Buzzword Convergence: Making Sense of Quantum Neural Blockchain AI | Hacker News Reader

Buzzword Convergence: Making Sense of Quantum Neural Blockchain AI

blog.stephenwolfram.com

75 pointsOberstKrueger38 comments

Since we're in a whimsical mood:

"There’s a function called Hash in the Wolfram Language, and for example applying it to the previous paragraph of text gives 8643827914633641131."

I was a bit saddened to see that this was not "applying it to this paragraph of text gives…", which would have been quite the party trick.

Would it be possible?

Probably. There are only 2^64 possible hash values (Mathematica's Hash function, with no method specified, produces a 32- or 64-bit output depending on platform), and there are many more strings than that which could reasonably be called a "paragraph". It's pretty reasonable to hope that there's a self-describing paragraph out there somewhere.

Hah, "could in theory" vs "could in practice" -- depending on the hash function, it could take hundreds or thousands of processor-years. Maybe Google could do it.

http://m.wolframalpha.com/input/?i=2%5E64+things+%2F+%281+bi...

Actually for any given paragraph, probabilty of hash(paragraph+hash) = hash is pretty high.

You mean that there exists at least one hash such that hashFunc(paragraph+hash) = hash? Yup, that probability converges to 1-exp(-1) = 63%. So it's no guarantee, but if you have some wiggle room in the paragraph (change the wording slightly etc) you can probably make it work.

It looks that the default Hash implementation depends on whether the system is 32-bit or 64-bit [1]. It's also not clearly documented what the default implementation of Hash really is.

On a 32-bit system, yes, someone could probably do it.

[1] https://mathematica.stackexchange.com/a/124340

Given that they control the hash function, it would be trivial: hard-code a special exception value for exactly that purpose.

What you are describing sounds similar to a "quine". A quine is a non-empty computer program which takes no input and produces a copy of its own source code as its only output.

https://en.wikipedia.org/wiki/Quine_(computing)

Here is an example: s = 's = %r\nprint(s%%s)' print(s%s)

It is very similar, except this one is computationally difficult to achieve if the hash function is worth its salt ;)

are there any practical uses of quine?

As a distraction it can make a nice tonic on a stressful day.

Yes: hiding malware.

https://softwareengineering.stackexchange.com/questions/1113...

More generally, https://stackoverflow.com/a/1764933/1462221 points out that DNA implements a very big and complicated quine.

I feel that a groundbreaking use of quines is simply yet to be realized. I have a feeling they will show up in AI (if not intentionally!).

Does DNA encode it's own logic? That seems testable.

One of the answers from your first link gives a link to this excellent article

https://link.springer.com/chapter/10.1007%2F978-3-540-92273-...

on "autocatalitic quines". The Introduction section explains very nice the history of uses of quines in artificial life.

There are some weird parts though in all this, namely that we may think about different life properties in terms of quines:

1) Metabolism, where you take one program, consume it and produce the same program

2) Replication, where you take one program, consume it and produce two copies.

But what about

3) Death

I thought about this a lot during my chemlambda alife project, where I have a notion of a quine which might be interesting, seen the turn of these comments.

A chemlambda molecule is a particular trivalent graph (imagine a set of real molecules, the graphs don't have to be connected), chemical reactions are rewrites, like in reality, when if there is a certain pattern detected (by an enzyme, say) then the patern is rewritten.

There are two extremes in the class of possible algorithms. One extreme is the deterministic one, where rewrites are done whenever possible, in the order of preference from a list, so that the possible conflicting patterns are always solved in the same way. The other extreme is the purely random one, where patterns are randomly detected and then executed or not acording to a coin toss.

Now, a quine in this world is by definition a graph which has a periodic evolution under the deterministic algorithm.

The interesting thing is that a quine, under the random algorithm, has some nice properties, among them that it has a metabolism, can self-replicate and it can also die.

Here is how a quine dies. Simple situation. Take a chemlambda quine of period 1. Suppose that there are two types of rewrites, the (+) one which turns a pattern of 2 nodes into a pattern of 4 nodes, the other (-) which turns a pattern of 2 nodes into a pattern of 0 nodes (by gluing the 4 remaining dangling links in the graph).

Then each (+) rewrite gives you 4 possible new patterns (one/node) and each (-) rewrite gives you 2 possible new patterns (because you glued two links). Mind that you may get 0 new patterns after a (+) or (-) rewrite, but if you think that a node has an equal chance to be in a (+) pattern or in a (-) pattern, then there is twice as possible that a new pattern comes from a (+) rewrite than from a (-) rewrite.

Suppose that in the list of preferences you always put the (+) type in front of the (-) one. It looks that in this way graphs will tend to grow, right? No!

In a quine of period 1 the number of (+) patterns = number of (-) patterns.

Hence, if you use the random algorithm, the non execution of a (+) rewrite is twice more probable to affect future available rewrites than the non-execution of a (-) rewrite.

In experiments, I noticed lots of quines which die (there are no more rewrites available after a time), some which seem immortal, and no example of a quine which thrives.

On practical computers implementing such thing is trivial enough to be borderline uninteresting (at least when done by low-level non-portable means).

The interesting "practical" application is in proving that such a thing can exist in given formal system. By the way the concept of fixed-point combinators (of which Y-combinator is particular implementation) is essentually the same thing. (And in fact such combinators are notionally better match to problem "produce string that contains result of this function applied to it in its contents" than quines)

[Edit: functional->fixed-point and reworded the Y-combinator remark slightly]

Come back to me when I can have serverless quantum neural blockchain AI.

Came here to find an “April fools”, instead it’s a mostly serious article, albeit a tad tongue in cheek.

So it did fool you.

Surprised that nobody's called out SingularityNet [1] as a serious venture along these lines (sans "quantum").

"By combining open source principles, blockchain integration, and leading minds in machine learning we will make AI a global commons for all."

[1]: https://singularitynet.io/

I don't know, I get the feeling this project will have a lot of hype but nothing useful come out of it in the end. The team behind it doesn't seem like they have had significant practical successes, or exits.

I am also pretty prejudicial against anything that includes the term "singularity." That term sets of my bullshit detector.

There are so many examples these days. Here's an IBM press release I just stumbled across a few hours ago[1]:

>The two have been able to leverage JD’s expertise in the application of artificial intelligence (AI), blockchain, big data and other new technologies to protect consumers...Recent testing by Walmart showed that applying blockchain reduced the time it took to trace a package of mangoes from the farm to the store from days or weeks to two seconds.

[1] https://www-03.ibm.com/press/us/en/pressrelease/53487.wss

> And the act of that measurement would in effect force the blockchain to pick a definite history.

Wouldn't there be divergence, the further back in history you go between picks, due to collisions? Or is this what the reversible nature of the hashes take care of?

An awesome glimpse into the coming buzzword singularity!

So it really is recursive turtles after all.

Reply on news.ycombinator.com