Posts

One systemic failure in particular 2020-05-28T05:25:58.335Z
Matrix Multiplication 2020-03-05T12:12:15.125Z
What types of compute/processing could we distinguish? 2020-01-18T10:04:03.380Z
Multiple conditions must be met to gain causal effect 2019-12-05T10:15:05.640Z
This sometimes helps to expose assumptions 2019-08-28T18:42:57.271Z
Limits of and to (artificial) Intelligence 2019-08-25T22:16:33.108Z

Comments

Comment by moritzg on What types of compute/processing could we distinguish? · 2020-06-09T13:48:51.292Z · LW · GW

This relates to: https://en.wikipedia.org/wiki/Computational_irreducibility https://en.wikipedia.org/wiki/Reversible_computing https://en.wikipedia.org/wiki/Computational_problem

Comment by MoritzG on [deleted post] 2020-06-01T18:20:19.918Z

"freedom to self-alter its own error function"

How? By changing the function alone or by changing the input to that function?

Comment by moritzg on One systemic failure in particular · 2020-05-29T16:31:54.355Z · LW · GW

"Styling" I will (and can) make the edits.

"parody" I call it a polemic analogy.

"seems, if not in conflict with" I think you noticed that there is no contradiction, but I agree that I need to clarify. Faced with a massive lack of information and the task to predict the future it is clear that it would be pure luck to make the best decision. Operating with that mindset might even be hindering.

" I must seek C*(A+B) at a lower cost." I was trying to get into what to choose / look for in a finite set with competition. A B C ... are terms of criteria that I estimate to be fulfilled to some degree. For simplicity they shall be binary logic terms. Every option that I have has more properties than I even know about and those I do know and find relevant, I either seek or avoid. Any term might contain many such properties. Knowing what others are looking and paying for and that the world is very complicated, I find it more sensible to intentionally not use the same function to assess options. Instead I must design my net to "fish" in other areas of the choice property space. This applies to HR or any other investment.

Comment by moritzg on One systemic failure in particular · 2020-05-29T16:28:35.024Z · LW · GW

Thank you for commenting.

"is offputting enough" That would be a sensibility of yours and not a rational argument.

"implication that young women are not competent, and the generalization overall, and the unstated implication that HR has anywhere near the power that you ascribe to it" I made no such statements. "Many" is not the same as "all". I include employees of headhunting companies as HR workers and these do have power when it comes to early screening including the assessment of qualification. I had plenty such talks where I could not even make the other understand what I do. Also I do use the term HR in a broader sense meaning the entire system only including those who work in the HR department, but not restricted to them.

Comment by moritzg on One systemic failure in particular · 2020-05-28T10:01:11.499Z · LW · GW

I lost the formatting when I pasted the text. I managed to switch to the Markdown interpretation and bring the list back.

Comment by MoritzG on [deleted post] 2020-05-25T06:45:36.113Z

I came across this:

The New Dawn of AI: Federated Learning

" This [edge] update is then averaged with other user updates to improve the shared model."

I do not know how that is meant but when I hear the word "average" my alarms always sound.

Instead of a shared NN each device should get multiple slightly different NNs/weights and report back which set was worst/unfit and which best/fittest.
Each set/model is a hypothesis and the test in the world is a evolutionary/democratic falsification.
Those mutants who fail to satisfy the most customers are dropped.

Comment by moritzg on Autism And Intelligence: Much More Than You Wanted To Know · 2020-05-22T08:58:35.774Z · LW · GW

When it comes to intelligence, rationality, depression, autism the evolutionary selection aspect is interesting, because we all know that the mentioned mental properties are lowering your chances to raise many children today.

https://www.google.com/search?q=%22falling+off+the+cliff%22+evolution+OR+selection+autism+OR+depression

https://www.psypost.org/2017/02/study-suggests-autism-risk-genes-favored-natural-selection-47876

https://evolution-institute.org/the-darwinian-causes-of-mental-illness/

Too much good quickly turns bad.

Comment by MoritzG on [deleted post] 2020-05-18T11:01:17.415Z

As we know and you mentioned, humans do learn from small data. We start with priors that are hopefully not too strong and go through the known processes of scientific discovery. NN do not have that meta process or any introspection (yet).

"You cannot solve this problem by minimizing your error over historical data. Insofar as big data minimizes an algorithm's error over historical results ... Big data compensates for weak priors by minimizing an algorithm's error over historical results. Insofar as this is true, big data cannot reason about small data."

NN also do not reduce/idealize/simplify, explicitly generalize and then run the results as hypothesis forks. Or use priors to run checks (BS rejection / specificity). We do.

Maybe there will be a evolutionary process where huge NNs are reduced to do inference at "the edge" that turns into human like learning after feedback from "the edge" is used to select and refine the best nets.

Comment by moritzg on I do not like math · 2020-05-02T05:56:39.181Z · LW · GW

Why did you not go for engineering (like me)? Still some math proves but no one listens and they will not test it either.

Twice I made the mistake to ask 'why' it is the way it is. All I got was "look at the prove, it works out". That is why have have little respect for mathematicians i.g..

Comment by moritzg on Matrix Multiplication · 2020-04-23T08:14:59.951Z · LW · GW

Because the SIMD approach is bad for 2D on 2D matrix multiplication NVIDIA has introduced:

Tensor Cores in the Volta architecture.

Article about it:

https://www.anandtech.com/show/12673/titan-v-deep-learning-deep-dive/3

Comment by moritzg on The Great Annealing · 2020-04-15T18:00:57.644Z · LW · GW

Or when you attended a gathering/event then read about it and think that that was an entirely different event.

"newspaper from ten years ago, ... what happened in the topic afterwards and then judge how informative the article was"

As a German you will know the saying: "Nichts ist so alt wie die Zeitung von gestern." -> "Nothing is as old as yesterdays paper."

N.N. Taleb calls it noise.

At fist it is either wrong or without consequence or propaganda, then it is outdated. A historian will find 99% of all "news" to be little more as an "interesting time piece" at best representative for the thinking and style of the era.

Comment by moritzg on The Great Annealing · 2020-03-31T17:04:00.338Z · LW · GW

Danke

Or you read: https://en.wikipedia.org/wiki/Propaganda_model

Comment by moritzg on How dangerous is it to test a vaccine without animal trials? · 2020-03-20T13:46:50.365Z · LW · GW

There is a non zero risk of death. There have been cases where some candidates had severe permanent damage and it took a while to figure out why. We are all slightly different.

Even with animal trials there is a risk. I do not think there is an answer to your question.

Comment by moritzg on Growth rate of COVID-19 outbreaks · 2020-03-12T08:21:05.220Z · LW · GW

Oh, I should have mentioned that this is also assuming that there is a constant factor between the total number of infected in the past to the number of currently infectious. Which is true as long as the spread is exponential, but that is the entire assumption anyhow.

Comment by moritzg on Growth rate of COVID-19 outbreaks · 2020-03-10T10:15:00.107Z · LW · GW

In Germany the data is ATM consistent with:

+22% infected per day which is exactly +3 people infected after one week after the infection by every infected. (This is assuming that there no imported cases, 7th root of ((1+3)/1))

Comment by moritzg on Epistemic standards for “Why did it take so long to invent X?” · 2020-03-09T10:59:04.373Z · LW · GW

"only a minority of people use a bicycle for anything other than recreation"

I guess my upbringing and surrounding is special (densely populated area in north Europe), but I know plenty of people who move in no other way (shopping, vacation, commute, everything). Before the gasoline motor scooter became wide spread, and poisoned the air in Asia, people used bikes all the time.

Comment by moritzg on Epistemic standards for “Why did it take so long to invent X?” · 2020-03-08T18:52:20.598Z · LW · GW

Please elaborate on why you think the bicycle "merely offer convenience or entertainment". I understand that people did not understand it's potential and thought of it as a toy for crazy people, but wasn't the same true for the gasoline-automobile? To me the bicycle is of great importance, not just/only for leisure, sport. I understand that the bicycle's value depends on the distance, flatness, wind, road quality traveled, but compared to a horse (that most did not have) it is so much better.

Comment by moritzg on Matrix Multiplication · 2020-03-05T19:56:27.967Z · LW · GW

The entire SIMD vector approach is good for many dot products but it is not the same as a systolic array for rank two on rank two multiplication.

If the job would be to multiply two 1024x1024 matrices then a systolic array of 256x256 MACs would be a good choice. It would work four times on 256x1024 by 1024x256 matrices for 1024+256 steps.

Comment by moritzg on Matrix Multiplication · 2020-03-05T19:38:54.185Z · LW · GW

To me there is a difference between the hardware for 1xN by Nx1 and MxN by NxM (with N > M > 1). Although any matrix operation is many 1xN by Nx1 dot products, doing them independently would be inefficient.
"If you do a matrix multiplication the obvious way, this results in dot products of rows and columns (one for each element of the resulting matrix). So it seems to me that improving matrix to matrix multiplication performance comes from improving the performance of dot products."
True, but not individual dot products, but the collective of very many dot products. Obviously you do not do it the obvious way as you would have to load the same data over and over again.

Comment by moritzg on Matrix Multiplication · 2020-03-05T19:35:44.344Z · LW · GW

An example of a systolic algorithm might be designed for matrix multiplication. One matrix is fed in a row at a time from the top of the array and is passed down the array, the other matrix is fed in a column at a time from the left hand side of the array and passes from left to right. Dummy values are then passed in until each processor has seen one whole row and one whole column. At this point, the result of the multiplication is stored in the array and can now be output a row or a column at a time, flowing down or across the array.
https://en.wikipedia.org/wiki/Systolic_array

Comment by moritzg on How does electricity work literally? · 2020-03-05T10:22:14.330Z · LW · GW

True, I had not claimed that all criteria could or have been met. Because of the noise and the heat I just the other day replaced the inductive load in some of my very old but still fully functioning kitchen counter lights, with modern switching current regulators. The 50 Hz produce a 100 Hz tone that had been bothering me for decades. But even some of those can be heard by some people. (Not me I am deaf to anything >10kHz)

It is a compromise in an area of sensory overlap but the human senses are not equally sensitive to all frequencies. Your hearing is way better at 3kHz. At your age you will still remember CRT monitors that would operate at 60 Hz at max resolution, bad but they did get used.

Comment by moritzg on How does electricity work literally? · 2020-02-28T16:54:21.632Z · LW · GW

Why 50/60Hz? It has to be too low to be heard, to high to be seen, high enough for transformation, low enough for low induction losses, low enough for simple rotating machines. Trains can not use 50/60 so they went with 1/3 (16+2/3 Hz or 20 Hz)
Grid frequency is controlled to +-150mHz if that fails private customers might get disconnected/dropped.
The time derivative of the grid frequency is a measure of the relative power mismatch.

Comment by moritzg on Have epistemic conditions always been this bad? · 2020-01-28T00:04:59.792Z · LW · GW

I observe a radicalization that is driven by what I call the concept of "counter crazy". It let to Trump but I have been aware of it for longer on the left. The idea is that by being more radical in the way, that you think the world needs to be, you could achieve that. It is compounded by the tribalism and identity culture.

The idea of "you can not speak on this because you are not a woman / ..." is recent to me. But has been expressed by the most intelligent female I know. It is a scary idea.

The idea of cushioning life is a generational thing and related to U.S. product liability law. No matter how dumb, misinterpreting, interpreting in bad faith or sensitive you are, others are responsible for your feelings no matter how unjustified. The Snowflake generation can not surprise anyone, we were watching as they were raised to be what they have become.

There is much more awareness of the issue thanks to the members of the "intellectual dark web" and comedians such as Ricky Gervais, Bill Maher, Joe Rogan.

The PC culture is nothing new. It has been impossible to talk about many topics for four decades. And obviously there are "good" reasons. People are unable to talk about these issues without confusing separate issues, values, preferences. To me it is painful to listen to the arguments because they are so confused and old. Most (including scholars and journalists) are simply unable and should indeed not talk on these issues because it does lead nowhere. What we have now is a result of not having talked about important topics for decades, massive preference falsification/concealment and having generations growing up in that environment.

Comment by moritzg on Limits of and to (artificial) Intelligence · 2020-01-27T09:19:13.843Z · LW · GW

I found this recent Dilbert cartoon to be a good summery of the issue with being smart in a complex random world:

https://dilbert.com/strip/2020-01-25

Comment by moritzg on Summary of "The Straw Vulcan" · 2020-01-23T17:21:06.319Z · LW · GW

The way you commented it is not clear what you are referring to. I did not understand your comment because I did not get "where you were coming from".

Comment by moritzg on Summary of "The Straw Vulcan" · 2020-01-23T12:16:51.783Z · LW · GW

Straw_Vulcan is an example of an attack of two of the three types of thinkers on another.

The moral-thinkers try to show their superiority. In Star Trek this is ever present. In all the stories morality and principles always win over rational compromise. The captains usually favor the best possible short term outcome over risk minimization and the long term. As it is fiction this always works out.

The three thinking types as formalized/categorized (to my knowledge) by Rao Venkatesh of ribbonfarm.

https://fs.blog/venkatesh-rao/

Venkatesh Rao: The Three Types of Decision Makers [The Knowledge Project Ep. #7]

I can hardly express how useful I found this to make sense of the world.

Comment by moritzg on What types of compute/processing could we distinguish? · 2020-01-22T09:34:18.430Z · LW · GW

Put that way I completely agree.

Comment by moritzg on What types of compute/processing could we distinguish? · 2020-01-21T09:59:24.734Z · LW · GW

Thank you, I should have thought of it in that (Time complexity) context. Time complexity is not just about how long it takes but also about the nature of the problem. Chess is neither P nor NP, but the question of complexity is certainly related.

Maybe my question is: Why can there be a Heuristic that does fairly well and is nowhere near exponential? Even a count of the pieces left on the board usually says something that only a full search can prove.

Comment by moritzg on What types of compute/processing could we distinguish? · 2020-01-20T12:29:31.946Z · LW · GW

Then you are wrong because since the search usually does not reach the chess mate state, there is always a scoring heuristic replacing the further exploration search at some dept.

I know and had read chessprogramming prior to your post, you are wrong to assume that I am a total idiot just because I got myself confused.

Comment by moritzg on What types of compute/processing could we distinguish? · 2020-01-19T20:24:37.360Z · LW · GW

Ok, let's go with chess. For that game there is an optimal balance between the tree search and the evaluation function. The search is exploratory. The evaluation is a score.

The evaluation can obviously predict the search to some degree. Humans are very bad at searching, still some can win against computers.

The search is decompressing the information to something more easily evaluated by a computer. A human can do it with much less expansion. Just a matter of Hardware or is it because the information was there all along and just needed a "smarter" analysis?

Comment by moritzg on Key Decision Analysis - a fundamental rationality technique · 2020-01-14T14:29:32.662Z · LW · GW

This reminds me of another issue. If you do make informed complicated decisions, the basis of these decisions might change over time. I struggle with that problem professionally. As an engineer I have to make complicated compromises/decisions. The trouble is that the situation changes all the time. The requirements and the means change. Without tracking why I made decisions there is no way to tell if those decisions still hold, because I do not even remember myself. The project becomes a zombie even before there are true legacy and hand-over issues. Usually decisions are incomprehensible later. We all know this and have though everyone else is an idiot, but often people had good reason to do it that way or lost track as described. Making changes to often reveals that there were reasons, but too late.

Privately you might find yourself in a place that you had reason to go into but those reasons went away without you noticing.

Comment by moritzg on Markets are Anti-Inductive · 2019-12-18T10:56:51.303Z · LW · GW

In reality there are smart penguins and dumb penguins and penguin news papers. The professional penguins will tell other penguins how great it has been going so they can get out before the ledge breaks of and they all fall into the water.

To realize those booked earnings you have to sell without causing the crash, so you have to setup potential buyers first. That is why I consider articles about investing into something in major papers the last warning before the crash. When I read that the only smart thing to do, is to invest into ... I know not too.

Comment by moritzg on Is the "business cycle" an actual economic principle? · 2019-12-13T19:31:30.867Z · LW · GW

You seem to think that the economy and markets are random without memory or state. You are the one with a fallacy called: "the map is not the territory".

Comment by moritzg on Is the "business cycle" an actual economic principle? · 2019-12-13T19:24:14.545Z · LW · GW

I think Liron only meant the times of growth with those 10%. Looking at the recent stock market you will clearly find growth that is much higher than the long term rate and higher than economy + inflation + "risk free return". In the last 10 years the annual rate was indeed 10.5% pa

Comment by moritzg on Causal Diagrams and Causal Models · 2019-12-12T18:53:10.301Z · LW · GW

There are two issues with it.

You can not figure out how something works by only looking at some aspect. Think of the blind people and elephant story.

But it still has a point because with a subsystem that makes predictions the understanding of a system by pure observation becomes impossible.

Comment by moritzg on Multiple conditions must be met to gain causal effect · 2019-12-09T19:16:22.478Z · LW · GW

Right, one could expand the clause indefinitely, that is kind of what I meant by "can only find what you are looking for". But that only means it is hard, not that it is bad to think that way.

I do neither think of it as logic nor as causal diagrams nor Bayesian nor Markov diagrams but simply as sets of some member type that may have any number of features/properties/attributes that make them a member of some subset.

When I wrote "A AND B" I wanted you to understand it as a dual logic clause, but only for simplicity.

The way I really think about it is: attribute magnitude to impact function and then some form of interaction function that is neither only AND nor OR but possibly both to some degree. We have to deal with negative correlation in some way, I do not see how that is possible if it is always OR.

right "language" in which to think [is] causal diagrams

They are nice on paper but I can not see how they are useful. To me they seem like some synthetic made up way to get the result, unfit to model the world. "If the world would not be as it is, it would be mathematically correct to do this." is so academic. As far as I understand it, the graph can not be cyclic. Since you do not know if the graph is cyclic and what factors are in the cycle you do not know which factors you must treat as an aggregate. The only directions known are those that go into the graph.

There is only one joint probability for cases where there were multiple causal paths to one feature/property.

Think of a hospital. Sick people go to hospitals, but sometimes people in a hospital will catch an infection that is only typical in hospitals.

A= person is sick
B= person is in hospital
C= person has hospital infection
C is a subset of A
A causes B
B causes C

How do you work with that?

"the fix is not to look for a giant and-clause of conditions [but] to build a gears-level model of the system, figure out the whole internal cause-and-effect graph"

I thought that was what I was suggesting. Instead of stopping at: "It has to do with gears." keep going to get more specific, find subsets of things with gears: "gear AND oval-shape AND a sprocket is missing AND there is a cardan shaft AND ..." But if indeed only things with gears are affected do not expand with "gears AND needs oil" because that already follows from gears.

Comment by moritzg on How good is a human's gut judgement at guessing someone's IQ? · 2019-12-08T20:39:06.400Z · LW · GW

" Does anyone know of any similar experiments that have been run? "

I can say with certainty that it has been done on a small scale, because I once saw a German TV documentary for which they had also gotten a small group (6-10) of both males and females and gotten them tested and then shown them the pictures of the others. Later they also gave them some facts, showed videos and asked them again.

The outcome was that there was a clear correlation between the guesses and the IQ-Test results and it got better the more information people had.

But the people were not representative but of above average education. In my own experience it is not the case that people always have a good intuition. I would guess that people are bad as soon as the person judged is much more intelligent than the judge. I think very smart people are much better at recognizing each other.

Comment by moritzg on Multiple conditions must be met to gain causal effect · 2019-12-06T16:20:43.782Z · LW · GW

I made up this story:
In a company there have been head injuries, so they brought in a medical student to investigate/research.
The researcher gathered all employees blood pressure, gender, age, and eye sight data.
The result was that mostly men were affected, with all other factors being what you would expect given the employees.
The company was forced by the insurance company to make helmets mandatory for all men due to their gender being a risk factor.
Because the engineers were all men they were over proportionally affected and did not like to wear the helmets, so they got together and demanded further research into what caused the injuries and how to remove the cause.
This time the secretary was tasked with the follow up because she knew Excel. She took her mail scale and measuring tape and went around asking everyone if they drank coffee or tea, measured the weight of the content of people's pockets and how high they were with and without shoes. To be thorough she did this for every week day separately.
After importing the previous data, she found many correlations between attributes and other attributes variances but what stood out were the correlations between injuries to Friday, pocket weight, gender, height with shoes in ascending order. A histogram of injuries per "height without shoes"-class showed a sharp increase at 6 feet. Being taller than 6' was clearly the cause.
After having presented her findings, one woman stood up and remarked: "But I am not 6', and it happened to me!" Counting the women taller than 6' the secretary found none.
---
I could go on but I think you get it and we can save us the time. After more searching they found that their 6 foot door frames were the best thing to change and that some women had been wearing higher shoes on Fridays.
My point is that gender was not the cause and especially "too low doors" AND ("over 6' tall" OR ("tall for a woman" AND "high shoes")) was the problem. Neither being a tall woman nor high shoes alone would have been causal in this scenario.
I would have loved to include wheel chairs in this but found it too complicated.

Comment by moritzg on If giving unsolicited feedback was a social norm, what feedback would you often give? · 2019-12-05T11:57:46.054Z · LW · GW

"This will result in unnecessary stress and misery in your life."

LOL, that is very close to what I told a girl once. You would think it is the most sensitive and reasonable thing to tell a person and a good way to put it. She did not call me names, but was not thankful either.

Comment by moritzg on Autism And Intelligence: Much More Than You Wanted To Know · 2019-11-23T19:05:59.237Z · LW · GW

"cases of autism that are caused entirely or mostly by normal genetics are associated with unusually low IQ (80% confidence) "

Only the research correlating genes and IQ-test results are objective.

All correlations between IQ and DIAGNOSED autism are skewed. People who are smart and have good enough speech skills, and thus are not too affected can hide their level of autism. People who are functional will not be diagnosed.

Lets assume, that autism is not an on/off deal but gradual and that there is a positive correlation with general intelligence, then the statistic will not include people who are below a high level of autism because they compensate.

Comment by moritzg on Autism And Intelligence: Much More Than You Wanted To Know · 2019-11-23T18:30:21.527Z · LW · GW

"autistic people ... generally have very low intelligence. One study ... autistic people had an IQ ..."

Unless you positively define intelligence as measured by some IQ-Test, I oppose that statement.

The entire discussion around intelligence would profit, if people would stop casually equating the two.

One is a test that have seen different ones of and some where out right bad others flawed, the other is a concept that can be described, but is much more often used than understood by the public.

Comment by moritzg on Fake Explanations · 2019-11-20T11:13:20.413Z · LW · GW

Had the teacher presented a dozen of dice all showing the same number and asked how this could have happened they would have been wiser.
But the situation is similar. In pure theory this could happen naturally, in that case doubting it would be a case of gamblers fallacy or not knowing the Anthropic principle.

If you encounter the impossible you should check your assumptions, but to say that a human like entity has caused this outcome is dangerous.

Comment by moritzg on Fake Explanations · 2019-11-20T10:28:09.288Z · LW · GW

When you are presented with a very unlikely outcome you have to accept it.

Had the teacher shown a dozen dice all showing the same number and asked how he did it, there would have been two answers:

2. You

Comment by moritzg on Stuff That Makes Stuff Happen · 2019-09-09T17:18:44.711Z · LW · GW

"universe is a connected fabric of causes and effects."

I do not think that the universe as a whole is one fabric of causes and effects. There are isolating layers of randomness and chaos upon which there are new layers of emergence. This is why we can model at all without having one unified model.

"Every causally separated group of events would essentially be its own reality."

Places outside our solar system are their own realities in that sense. We have no effect there. Only maybe someone is there to amplify our radio signals.

Comment by moritzg on Limits of and to (artificial) Intelligence · 2019-09-05T20:23:44.431Z · LW · GW

I found these two articles on AI's mental health:

"Can Artificial Intelligences Suffer from Mental Illness? A Philosophical Matter to Consider"

Hutan Ashrafian

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5364237/

"Does my algorithm have a mental-health problem?"

Thomas T Hills is professor of psychology at the University of Warwick in Coventry, UK.

https://aeon.co/ideas/made-in-our-own-image-why-algorithms-have-mental-health-problems

Comment by moritzg on Why so much variance in human intelligence? · 2019-08-26T11:37:41.324Z · LW · GW

I think intelligence is much like homosexuality ...

... in that, it mostly benefits the tribe/gene-pool, but not the individual.

Being of average intelligence you are more intelligent than a good portion of the population and that helps you, just as being sub-average might be a hindrance in some situations. But being that much more intelligent does not help that individual much.

One does not have to be intelligent to profit from the intelligence of others. "We flew to the moon." No, *we* did not. We did not find Antibiotics, but we have much more breeding success because of it.