Hacker News Reader

Comment by ghostcluster | Hacker News Reader

I found the author to be slightly irritating on several occasions, dropping veiled references to Valleywag-style anti Silicon Valley memes, and then I got to the part where he regurgitates that idiotic article about the brain not processing information, and there being something magical about human brains that cannot be simulated [0].

He is right about his claim of having no right to be called a “director of research", as it seems to me his skills center on cribbing thoughts pulled from other people's thinkpieces. It's clear that he doesn't have a deep background in either neuroscience or engineering and that he was brought to the company from a background in business journalism.

In his condemnation of the state of AI research, there is no mention of AlphaGo, or a description of the teachable pattern recognition techniques that have swept the deep learning scene over the last 6 years.

I'm sorry to be so harsh, but there is a certain tone to this piece, "let's hate all those startup a*holes", "Mark Zuckerberg can't write like F Scott Fitzgerald because his knowledge of liberal arts is too limited, unlike mine" that seems like a snooty class signaler among a certain hipster set.

There is a compelling story in here, but to me the general attitude is just condescending to everyone around him.

[0] https://aeon.co/essays/your-brain-does-not-process-informati...

View on HN

I dunno. The article seemed like a pretty good riff on the Great Gatsby and the dark side of Silicon Valley culture. The writing is promising.

He doesn't present an honest understanding of his own field, or the field of neuroscience, or the ongoing developments in the technology surrounding his own business, or its implications.

Even without extrapolating from the pattern recognition tools we have today, whole classes and ranges of jobs can be fully or partially eliminated.

Here is what he says about the state of AI:

> Even the most eye-catching successes claimed for AI in recent times have been, on closer inspection, relatively underwhelming. The idea that an autonomous superhuman machine intelligence will spontaneously spring, unprogrammed, from these technologies is still the stuff of Kurzweilian fantasy. Forget Skynet; at this stage it’s not certain we’ll even get to Bicentennial Man.

> These techniques might replicate discrete functions of a human mind, but they cannot capture the mind’s totality or what makes it unique: its creativity, its genius for emotion and intuition. There’s something else going on."

"the brain has spiritual magic"

Compare that to quotes from real live top human Chinese Go players defeated by AlphaGo last year:

> “After humanity spent thousands of years improving our tactics, computers tell us that humans are completely wrong,” Mr. Ke, 19, wrote on Chinese social media platform Weibo after his defeat. “I would go as far as to say not a single human has touched the edge of the truth of Go.”

? “AlphaGo has completely subverted the control and judgment of us Go players,” Mr. Gu, the final player to be vanquished by Master, wrote on his Weibo account. “I can’t help but ask, one day many years later, when you find your previous awareness, cognition and choices are all wrong, will you keep going along the wrong path or reject yourself?”

http://archive.is/qCwn8

AlphaGo is impressive, but it is an impressive parlor trick not impressive as 'artificial intelligence'

Playing a game with fixed rules and a finite set of potential states is something a computer can do.

Designing the computer that does that is intelligent.

There is no connection between the two and one does not lead to the other.

The fact that some of the people developing 'artificial intelligence' have such a limited understanding of what intelligence is no doubt contributes to the mocking tone of some critics.

I'm looking forward to the day when we have a computer that passes the Turing test but comments like yours still show up.

Scott Aaronson: “So this is about the state of the art in terms of man-machine repartee. It seems one actually needs to revise the Turing Test to say that if we want to verify intelligence in a computer, then we need some minimal level of intelligence in the human interrogator.”

https://www.scottaaronson.com/democritus/lec4.html

That’s just the anthropic principle restated.

Plot twist : parent is an AI !

> He doesn't present an honest understanding of his own field, or the field of neuroscience, or the ongoing developments in the technology surrounding his own business, or its implications.

Yes, but I don't think you are, either.

The fact is that chess and Go abilities aside, we are probably not even close to insect-level intelligence, and we don't have a clear path of getting there soon -- let-alone anything human-level. Current state-of-the-art so-called AI is basically powerful statistical regression algorithms, that are heuristic improvements over core algorithms invented in the 1960s, and there have been few theoretical breakthroughs since then (far fewer than in most other fields), so much so that many consider machine learning to be in the pre-science or pre-theory stage, being mostly about collecting data and trying out heuristics. It's silly to deny recent successes -- largely due to better hardware (although hardware progress is slowing down quickly) -- but we are behind, not ahead, of where we thought, even as recently as the 1990s, we'd be by now.

At this point we have no idea what role statistical regression plays in intelligence or whether we're even in the right direction. That statistics has become synonymous with intelligence (it used to be synonymous with lies) is certainly a cultural phenomenon that is not directly related to our actual knowledge of the field.

That computers perform some mental tasks (certainly more and more of them) better than humans has been a fact of computing since the 50s, and often a cause for wild claims. The invention of neural networks in the 40s and their implementation in digital computers in the 50s led some very respectable people (like Norbert Wiener) to declare that the problem of the brain will be solved in 5 years. The pragmatic Alan Turing thought that was ridiculous and predicted it would take 50. It's been almost 70 years and we haven't yet reached insect-level intelligence or anywhere near a complete understanding of the insect brain, so at this point, any claims that we are on the cusp of something, or starting to believe that our statistical regression algorithms reflect the beginning of intelligence is... misplaced.

On the other hand, it seems like we have not learned the lessons of misplaced confidence in AI, despite our relatively slow progress, and things are worse now as that we actually have some algorithms that are useful in certain restricted domains that we insist on calling AI, thus causing people to use them in domains where they are not only useless but downright harmful. In the meantime, some people draw attention to the dangers of real AI -- which may be anywhere between decades and centuries away (I believe we'll get there some day but we have no idea how or how soon) -- while distracting from the very real and already present dangers of "AI".

The Blue Brain project managed something “as big and complex as half of a mouse brain” in 2007, so I think your claims of not-even-insect-level are outdated.

The OpenWorm project doesn't even manage a worm though.

So? OpenWorm is (barely) crowd-funded and trying to simulate the entire body not just the brain. Seems like a decent effort given their resources, but on a scale of Radioactive Boy Scout to Manhattan Project they’re a Farnsworth Fusor.

(Which is to say “I ought to volunteer”).

Not so sure that claim is right : http://www.artificialbrains.com/blue-brain-project

Seems like they managed a honeybee (but I am not sure that it ran in real time or how they validated that) but were hoping for a rat brain.

I'm quite sceptical - I don't believe that there is a good understanding of how a single neuron functions, or agreement on the taxonomy of neurons or an understanding or agreement on their interactions and arrangement apart from in a part of the vision system where there do seem to be some good models.

Thanks for the link. I noticed it was “As of August 2012”, which despite being old still falsifies my prior belief. (Gell-Mann Amnesia, I incorrectly relied on the press for science). Unfortunately I can’t seem to find anything more up-to-date and just get more confused journalism.

That said, isn’t the point of the Blue Brain project to answer your skepticism? When we have all those things, the only thing remaining is to see what can be left out of the sim without compromising the behaviour?

> hardware progress is slowing down quickly) I would be interested to know more about this. I haven't heard yet that the progress of GPUs cores for example is declining quickly ...

The problem is Amdahl's law. You can only parallelize so much. While the brain is certainly extremely parallelized, neural nets do not employ the same algorithms as the brain, and so, unless we find algorithms that are more amenable to parallelization, Amdahl's law is going to get us.

Most modern neural networks implementations are parallelized. And that is why we can run them extremely well on the GPU. For example Volta GPUs delivers 5X increase in deep learning training compared to prior generation NVIDIA Pascal architecture. This is why I was asking for clarification about the hardware claims.

Of course they are, but they're not as parallelizable as the brain, which is why they're subject to Amdahl's law.

> we are probably not even close to insect-level intelligence

I think the problem is that we do not have a slightest clue what is (even insect-level) intelligence (or consciousness, which is often mixed up in the discussion).

That's right. Some have tried describing intelligence as a general problem-solving skills, but this is clearly false. Humans are terrible at finding even approximate solutions to NP-hard problems, which are certainly general and very common. It seems like intelligence is an ability to solve many problems that humans and animals face, but no one has characterized it more precisely, AFIK.

Where is the AI that can fold laundry (clothes, linen, towels)? Do laundry (sort, pre-treat, load, unload, clean lint filter)? Do dishes (clear table, scrape food into compost or trash as appropriate, separate to recycling as appropriate, load, unload, put up)? Keep a lawn (mow, edge, trim hedges, move trimmings to compost, trim trees)? Put up Legos after a 5 year old? Pick up around the house and tell you where it placed or last saw an object when queried?

And where is the AI in a humaniform that does all of the above?

There are tentative steps towards some of those activities, but we’re still in the early years with imbuing our machine intelligence models with the equivalent of our kinesthetic sense, object recognition and classification, and natural language interaction. And it is far from clear that we can get there with purely current statistical heuristic-oriented technology. We can only try, but the amount of effort required just for folding clothes to date reminds me of the elaborate Ptolemaic models, or as if we’re trying to build Excel by poking ones and zeroes into memory.

More tinkering required, be back later.

Laundry folding robot with actual smarts to it: https://m.youtube.com/watch?v=DzK38ylMTXk

Inferior to a human, sure. But a start.

Ha and theres this: http://dalybulge.blogspot.co.uk/2018/04/cyberpunk-sabpk.html

This is the wrong way to think about it to be frank. Animals can do things humans can't that doesn't mean that humans can't come up with ways to make what those animals can obsolete. Humans don't need wings to fly, we don't need to be able to run faster than a leopard to be able to move faster than a leopard.

AI will not always represent itself in a one to one relationship with humans to be able to compete or outcompete us. Just as an example a lot of things have become digitalized which have rendered many elements that used to exist in physical form into digital form, music is a good example of that.

So sure there are areas that machines aren't as good at yet because they haven't practiced it enough but it's literally just a matter of training and improving not some fundamental problem that can't be solved.

> ...but it's literally just a matter of training and improving not some fundamental problem that can't be solved.

I heard this echoed many before when pawing through the library stacks in my uni days looking through the littered corpses of AI trends in the past. I believe that we will eventually get to strong AGI. But after either reading or seeing in-person the hype machine sprout up and wither around symbolic programming, semantic programming, neural nets, fourth generation, expert systems, perceptrons, Connection Machine, etc., I'm gun-shy around any proclamation that achieving strong AGI is "just a matter of...<insert-single-solution-space>". The results so far seem to indicate pure cognitive processing is very amenable to the toolbox we have built up to-date in AI research, hence the breakthroughs in game playing.

Manipulating and interacting with the material world and humans however, and the results are a little patchier; I suspect we have lots more work and research ahead of us than we currently realize. When we do get some initial results like the laundry-folding machines, they're single-purpose and uneconomic for mainstream middle-class adoption (not to speak of working-class), and often with lots of attached caveats like Tesla AutoPilot. Instead of all these discussions of whether or not we will get strong AGI, I prefer to see everyone assume it will happen, and when we don't get the incremental result we were anticipating, say, hmm, that's interesting, I wonder why...

I want to see the hype tamped down to the point we can steadily chip away at the overall problem space, and accelerate AI research results and organically reach strong, economically-available AGI sooner than continue experiencing the disappointing two-steps-forward-one-step-back our industry seems to so far historically take in this field. The hype says we're a sprint away from unlocking all sorts of benefits promised by strong AGI, when we are better served accepting the organic incremental benefits as they occur during our acknowledged marathon, and using those incremental benefits as stepping stones to greater understanding.

It took billions of years to evolve us. We have only been working on this seriously for less than 100 years and the progress is reminiscent of the Cambrian explosion.

Again humans evolved from dumb atoms why is that easier to believe?

We don't create factories around people. Reinvent fashion, kitchens and house plans to fit the machine. That's very doable. Restrict the solution space to find the answer. (Let Marketing handle the user acceptance issues)

I believe there is an alternate reading you may be missing. Viewed in a different light the example of Go you cite is not that different from John Henry vs. the steam engine. The history of industrialization is the replacement of humans by machines for specific tasks. People tend to oversell current AI technology as somehow deviating from that long path. It's a fair point to make.

For me the interesting theme was the exploration of the character and dubious success of the mysterious Jim, who is using his connections to ride a wave of poorly understood and possibly malevolent technology to a grand house in the country and membership in the upper classes--almost like flotsam on the rising tide of economic progress. There's a lot of Gatsby in there and some hints of Graham Greene's quiet American as well. Just as you can argue that AI the technology is part and parcel of industrialization, its social effects recall recurring conflicts in American society and culture that authors like Fitzgerald have been exploring for 150 years.

As for other critiques of the style from the Hackerati I would just say yes, it seems like the work of a young writer. Good writing is hard to achieve and it's typically preceded by a lot of bad writing. To paraphrase Senator Palpatine, we will watch his career with great interest.

Fixed: typo/it really is hard to write well

The article you reference makes the same mistake as Searles Chinese room argument. It somehow assumes there is something magical about the human mind.

I had a brief argument with Robert Epstein (the author of that article) because I find the argument that humans don't actually store information to be quite misleadning and missing the point.

The most obvious mistake is this:

"We don’t store words or the rules that tell us how to manipulate them. We don’t create representations of visual stimuli, store them in a short-term memory buffer, and then transfer the representation into a long-term memory device. We don’t retrieve information or images or words from memory registers. Computers do all of these things, but organisms do not."

This is both true and false. It's true we don't do that like a computer but its wrong to claim that computers fundamentally do that too.

That happens several layers of abstraction up and so a computer fundamentally doesn't actually store an image or a word either it manipulates atoms and turns circuits on and off and several layers of abstraction up it gets translated into meaning first by machines then by humans.

Anyone who have a hard time believing machines can become sentient should first ask themselves why they have a harder time believing that than accepting that dumb immaterial matter somehow have become the pattern recognizing feedback loops that are us.

Searle's argument isn't that there is something 'magical' about the human mind, it is that biological systems are fundamentally different from mechanical or digital systems and that sentience (not intelligence!) is a unique feature of biological systems.

This argument deserves much more recognition than it gets because 1. it's at this point still empirically true, we have not observed non-organic sentient life and more importantly, because not Searle but everybody else employs 'magic'.

Searle's point is simple. Computation is subjective. Electricity flowing through a machine doing complex things is just a physical process like anything else. You (the sentient observer) classify that physical process as meaningful, but a computer is no more 'computing' things than a falling pen computes gravity.

So sentience really is related to physical agency and sensory experience in the world, which creates conscience in organic brains. That doesn't imply complexity or intelligence or understanding. Syntax and Semantics are different things. Your pocket calculator processes the syntax of mathematics, but it does not understand the semantics of mathematics. A compiler processes symbols according to rules, but it does not understand the meaning of the computation, it has no cognition. It might be very good at what it does, but it has no capacity to understand. That's the essence of the Chinese room, and it's still a convincing argument.

An even stronger point might be made, namely that sentience actually limits intelligence. That it requires a degree of slowness and introspection that is unsuited for fast decision-making. For a fictional treatment of this, Blindsight by Peter Watts is an excellent read.

It really don't in my opinion.

Of course, the person in the room doesn't understand Chinese just like the individual neuron in a Chines person doesn't understand Chinese.

It's the entire house where the person in the room is just one part of the entire system.

And yes he does imply that there is something magical about the way humans are pattern recognizing feedback loops vs. machines as humans came from immaterial matter ourselves if he doesn't his argument simply do not hold up as there is nothing magical either in the way human conscience is a byproduct of simpler systems all forming to become a scentient one.

If you can buy that humans have evolved from dumb atoms then you have too look that not how humans and "machines" are different but how they are the same.

The samenes is that we are pattern recognizing feedback loops and that our sentience comes out of something non-sentient.

So either something magical is in play or there is nothing that hinders machines to become sentient either from what we know.

A human consists of milions of sub-systems like a calculator and yet we are somehow sentient.

Furthermore there is no know upper limit to how complex silicon based systems can be and so the right answer really is if anything "we don't know" not "because the person in the room doesn't understand chinese it proves that systems can" the person in the room is not the system the entire house including everything happening outside of the room.

In other words unless searle is claiming magic at some level nothing, absolutely nothing indicates that machines can't become sentient.

With regards to your last point then that's the wrong way to look at it.

A better way to understand why it's possible is to start from omniscience and then realize that omniscience means you are aware of everything and thus have no perspective where as all systems that can handle information potentially can become sentient the more complex they become.

This is a very interesting interpretation of Searle's argument (which also has some overlap I believe with some of Douglas Hofstadter's ideas), and I will begin to read Blindsight shortly, as it seems a very intriguing novel.

> idiotic article about the brain not processing information

How about the brain creates information from constant interaction with the world based on the kinds of bodies we have and our needs/wants? This information doesn't exist as information until the brain creates it. Information is the product of minds. It doesn't exist in the world on it's own to be processed. As such, the brain is something other than a computing device. Computers exist because we figured out how to arrange physical systems to process information that's meaningful to us. But to nature, it's just a physical system (and not even that, since physics is a model of nature we create).

That's Jaron Lanier's paraphrased argument against thinking of the brain as a computer. To say that information exists in the world to be processed is to make a metaphysical commitment that information exists ready made for us.

> and there being something magical about human brains that cannot be simulated

It doesn't have to be magical. There are different philosophical views on the world and the mind which lead to different conclusions. If one takes the hard problem of consciousness seriously, then consciousness cannot be computed. Not because of magic or the supernatural, but just because consciousness is not computable, since computation is itself an abstraction (Turing machines don't exist on their own anymore than do any other mathematical systems). Unless your metaphysics falls along the lines of Tegmark, Plato or Wheeler (it from bit).

Instead you can think of The brain as an information creator. We give meaning to the world. We build models. The world itself just is, it's not information, math, physics or symbols.

"Not computable" is itself a strong claim that's never been proven.

What is true is that nobody has done it yet. The process is a mystery in the sense that it's not understood, which means that we don't know if it's computable or not.

The argument at the moment seems to be "define a problem that a computer can't do that a human brain can"... "I can't because expressing that problem is beyond the machinery I have developed for cognition, and it may always be".

What is certain is that there are uncomputable problems, but are any of the problems that humans solve in order to speak, act, socialise uncomputable? Some people think that because they are solved within the physical universe then they must be computable but that implies that the physical universe can be simulated (in principle) by a universal turing machine, since we can express problems (using the machinary of the universe) that a universal turing machine can't do then there is a gap that permits the possibility that there may be some process which is not simulatable by a universal turing machine.

In my belief free will/autonomy/initaitive/creativity are expressions of that process, but belief is not an argument.

> Some people think that because they are solved within the physical universe then they must be computable but that implies that the physical universe can be simulated

If you believe the brain exists in the physical universe, that means you can build a physical system that also solves the same problems.

Yes : but...

- "It" wouldn't be a "computer"

- you / someone would have to be able to understand it, which might be impossible (for a human)

- you would have to be able to construct it, which might be very very technically hard

but yes (ish)

> If you believe the brain exists in the physical universe, that means you can build a physical system that also solves the same problems.

Sure, but in the case of computation, is it possible to make a computing system that has subjective experiences? Maybe consciousness isn't something that can be expressed in computational terms, because computation is itself based on abstraction.

There is no way to define subjective experience such that you can tell whether something has them or not. The question is meaningless.

I think that is extremely austere, and misses a large part of the value of science and philosophy. In some ways the questions of cosmology matter far less to the vast majority of humans, and have in practical terms less qualification and quantification for those people than subjective experience communicated by drama, poetry and art. Stating that this is meaningless ignores the suffering and joy of humanity, and makes an unwarrented and low utility judgement about the set of beliefs that are legitimate in finding how things are.

Computers interact with the world too. I'm looking at a screen that produces patterns of light based on the internal state of my computer. How is this different from a brain interacting with the world? The brain is a finitely sized hunk of matter and matter seems to follow laws. We currently have no reason to assume that those laws can't be simulated by sufficiently sized computer, so anything observable the brain does a computer can do too.

Yes, computers are physical systems. But what does their interaction mean without humans around to interpret their output?

the tone is pretty standard baffler style. its meant precisely as a provocation, thats their whole thing. Its also an openly leftwing publication fwiw. Personally I find it way more refreshing and honest, then, say, the NYT op ed pages, in terms of being honest about why they take the subjects they do and why they present them in the way they do.

I enjoyed it and I think that it is helpful in the sense of calling "naked man" at the emperor. It doesn't do any harm at all for the AI community or startup community to look at itself and think hard about what it's doing and saying.

This time round there will be no million fold increase in compute power to bail everyone out!

If you want to rise to an 'empereor has no clothes' caliber piece, it would help to demonstrate a comprehensive understanding of the fields you're criticizing, and not arrogantly cite bad science essays, and not ignore the actual state of the art techniques in that domain.

You need to present the best arguments from the side you want to critique and then prssent a case why you think they are wrong. Calling people names and avoiding difficult challenges to your thesis is not the way to do it.

Well - I'm not so sure, you're setting a very high bar which makes it difficult for people with a different background to make points (badly in your view, but pretty well in mine) which the community needs to hear.

This isn't a Ph.D. exam, this isn't a thesis - it's an outsider calling BS. I'm not impressed by the counter arguments advanced so far. Let's be honest, Alpha Go and Alpha Go Zero are surprises in that they have shown that Go isn't as astonishingly difficult for approximate search - which everyone thought it was - but until we see the real world applications it's all of intellectual interest.. which is the point of the article.

There are a lot of folks who I respect making claims similar to the company that is featured in the piece, I'm really disappointed by that because everthing that we know about learnability is ignored with the cry "we've got deep networks now". We've don't know why dnn's generalise as well as they do but shouldn't, but it's no excuse to just abandon our sanity and go out and bet large amounts of other people's money on them doing things that they can't.

This money, btw, should be spent on hospitals and roads, not on providing near 7 figures for these people.

If you don't even understand the basics of what's possible in a field and what's not, you can't make a convincing refutation of that field by just calling certain people names and citing bad science essays that you also don't understand.

Reply on news.ycombinator.com