This goes in the "don't watch while stoned" pile.

I konw such cases Chomsky talks about. Richard Feynman told this story once about quantum physics, it goes somehow like: The mayan scientist priests observe the sky. They observe it year by year, and after many years they come up with a prediction for the king: Put 365 beans in a bowl and take one out every day. When the bowl is empty, the stars will be at the exact same spot they are now. The king says. Ok. but what do beans have to do with stars? Feynman said, we are in the same stage with quantum physics. We can make good predictions, but have no idea at all how it works.

yes, that was precisely the question that was on my mind.

GET READY FOR A COMMENT SECTION FULL OF AGGRESSIVE PEOPLE INSECURE ABOUR THEIR INTELLIGENCE.

"I don't do the wiggle-dance, because I'm not a bee." -Noam Chomsky, at a talk I saw him give about the evolution of language

Persons whom, mock Noam Chomsky, for speaking on topics that he is not a scholar in simply fail to see how much time and real effort he puts in each day to understand and explain them. I have not known a more hard working scholar and activist. Disagree with him that is fine. But, do not belittle his efforts that is simply being utterly dishonest.

I take Noam's comments here to be the distinction we now make between prediction and inference. A lot of the machine learning techniques we use today use models that become so complicated that they cannot be interpreted. The goal of these models, however, is prediction; these models are successful because they can predict things very well. Classically, science has been a field of inference, where we try to infer how systems of nature function. In this way, some scientific disciplines are becoming less interested in problems of inference and more interested in problems of prediction.

That was the longest question I've ever heard in my life.

I just decomposed listening to this question

For those wondering what the hell Chomsky is talking about with respect to the measure of success, it may be helpful to see the video below. This shows a neural network (one of the machine learning techniques Pinker mentioned) learning to make speech-like sounds. This sort of thing could be useful in lots of ways. It's clear that with a bit more work, you could have several tracks of this overlaid to sound almost indistinguishable from an actual crowd of people in conversations. Could be useful for making movies, for example. It's also clear, though, that there is no semantic content in the speech. It doesn't mean anything, even if it sounds great. In particular, even if it generates sentences, or hour-long lectures on particle physics, there's no meaning in them because they were drawn at random, in a sense, out of a clever neural network hat. This is Chomsky's point. You can call a neural network successful when it generates speech-like sounds, but that doesn't get you any closer to teaching a machine to produce _meaningful_ speech. In much broader terms: Pinker asked Chomsky to what extent machine learning has been successful in processing natural language. Chomsky said that the only successes he was aware of involved machine learning that started with a substantial injection of human knowledge, where machine learning from scratch has produced nothing interesting (in language). kzfaq.info/get/bejne/hK1df6SFp8Cycqc.html

I watched it without sound. Thought I was watching Godfather

Strangely enough, most people in this comment section simply don't get it that Chomsy and Pinker are in full agreement here regarding the necessity of building in biases for upping the efficiency of pattern-extracting algorithms. Responses can be roughly divided into 3 categories; 1) ain't got the slightest clue what what these guys are talking about because either too convolutedly and/or boringly stated 2) Chomky's a politically biased intellectual who can't just accept his GU wasn't of any use for language processing devices 3) Pinker was all strung up for being face-to-face with the man himself

that is the most long-winded question i've ever heard in my life

Chomsky always tries his best to explain clearly things as he has come to understand them.

Does Steven Pinker look nervous to you?

6:08 A ball will not roll down a frictionless plane, it will slide down without turning.

This would be a good video to use to analyze Noam's hand gestures and non-verbal communication.

Noam's dry irony was hilarious. The people around him didn't seem to get it. Or maybe they did....

Chomsky is such a boss. A.I. "scientists" exposed as unscientific dreamers.

As an engineer I can see that this is an engineering approach to making a device that works. And I guess that AI is still in that phase. There isn't a complete scientific model for AI or the physical process of parsing speech, so these guys are doing what engineers have always done, they fiddle around until it works and let someone else develop the theory latter.

Trying to understand Latin would have been easier than following that exchange. I don't even know why I clicked on this video. So much hope.

Chomsky basically summarized the state of machine learning perfectly. We are getting these statistical models that are becoming more and more accurate in prediction/classifying previously unseen examples. However, what we are getting are approximate models to human counterparts. It's more of an engineering solution where we can use these machines/models to build useful applications. However, we are no closer to understanding how human brain works.

Usually Pinker is very articulate, but here he seems to stumble when asking his question. Is he nervous in front of Chomsky?

This is why I dropped out of English language studies and got an English lit degree instead. I really don't understand what these two are talking about, but I know they are both geniuses in the field of linguistics.

I'd like to ask Chomsky what he thinks about vocal fry.

And I thought he was going to ask him what his favourite colour is

That was actually a great point. Since there were never supercomputers and myriads of data in the past, scientists were limited to creating their own theories that had a good predicting value. However, the role of science seems greater then that. It's getting the truth about the world with any means possible. If that means running statistics on large amounts of data then so be it. Predicting value without any underling theories (though machine learning can often point you in the direction of generalizable patterns or theories) is still a value.

Wow. Yannis Varoufakis said roughly the same thing about the pure economics models with their lack of relation to real world economics, in a later discussion with Noam Chomsky on the state and mechanics of the European Union. Another eye opener.

Translation: Pinker: 'Many of the recent advances in AI programming and cognitive sciences are down to the use of probability and statistics based on learning but your fields - generative grammar and semantic memory - do not use these techniques at the moment. Since AI and the cognitive sciences are analogous to your work, do you think they might in the future also employ statistical and probabilistic methods?' Chomsky: 'We've used statistics before with more failures than successes. For example, work was done on statistical analysis (of the sound of speech) to identify boundaries between words which failed. If you combine this with existing audible features that we know of (unsure about these) then identifying word boundaries statistically becomes more successful [this is not really addressing Pinker's question]. There is work being done using the sorts of techniques that AI and cognitive science uses (Bayes Theorum etc.) without actually looking at the structure of the language. They have an odd definition of success.' Chomsky then digresses to making an analogy of the use of these techniques in modelling bee behaviour. He describes the difference between a bee scientist looking at individual behaviour versus taking video of many bees and modelling the aggregated behaviour. What you get are approximations of bee behaviour that are better than what the bee scientist gets. He then suggests that it is strange that when people write papers about this modelling they claim 'success' in their modelling and Chomsky knows of no other science where approximations are deemed as success. I am unclear how this relates to the work being done using statistical analysis and probability in linguistics which he referred earlier. I would like to note that statistical analysis, Bayes Theorum etc. are very useful in a host of modelled systems and that there are many, many branches of science which rely on approximation of behaviour and complex systems.

Once you reach a certain level of intellectual sophistication, once you've known you field so well that you see its connections to philosophy, neuroscience... and once you developed great work ethics of reading enormously, of retaining information like him (encyclopedic memory) and you know the means by which to research, to connect and extract conclusions, you can be a humanist, a person whose interests and depth of knowledge covers many fields of interest... Chomsky is extremely efficient in reading, memorizing and writing.... He wrote many many books! He is one of the top humanistic minds of our era!

So basically, is he just saying you can either observe a lot and make fairly accurate predictions without understanding the concepts (probabilistic method), or alternatively you can build a simplified model and improve understanding of underlying concept slowly, but get crappy results in the meantime(his method)?

Mayor Michael Bloomberg thank you for having me .

Thanks for posting. Chomsky is such a brilliant man and it's amazing to think of just how long he has been a voice of reason.

When was that discussion? (Not the upload.)

That was a good fucking answer, and that’s what all this evolution has done for me, writing this on YT to add data to the algorithm

Chomsky is so bright it is alarming. How can he remember all this stuff!

Chomsky is the master of takedowns. I think he's basically saying, at the end, the "approximation of unanalyzed data" is novel when conceived as a sort of system without parameter or purpose. You can collect lots of randomized data that seemingly has no meaning, and from the perspective in which it is captured, after breaking it down into arbitrary units connected to nothing in particular, get some idea of what 'generally' will or is likely to happen in the future. But its purpose as collected data is simply to be analyzed. It has no bearing on anything meaningful to human beings.

Emotionally, Chomsky's attitude feels right. Its hard to see brute-force computing approaches as being an appropriate measure for science, where theory and carefully designed experiment 'ought' to be the way advancement is made. However on the longer view, I don't think enough credence is given to the influence of technology itself. Chomsky formulated generative grammar when the mathematical tools were available from computer science. Today, raw computing power give rise to big data approaches. Obviously their work strives to be an incredibly nuanced synthesis of linguistic evidence, brain studies, evolution etc. But there is also the sense that when a new technical approach arises, there is a tendency to push it to its limits, to harvest any low-hanging fruit, and see how things look from a new vantage point. And its my belief that in the end, its the new technique that advances things, more than the intellectual grand synthesis.

Chomsky seems to like wearing jeans. I saw him in jeans in so many conferences, whereas other people would dress formally.

Took the words right out of my mouth, Steve.

It's always a pleasure to listen to Professor Chomsky discuss linguistics. The field has widened greatly since the 50's into computational parcels where brute force methods can be applied. The Professor maintains the layer of abstraction necessary to provide an overview of the nature of the problems​ being tackled.

What did i just watch?

RIP Marvin :'(

MArvin Minsky kept quiet. He's on Web of Stories (KZfaq) trashing Chomsky's whole project as essentially a waste of time

Noam looks like a gangsta here

0:29 = the typical sideline on a Saturday afternoon at my grandpa's bacci court.

Yes, it could be used to predict but would it produce any understanding?

Seeing this comment field, I feel like I've stumbled into a room full of angry people, and yet I cannot see any reason for anger. Why the hate here?

Chomsky is in his late 80's in this video!

Noam's "novel notion of success" at 5:00 seems to be an argument in favor of science as _statistical inference about a physical model with many training examples_ as opposed to _controlled experiments on approximations of reality inside a laboratory_. For example, you can predict the location of the moon relative to the Earth at a given date and time by 1) applying the laws of motion under ideal circumstances, e.g. the moon occupies a plane with a constant z-coordinate that bisects the Earth; or, 2) record video of the night sky for several months, convert the video into quantitative data, then feed the data into a statistical data structure and let it learn.

Mr Chomsky, sharp as ever.

chatgpt says: this didn’t age well

If bee scientists, through laborious experimentation and analysis are able to predict swarm behavior with 95% accuracy, and a machine-learning model has been built that can predict swarm behavior equally or better, what's a key difference? The former results in _knowledge_ and _understanding_ (i.e., true science), the latter, simply information transmission. Taking a look around at our current culture, one can see how important the distinction is.

Is there anyway to get the entirety of this talk?

Reading the comments. the best way to understand something is to read something you don't like. Trite, but true.

When people are outraged by what you say - but they can't explain why - there's a good chance you're right. See angry sputtering comments below for examples.

amazing minds right here

Where’s the link to the full talk?

You would think a linguist would be easier to listen to.

Flash! Chompsky (I know, I know) says a ball rolling down an inclined plane is a thing that never happens in nature. Huh? I better go tell my granddaughter whose favorite thing at the moment is rolling a ball down the inclined plane in the front of my daughter's house, ie., the driveway, that she is doing the impossible. I doubt she will believe me.

6:35 Statistical analysis, unknown in the history of science? What about thermodynamics?

That was one of the most poorly worded question I have ever heard. It was basically Pinker trying to sound smart instead of smartly asking a question.

I like when he talks about things other than politics

Translation: While solely using statistical analysis in machine learning has been shown to work, within a limited scope, it appears to be much more successful when integrated with other techniques.

Pinker stumbles a bit because of nausea...he was with Chomsky, and that would make anyone sick.

Living things without brains; plant, animal and fungi basically cope with their environment through stimulus/ response and selected physical adaptations. Brains developed and allowed for more immediate adaptation through learning and instinct. The genetic language and math that humans have developed has is orders of magnitude greater in that these two adaptations allow humans to modify the future by greater pattern recognition from math and the ability to use language to enlist the cooperation of others in completing projects that surpass the scope of any individual’s existence.

I don't get this but question whether I should. Thanks.

first off, about the question asked....huh? i feel like Chomsky understood it but....i couldn't tell from his answer.

I don't suppose this whole conference is online in full anywhere, is it?

The math is already there, The brain uses statistics in form of stress. You can avoid math in your models only if you use premolded blocks. How enginners do a basic lowpass filter to alter a reproduced sound? They add a capacitor in the wire to the speakers. The math is there, but you don't need it to make the filter work. How do programmers the same filter: The sound records are not sulpted in solids, the sound waves are encoded. You need strong math (Fourier, I suppose) to transform the data before convert it to waves.

So Chomsky is guru in linguistics, created highly popular but controversial theories... And right now tries to explain why if his ideas about linguistics are so great, people who actually deal with computer translation have no use for it and rely instead on statistics. Maybe Occam's razor should be applied?

Language: The most recent and most important language is the language of "Texting". Textos is truly a language and must be considered as important, even More important than most of the tribal languages in Africa and the dialects spoken in many sub cultures on earth.

what an epic fucking answer

I haven't a clue what they are talking about. which is quite frustrating.

Fantastic

wheres the full video

Chomsky , the Prophet of Human future.

Ok, how is anything Chomsky said that complex? Bottom line, you can use statistics to predict what sound come next with a big sample, but it's better if you understood the language? Like duh? I think there are some fan boys here giving him more credit than is warranted

I don't understand why science wouldn't be interested in finding paterns and predicting the world.

who asks the question, "ill let you complete the sentence" ? thats not a question

all of the smartest men in history were those not afraid to be wrong.

Noam was 100% incorrect. He was so sure of his opinions. I know he is old now so reflecting upon that may not be possible, but maybe it can be used as a cautionary tale for others.

didn't understand the question, but what he said about novelty physics problem solving was embarrassingly wrong.

is that DR GOLDSTEINBERG?

WHY DOES SOUND LIKE HES ABOUT TO FUCKING CRY LMFAOOO

Why do I think Noam looks like Warren now ?

The hair is a bit out of style no?

I'm really disappointed in Pinker here. That was incredibly convoluted and abstruse, and didn't need to be. Who's he trying to impress? Normally I'm impressed by how easily and clearly he presents material.

Those little bottles of water are just daft. Whats the point?

Mr Chomsky is too old

The issue is that statistics yields prediction without explanation. While Chomsky used computer detection of word boundaries in speech as his example, Google Translate is another application. Nothing surprising here: If you know something about the laws governing the behavior of language, you can get better predictions, and with fewer data points, then if you use only raw brute force. I think Chomsky gets thrashed because of his nutty Marxist politics more often than on the merits of what he says about language, his own field, by people who aren't linguists. That reduction hasn't proceeded as far in language as it has in physics doesn't alter his basic point, however.

📍5:38

For all you idiots calling Noam an idiot for not understanding him, he's talking about different methods of linguistic optimization for AI/algorithms like that little CC button you see in the video so you can read what he articulates. It's quite comically meta in a way. I just refuted some idiots on this "refuting idiots" channel.

I understand how my brain works, I poses one and have done for the last 68 years, and the 'expert' is me, academia is speculative innocent ignorance.

I put this German sentence in Google Translate: "Der Frau gibt der Mann das Auto". This English sentence came out: "The Lord gives the man the car." I don't know what methods Google Translate uses, but that's a bogus translation.

These guys are woke af

Anyone know what he said.

Pinker is asking if chomsky's theory will be of use in the future, because unlike the new theories (big data), it has no statistiks and propability in it. Chomsky answers that statistics and propabilities are too superficial and that you can't get understanding out of it (only statistically data), and that it is therefore not very usefull. Pinker framed his question in a weird way, but he wanted to kind of embarras Chomsky, because he thinks his theory is a failure. But Pinker embarrast himself.

I like the name of your page

Was that even a question?