Transfuturist

Karma: 341

Transfuturist 5 Mar 2013 23:58 UTC
0 points
in reply to: BlazeOrangeDeer’s comment on: Open thread, February 15-28, 2013
Chaotic classical simulations? Could you elaborate?

Transfuturist 6 Mar 2013 0:50 UTC
0 points
in reply to: BlazeOrangeDeer’s comment on: Open thread, February 15-28, 2013
I think there seems to be a mismatch of terms involved. Ontological probability, or propensity, and epistemological probability, or uncertainty, are being confused. Reading over this discussion, I have seen claims that something called “chaotic randomness” is at work, where uncertainty results from chaotic systems because the results are so sensitive to initial conditions, but that’s not ontological probability at all.

The claim of the paper is that all actual randomness, and thus ontological probability, is a result of quantum decoherence and recoherence in both chaotic and simple systems. Uncertainty is uninvolved, though uncertainty in chaotic systems appears to be random.

That said, I believe the hypothesis is correct simply because it is the simplest explanation for randomness I’ve seen.
What links here?
- Transfuturist's comment on Are coin flips quantum random to my conscious brain-parts? by Academian (6 Mar 2013 0:53 UTC; 0 points)

Transfuturist 6 Mar 2013 0:53 UTC
0 points
in reply to: OrphanWilde’s comment on: Are coin flips quantum random to my conscious brain-parts?
I’ve addressed this here.

Transfuturist 6 Mar 2013 1:15 UTC
0 points
in reply to: BlazeOrangeDeer’s comment on: Open thread, February 15-28, 2013
I don’t think they refer to Bayesian probability as probability. The abstract is ill-defined (according to LessWrong’s operational definitions), but their point about ontological probabilities originating in quantum mechanics remains. It, I think, remains intertwined with multiverse theories, as multiverse theories seem to explain probability in a very similar sense, but not in as many words or with such great claims.

Also, in a classical simulation, I would not see probability working as it should to be obvious at all. In fact, it’s quite difficult to imagine an actually classical system that also contains randomness. It could be that the childhood explanations of physical systems in classical terms while seeing randomness as present is clouding the issue.

Whichever way. I don’t think it’s really worth much argument. Just as a basis in probability theory.

Transfuturist 8 Mar 2013 19:28 UTC
−2 points
on: Destructive mathematics
The very big problem with this proposition is that as soon as you take all propositions to be true, you have no method for determining the falsity of any propositions. From any contradictorily accepted propositions any statement can be determined true.

To make this work, you would have to start with a small set of unprovable axiomatic negations, and then build from there. In other words, there is no essential difference between “destructive mathematics” and constructive mathematics, and destructive mathematics has to take a useless rigamarole around the concept of truth and falsity and rebuild constructive mathematics. All statements cannot be assumed to be true or false, they are indeterminate. Formal logic takes axioms and derives what is determinably true and false from those parameters.

I’m sorry, but this concept is useless.

Transfuturist 8 Mar 2013 22:39 UTC
0 points
in reply to: CronoDAS’s comment on: Destructive mathematics
I still don’t see a point in assuming every statement to be true. It seems more like a gimmick than anything else. Even without the principle of explosion, there must be a distinction between what is proved to be not false and what isn’t. What use is there in assuming everything to be true?

I see no point in this theory. The application to MWI doesn’t really make sense, and even if it did, that’s no reason to give this proposition any credence. The Tegmark hypothesis is also misunderstood; it states that all well-formed mathematical structures complex enough to have self-aware systems subjectively exist to those systems. I am not sure this can be proven, but I see even less of a connection to “destructive mathematics” than MWI.

How is this useful to logic?

Transfuturist 24 Jul 2013 2:06 UTC
0 points
in reply to: Rob Bensinger’s comment on: The idiot savant AI isn’t an idiot
I don’t understand the premise of 3, even when looking at the five theses. I saw it restated under the second lemma, but it doesn’t seem an enormously difficult problem unless the most feasible approach to self-modifying AI was genetic algorithms, or other methods that don’t have an area explicitly for values/goals. Is there anything I’m missing?

Transfuturist 30 Jul 2013 2:35 UTC
7 points
on: Open thread, July 29-August 4, 2013
I believe I’ve encountered a problem with either Solomonoff induction or my understanding of Solomonoff induction. I can’t post about it in Discussion, as I have less than 20 karma, and the stupid questions thread is very full (I’m not even sure if it would belong there).

I’ve read about SI repeatedly over the last year or so, and I think I have a fairly good understanding of it. Good enough to at least follow along with informal reasoning about it, at least. Recently I was reading Rathmanner and Hutter’s paper, and Legg’s paper, due to renewed interest in AIXI as the theoretical “best intelligence,” and the Arcade Learning Environment used to test the computable Monte Carlo AIXI approximation. Then this problem came to me.

Solomonoff Induction uses the size of the description of the smallest Turing machine to output a given bitstring. I saw this as a problem. Say AIXI was reasoning about a fair coin. It would guess before each flip whether it would come up heads or tails. Because Turing machines are deterministic, AIXI cannot make hypotheses involving randomness. To model the fair coin, AIXI would come up with increasingly convoluted Turing machines, attempting to compress a bitstring that approaches Kolmogorov randomness as its length approaches infinity. Meanwhile, AIXI would be punished and rewarded randomly. This is not a satisfactory conclusion for a theoretical “best intelligence.” So is the italicized statement a valid issue? An AI that can’t delay reasoning about a problem by at least labeling it “sufficiently random, solve later” doesn’t seem like a good AI, particularly in the real world where chance plays a significant part.

Naturally, Eliezer has already thought of this, and wrote about it in Occam’s Razor:

The formalism of Solomonoff Induction measures the “complexity of a description” by the length of the shortest computer program which produces that description as an output. To talk about the “shortest computer program” that does something, you need to specify a space of computer programs, which requires a language and interpreter. Solomonoff Induction uses Turing machines, or rather, bitstrings that specify Turing machines. What if you don’t like Turing machines? Then there’s only a constant complexity penalty to design your own Universal Turing Machine that interprets whatever code you give it in whatever programming language you like. Different inductive formalisms are penalized by a worst-case constant factor relative to each other, corresponding to the size of a universal interpreter for that formalism.

In the better (IMHO) versions of Solomonoff Induction, the computer program does not produce a deterministic prediction, but assigns probabilities to strings. For example, we could write a program to explain a fair coin by writing a program that assigns equal probabilities to all 2^N strings of length N. This is Solomonoff Induction’s approach to fitting the observed data. The higher the probability a program assigns to the observed data, the better that program fits the data. And probabilities must sum to 1, so for a program to better “fit” one possibility, it must steal probability mass from some other possibility which will then “fit” much more poorly. There is no superfair coin that assigns 100% probability to heads and 100% probability to tails.

Does this warrant further discussion, if at least to validate or refute this claim? I don’t think Eliezer’s proposal for a version of SI that assigns probabilities to strings is strong enough, it doesn’t describe what form the hypotheses would take. Would hypotheses in this new description be universal nondeterministic Turing machines, with the aforementioned probability distribution summed over the nondeterministic outputs?

Transfuturist 30 Jul 2013 4:21 UTC
2 points
in reply to: Qiaochu_Yuan’s comment on: Open thread, July 29-August 4, 2013
Can probabilistic Turing machines be considered a generalization of deterministic Turing machines, so that DTMs can be described in terms of PTMs?

Editing in reply to your edit: I thought Solomonoff Induction was made for a purpose. Quoting from Legg’s paper:

Solomonoff’s induction method is an attempt to design a general all purpose inductive inference system. Ideally such a system would be able to accurately learn any meaningful hypothesis from a bare minimum of appropriately format- ted information.

I’m just pointing out what I see as a limitation in the domain of problems classical Solomonoff Induction can successfully model.

Transfuturist 30 Jul 2013 4:54 UTC
−2 points
in reply to: Adele_L’s comment on: Open thread, July 29-August 4, 2013
To come up with MWI, it would have to conceive of different potentialities and then a probabilistic selection. I don’t know, I’m not seeing how deterministic Turing machines could model that.

Transfuturist 30 Jul 2013 19:01 UTC
−1 points
in reply to: Kawoomba’s comment on: Open thread, July 29-August 4, 2013
Trivially? I was under the impression that it involved up to a polynomial slowdown, while probabilistic Turing machines can simulate deterministic Turing machines by merely having only a single probability of 1 for each component of its transition function.

Transfuturist 30 Jul 2013 19:36 UTC
0 points
in reply to: Kawoomba’s comment on: Open thread, July 29-August 4, 2013
Well, wouldn’t that be because it’s all theorizing about computational complexity?

I see the point. Pseudorandom number generators would be what you mean by simulation of nondeterminism in a DTM? Would a deterministic UTM with an RNG be sufficient for AIXI to hypothesize randomness? I still don’t see how SI would be able to hypothesize Turing machines that produce bitstrings that are probabilistically similar to the bitstring it is “supposed” to replicate.

Transfuturist 1 Aug 2013 6:02 UTC
5 points
on: More “Stupid” Questions
I am a transhumanist and a futurist, but I’ve been depressive recently while thinking about the far future. This rarely happens. I found myself being scared of getting smarter due to a Singularity-like event. I was also scared by the arbitrariness of our goals and values. Simply put, I don’t fit in to the present. I’m theorizing about intelligence, reading scientific papers, and participating very modestly in the brony fandom. I’ve made it my life’s goal to make major steps towards safe AGI. Living to the point past that, I see aimlessness. Besides my one creative technological skill, I am mainly a consumer. That leads to my concern of getting smarter.

I mostly read stories, take in stories, participate in stories. Stories are my life. I want to be able to appreciate the stories we have now in the future. And I’m concerned that upgrading to transfuturist levels of intelligence will make the types of stories we have now incredibly banal and obvious, for many good reasons. Predictable, boring, and worthless.

It’s not really a question, but I’d appreciate any other perspectives. Please?

Transfuturist 11 Aug 2013 5:51 UTC
0 points
in reply to: JGWeissman’s comment on: The Irrationality Game
Link?

Transfuturist 13 Aug 2013 15:54 UTC
2 points
in reply to: new_throwaway’s comment on: New Monthly Thread: Bragging
Please elaborate. Algorithmic trading?

Transfuturist 27 Aug 2013 2:54 UTC
2 points
in reply to: jamesf’s comment on: Open thread, August 26 - September 1, 2013
This sounds exceedingly interesting. Please clarify on the “few days on ice,” does that mean they were cryogenically frozen, or...

Transfuturist 31 Aug 2013 1:52 UTC
3 points
in reply to: David_Gerard’s comment on: Open thread, August 19-25, 2013
The biggest problem I have with racists claiming racial realism is this.

Transfuturist 3 Sep 2013 21:00 UTC
−3 points
on: The genie knows, but doesn’t care
What if the AI’s utility function is to find the right utility function, being guided along the way? Its goals could be such as learning to understand us, obey us, and predict what we might want/like/approve, moving its object-level goals to what would satisfy humanity? In other words, a probabilistic utility function with great amounts of uncertainty, and great amounts of apprehension to change, or stability.

Regardless of the above questions/statement, I think much of the complexity of human utility comes from complexities of belief.

If we offload complexity of the AI’s utility function into very uncertainly defined concepts, and give it an apprehension to do anything but observe given such little data… I don’t know, though. This has been something I’ve been sitting on for a while, lambast me.

As one last thing, I think the best kind of FAI would be a singleton, with a metautility function, or society’s utility function. I think one part of Friendliness would be determining a utility function for society, as to how people can interfere with each other in what circumstances, and then build the genie’s utility function in the singleton’s constraints.

Please critique. If my ideas are as unclear as I think they may be (I’m sick), please mention it.

Transfuturist 3 Sep 2013 21:12 UTC
0 points
in reply to: Michelle_Z’s comment on: True Optimisation
We’re going to be facing a big underpopulation problem in the next century or so.

Transfuturist 3 Sep 2013 21:52 UTC
3 points
in reply to: Said Achmiz’s comment on: Rationality Quotes September 2013
The effect of the fallacy can be implied, can’t it?