brent Meeker

unread,
May 30, 2020, 4:14:17 PM5/30/20
to everyth...@googlegroups.com
You could ask the same question about physicists.

Brent
--
You received this message because you are subscribed to the Google Groups "Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email to everything-li...@googlegroups.com.

Bruno Marchal

unread,
May 31, 2020, 3:37:22 AM5/31/20
to everyth...@googlegroups.com
On 30 May 2020, at 19:24, Lawrence Crowell <goldenfield...@gmail.com> wrote:

I wrote a paper recently for publication on how the unital set of QM is a Cantor/fractal set that is fundamentally incomputable.

I use the Cantor (triadic) set, or the Baire space. But it is more the measure on the possible local input puts. 

I am not sure what you mean by a set (or real) being “fundamentally incomputable”.

The Church thesis is on function defined on natural numbers.  On the reals, there are as much definition of computability than there are mathematicians. But to get the measure, it makes sense to study the subset of the real line structured by some degrees of complexity, and some results related this to the arithmetical and analytical hierarchy in recursion theory (and their has some importance for the extraction of physics (the self-referential measure on possible inputs) from “the head of the universal machine” ).




This is a measure of nonlinearity a quantum system is forced into, say with gravitation or with einselection into classicality. To compute it requires a single algorithmic system for computing all p-adic sets, where the theorem by Matyiaesovich is a form of Gödel's theorem that illustrates this does not exist.

What is the relation between p-adic sets and Matiyasevitch theorem? (I see the relation with Gödel, but only with dioplnatine polynomial (with integers). I don’t see the relation between non linearity and non computability. You need to elaborate perhaps, or give a link to some draft of your paper.



It corresponds to the unobservability of hidden variables, or that they are nonlocal, and establishes entanglement symmetries as topological indices or obstructions. This might mean we are saved by the Bell, here Bell's theorem in a sense, from the invasion of the robots.

You need to clarify this.



It will be some time I think before AI systems can work through self-referential inference. 

That is weird. Have you read my paper on "amoeba, planaria and dreaming machine”. Self-reference is just not avoidable. Once a universal machine can believe in the induction, like PA, ZF, any of their consistent extension, they are already as self-referential than you and me. They obey to the theology G* (with the same physics as us (Z1*, X1* and S4Grz1).

Bruno



--
You received this message because you are subscribed to the Google Groups "Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email to everything-li...@googlegroups.com.

Bruno Marchal

unread,
May 31, 2020, 3:46:09 AM5/31/20
to everyth...@googlegroups.com
On 30 May 2020, at 21:51, ronaldheld <ronal...@gmail.com> wrote:

Is the AI discovering some Physics or just fitting data which produces equations that look like physical laws?

The question is if the equation obtained makes the good predictions, on any, or at least “many” different data. 

It is easy to make a machine predicting an earthquake the day before an earthquake, by making it predicting everyday that there will be an earthquake …

My point is that such AI research avoid the metaphysical problem: even us cannot use physics to predict an eclipse, in a way coherent with mechanism. Physical laws works only if they solve the measure problem, which is typically not the case today, except QM is promising, as we can see from the study of self-reference. The problem is in making negligible the measure of the aberrant histories (which are executed in arithmetic).

Bruno




     Ronald
On Thursday, May 28, 2020 at 12:20:49 PM UTC-4, Philip Thrift wrote:


https://www.facebook.com/461616050561921/posts/3107668729289960/


We just posted a new AI paper on how to automatically discover laws of physics from raw video with machine learning. For example, we feed in the video below of a rocket moving in a circles in a magnetic field, seen through a distorting lens, and our code automatically discovers the Lorentz Force Law. It took Silviu and me about a year to get this working, by using ideas inspired by general relativity and the the theory of knots in 5-dimensional space, so we're excited to be done!  https://arxiv.org/abs/2005.11212

@philipthrift




--
You received this message because you are subscribed to the Google Groups "Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email to everything-li...@googlegroups.com.

Lawrence Crowell

unread,
Jun 1, 2020, 12:46:14 PM6/1/20
to Everything List
On Sunday, May 31, 2020 at 2:37:22 AM UTC-5, Bruno Marchal wrote:

On 30 May 2020, at 19:24, Lawrence Crowell <goldenfield...@gmail.com> wrote:

I wrote a paper recently for publication on how the unital set of QM is a Cantor/fractal set that is fundamentally incomputable.

I use the Cantor (triadic) set, or the Baire space. But it is more the measure on the possible local input puts. 

I am not sure what you mean by a set (or real) being “fundamentally incomputable”.

In the case of a Cantor set the "dust" can be reached (mapped by a function etc from one point to another) by a p-adic representations, but different elements have different p-adic groups. The Matiyaesivich results shows there does not exist a single method for all p-adic groups, Here the fractal is identified with the p-adic set. Actually this is not so much a fractal, but rather the complement of a fractal. 

I am really more interested in the impact with standard math and what might be called practical results. Issues with the ontology of mathematics I am less concerned about.

LC
 
To unsubscribe from this group and stop receiving emails from it, send an email to everyth...@googlegroups.com.

Bruno Marchal

unread,
Jun 2, 2020, 6:41:46 AM6/2/20
to everyth...@googlegroups.com
On 1 Jun 2020, at 18:46, Lawrence Crowell <goldenfield...@gmail.com> wrote:

On Sunday, May 31, 2020 at 2:37:22 AM UTC-5, Bruno Marchal wrote:

On 30 May 2020, at 19:24, Lawrence Crowell <goldenfield...@gmail.com> wrote:

I wrote a paper recently for publication on how the unital set of QM is a Cantor/fractal set that is fundamentally incomputable.

I use the Cantor (triadic) set, or the Baire space. But it is more the measure on the possible local input puts. 

I am not sure what you mean by a set (or real) being “fundamentally incomputable”.

In the case of a Cantor set the "dust" can be reached (mapped by a function etc from one point to another) by a p-adic representations, but different elements have different p-adic groups. The Matiyaesivich results shows there does not exist a single method for all p-adic groups, Here the fractal is identified with the p-adic set. Actually this is not so much a fractal, but rather the complement of a fractal. 

I don’t see the relation between p-adic number and Matiyasevic’s result (which I have studied in all details, notably in Matiyaevic book, which does not mention p-adic numbers). I would appreciate some references.



I am really more interested in the impact with standard math and what might be called practical results. Issues with the ontology of mathematics I am less concerned about.

I am interested in the mind-body problem, and my point is that with mechanism, the physical laws are at a very well precise place in the “head of the universal numbers”, so that we can test mechanism experimentally, and as it predicted quantum mechanics/logics, it makes sense to say that the evidence is for mechanism (with 0 universes, but many-histories/dreams).

Bruno

We are not human beings having spiritual experiences, we are spiritual beings having human experiences (de Chardin).




To unsubscribe from this group and stop receiving emails from it, send an email to everything-li...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/everything-list/345b9380-dee6-4052-af94-9eb37433bb1f%40googlegroups.com.

Reply all
Reply to author
Forward
0 new messages