Hacker News Reader

Comment by SeanLuke | Hacker News Reader

Cryptocurrency, welcome to AI's long nightmare.

AI has forever had buzzword-complaint terms. Consider "Agent", which describes a very specific thing (namely autonomy), but that didn't stop everyone and their dog outside the field from calling their software a "database agent" or a "compiler agent" or whatnot, equating "agent" with "program". Indeed there's an entire non-AI field called "Agent Based Modeling" -- of which I am a participant -- which completely misunderstands the term. So AI gave up and switched to "Intelligent Agent", which was unfortunately an even more buzzword-compliant word. Now we started seeing "intelligent graphics agent" or "intelligent login page agent". Finally in the late '90s AI moved to "Autonomous Agent", a word straight from the Department of Redundancy Department. But it worked! Nobody outside of AI seems to use it: my theory is that most people don't really know exactly what autonomous means.

View on HN

I mean... "crypto" !!! Can you imagine what cryptographers had to witness every day for a couple of years now, while the term was increasingly skewing towards a completely misleading meaning?

It's like if people involved in:

- ATMs

- eCommerce

- Foreign intelligence and signals

or any other similar field where crypto is essential started calling themselves 'crypto'. Yes, these field use crypto, but no, they're not crypto. Just like cryptocurrencies aren't crypto.

This is crytpo: https://cr.yp.to

This is crypto: https://crypto.stackexchange.com/.

Things that use crypto are not crypto.

The meaning of words change when the rest of humanity takes it for their own. Claiming "crypto are not crypto" will just leave you behind as the guy with his fist in the air cursing the world.

Nuclear. Hacker. Patriot. Rubber. The audience matters. Besides, I doubt that any cryptanalyst would mind writing out cryptography any more than they'd mind writing out TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256.

I'm stating crypto is cryptography. There's not multiple cryptographies as far as I'm aware so 'crypto are' doesn't make sense.

And yes, words change over time, but in that case I can just refer to (all) cryptocurrency as 'Bitcoin' since that's the common definition. In fact it's far more popular to refer to cryptocurrency as Bitcoin than it is to call that 'crypto'. Short version: this one isn't settled yet.

How did nuclear, patriot and rubber change?

>this one isn't settled yet

I feel like it almost is. Now even the non-technical crowd is amped up for cryptocurrency, and could really not care less about cryptography.

Judging from the mad rush into investing in “cryptos” I disagree that this isn’t settled.

You just may not like how it settled.

> this one isn't settled yet

It pretty much is. Cryptography is a niche of computer science and used the word for a while but now that humanity has chosen to use the word for the entire field of distributed and trustless systems it will now and forever mean "cryptocurrencies".

The larger group pretty much always wins and this case is no exception.

Nuclear was a branch of physics, then became a weapon, then became a US senate procedural change "going nuclear". Patriot was an anti-government rebel, then a pro-government supporter, then a missile, then an act to boost government and corporate controls within digital content, and sometimes an American football player. Rubber was a tree, then a commodity, then a condom. "A rubber".

Nuclear still means to do with nuclear energy: it being used as a simile doesn't change that meaning.

'patriot' means someone who stands for what is perceived to be American values. A missile being named after it doesn't change that.

Rubber is more interesting and perhaps a better comparison 'are you wearing a rubber?' during sex or 'when the rubber hits the road' in a speech, or 'do you have a rubber' during an exam.

So lets let's say someone says they 'specialise in rubber': it seems reasonable to believe they mean the material: they don't mean condoms or tyres or erasers.

Very bad examples. You’re just saying that a lot of common words have multiple meanings based on the context. None of those meanings is prevalent as in using crypto now for shortening cryptocurrency.

I think that “Rubber” being used to indicate a condom is just a very plain synecdoche. All the other meanings are there, still. Plus, rubber’s main meaning is still referring to the material

> Things that use crypto are not crypto.

It's questionable that these thing even use "crypto". Cryptocurrencies are named so because they use "cryptographic signatures". While the modular arithmetic used here is very similar to that used in "encryption", "hidden writing" doesn't take place, as the name might suggest. Signatures are more of a dual concept to cryptography than actual cryptography.

(One more note: applying a hashing algorithm may be considered "one-way encryption". In areas where this is used for signatures, an argument might be made that this is cryptography).

It's still common to hear barely informed business stakeholders freely interchange hashing, encoding, and encryption. "Are passwords encrypted in the database?" "Uh...no?" "OMGWTF?!"

Cryptozoologists are also feeling the pain.

And just as AI is being overused to solve problems, so too is the "blockchain", inevitably making users lives worse in the long run.

> AI is being overused

double-checks URL bar to make sure he is still on HN

Downvote me all you want but I think the tech sector is currently suffering a veritable Machine Learning craze.

Give you the benefit of the doubt and assume that last bit was sarcasm?

Your comment sounds as if you thought that AI researchers had invented the term “agent”.

They didn't? Next you'll be telling me the word dates from Middle English. :-)

I'd guess it's the other way around.

'Autonomous' is an unusual enough word that you remember what it means (especially once you understand the 'auto' prefix), or look it up because it's strange.

'Agent' is so familiar though that I can imagine lots of people don't attach its literal meaning to it in the sense of 'having agency'. Popular uses are things like 'secret agent', 'cleaning agent'... "probably means like, 'person', or 'thing' or something".

> Consider "Agent", which describes a very specific thing (namely autonomy), but that didn't stop everyone and their dog outside the field from calling their software a "database agent" or a "compiler agent" or whatnot, equating "agent" with "program".

Actually, the only problem illustrated in your complaint is that you are unaware of the meaning of the word "agent".

https://en.oxforddictionaries.com/definition/agent

Considering how the definition of "agent" is "A person or thing that takes an active role or produces a specified effect", it is quite clear that an agent isn't a program, but programs (or modules or components) are quite obviouly agents. In fact, each and every entity in a sequence diagram is an agent.

> you are unaware of the meaning of the word "agent".

Imagine that! Could it be possible that as a native English speaker, I am entirely unaware that "agent" means something different in general English usage? :-)

Look, AI has a specific coined usage of the term. The outer computer science mis-usage is not derived from the traditional English meanings of agent at all, but rather is very intentionally derived from the AI usage to take advantage of its buzz. I'm sure you can appreciate the difference between coinage and intentional misappropriation.

AI even has this problem among it's researchers. What the fuck is a neural network?

I was also curious about the use of 'neural networks' in deep learning or whatever. My PhD research was working on problems in computational neuroscience.

I think the use of "neural networks" is because the topology of these deep learning networks is similar to what is observed in biological brains.

Brain networks and neural networks are not similar in structure or function, other than the superficial abstraction of “nodes and connections” (graphs).

They’re about as related as genetic algorithms are to actual genetics.

Which is what he's saying, the use of the term is because of the extremely-vague-homeomorphism implied by the structure

Unfortunately, the word choice can lead to "they've gots em a neural networks and an AI singularities and them robots coming to take over"

>extremely-vague-homeomorphism

Analogy is the concept in its full generality.

If only. Analogy is more like the bastardization of an Ideal Form

Thats a very foreign view of the concept for me. I think of analogy as a non-rigorous functor.

A mapping between concepts and relationships among concepts in one context to concepts and relationships among concepts in another context.

Genetic Algorithms and Neural Networks are clearly inspired by the scientific fields of Genetics and Neuroscience.

I'll disagree with this. Cybenko's approximation for a multilayer perceptron is simply a concrete expression for Kolmogorov's representation of a continuous function, which was created in order to solve Hilbert's Thirteenth problem. This has absolutely nothing to do with the brain. Yes, Cybenko uses the word "neural network", but just look at expression (1) in his paper "Approximation by Superpositions of a Sigmoidal Function" and compare that to expression (1) in Kolmogorov's paper "On the Representation of Continuous Functions of Many Variables by Superposition of Continuous Functions of One Variable and Addition". The brain diagram used by most neural network books is really just a graph diagram of a matrix-vector multiplication used in the superposition. While I can't claim to know everyone's inspiration for their research, there really is a mundane explanation and history of these algorithms that has nothing to do with the brain.

As a data scientist dating a neuro-scientist, I can confirm that the two fields are quite apart in practice (except from the occasional image processing of fMRIs) — however, Cybenko’s work started in the late 70s; the inspiration for using a formalism of unit perceptrons organised in layers (as well as the convolution structure) comes from a paper on cat’s vision system published in 1962 [1]. As far as I’m aware, all key papers (Hinton, Le Cunn) cite Hubel & Wiesel as their explicit inspiration.

The current practice has moved far from neuro-science, but I don’t think it’s pedantic when discussing machine learning to talk about “artificial neural networks” and acknowledge that neuroscience is still working on understanding the structure of brains. Some people, focused on their research, didn’t have that discretion. I wouldn’t imitate them.

[1] http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1359523/

Neural Network describes a class of algorithms. It's often misused in marketing (buzzword), but there is an engineering definition used by people that practice machine learning.

A perceptron, the basic unit in a neural network, has one or more inputs, an algorithm for processing the input, and one or more outputs. Layers of perceptron are often used so that the original input gets transformed over each layer. The most common output for a neural network is a classification. The major variants of neural networks are ways a perceptron's algorithms change, the types of algorithms inside a perceptron, the number of layers and the number of inputs/outputs. Neural Networks are computationally heavy to train, but not computationally heavy to use once created. They are specifically good for working with image data.

http://natureofcode.com/book/chapter-10-neural-networks/

Sorry, I'm not actually confused as to how the word neural network is used. I'm saying that it's stupid nomenclature. But thanks for clarifying anyway :).

Are you referring to the distinction between the software frameworks known as an artificial neural networks and biology, specifically neuroscience?

AI has had more than one winter, you know.

Yes, he's probably aware, but hopefully we're not coming to one. Machine learning was a lot less useful in the 1980s. There's still a lot of amazing (and unfortunate) uses for machine learning for government and business left to be plucked.

The only thing that is going to die for now is any trace of the idea from non-technicals is that if we throw $10mm at it, it's going to come alive.

Big hype cycles typically have lower low point as well as higher high points. I don't know if another AI winter is around the corner, I hope not, but it is definitely a possibility. I am much more pessimistic about blockchains.

There won't be because it still actually is useful governments and businesses. They can tangibly use it to increase their functioning/profits - that was not the case during the AI winter. Computers sucked, data sucked, and there was no internet. I hope you're right in that the attention it is getting decreases. A reasonable limit of what we can actually do with what we currently have is probably within the next 15 years. Having 50 deep learning moocs that teach introductory matrix multiplication isn't really benefitting much either. Hopefully things come to a better equilibrium soon.

I'm very pessimistic about blockchains but I'm not sure if they're going away or not, either. Regardless of regulation, there's a lot of people screaming about blockchains and that doesn't seem to be changing: “I believe what really happens in history is this: the old man is always wrong; and the young people are always wrong about what is wrong with him. The practical form it takes is this: that, while the old man may stand by some stupid custom, the young man always attacks it with some theory that turns out to be equally stupid.”

> There won't be because it still actually is useful governments and businesses.

This was actually true before, AI has been useful, it has always gotten more useful over time, AI has always had limits, and AI winters occurred because expectations dramatically exceeded those limits. For there to be a winter/crash, hype just has to increase to a point that is way beyond what we can hope to accomplish in the near time. It doesn't help that the Chinese are taking the place of the Japanese, who in turn were a significant cause for the last AI winter.

Reply on news.ycombinator.com