Not sure I buy this. The key difference between these algorithms and the brain is in the understanding of language. The language we speak and think with conveys meaning. These algorithms have no idea what the meaning of any text is, rather, they are just looking for patterns and combinations of patterns using huge libraries of language data. If anything, this is good mimicry. Maybe this is why NLU has split from NLP.
More is different. Even when things share the same base units they can have completely different qualitative properties at scale based on the structure and environments. You and the sun are both just leptons and baryons, but that's not very helpful.
In this case, these models have vastly different architectures, learn language is vastly different format, and are given massive corpora of words written from millions of random perspectives instead of receiving words from one perspective via targeted communications.
On what basis is there to expect such a system to magically develop the expressive intentional characteristics of human language when we can't even find another animal who seems to have this ability?
From my reading of where neuroscience is at the moment, it is a tenable hypothesis that the posterior neocortex (a major functional component of human cognition) is essentially a statistical inference machine.
There are a number of facts suggestive of this viewpoint:
For one, the apparent generality of the algorithms, resulting in a great deal of neuroplasticity and data-agnosticism. For example, you can wire visual input into auditory cortex and that auditory cortex will learn to see. [1]
For two, the similarity of Gabor filters learned automatically by statistical inference algorithms and the response properties of visual neurons. [2]
For three, the fantastic success, relative to other techniques at least, of such algorithms on natural language tasks such as machine translation.
Even its deficits are suggestive, such as the fact that the neocortex, like most statistical inference algorithms, can't easily do one-shot learning — for that you need a hippocampus.
Present-day machines obviously lack many abilities which humans have. Until we have a fully-functional AGI, it's hard to know for sure that there isn't a major missing piece which is nothing like the algorithms now under development. However, I'm not aware of any other algorithmic hypotheses of how the mind works which seem to so naturally fit the neuroscientific facts.
[2] Olshausen and Field (1996). Emergence of simple-cell receptive field properties by learning a sparse code for natural images. https://doi.org/10.1038/381607a0
The simple answer is no. Human understanding is more than just patterns. This would be akin to seeing the world through digital camera and thinking that everything just boils down to different patterns of red, green and blue.
What is the complex answer then? The parent's question is salient. Using your analogy as well, you could say humans see the world through the visual spectrum and reason about the world through that, although there is certainly much more to be observed beyond that spectrum - yet that is clearly sufficient for some subset of understanding.
There are entire academic fields and researchers of cognitive science, linguistics and neuroscience trying to unravel this. The question seemed to be made without taking a minute to appreciate this fact.
The meaning of human language is to do with having an effect on the world, having a perception of the world, and having motivations related to the world - such as conquest, subjugation, hearing lamentations etc.
It seems possible to have the information content of language without the above extension in the world. Information is "just" combinations of patterns.
Humans need surprisingly little data to learn language. If we can discover our internal presumptions (a priori probabilities), AI mightn't need so much data either.
I'm working on a side project right now where one of the applications is a platform to have AI engage in negotiation without a general purpose natural language. My project is an extension of the order writing system in Diplomacy into a domain specific language. I call it Dipspeak (like Newspeak from 1984). Dipspeak is a language for a universe where the only physical things are bits of plastic on a map of Europe and the only physical interactions between bits of plastic are hold, move, support and convoy.
This thesis implies language carries all of the meaning required to have an effect on the world.
To use a crude metaphor, language could be just an API schema. Is there content accessible via the schema that is more meaningful than the schema itself? For APIs, yes. I think so for language too.
Not to mention the possibility of side channels. That, at the same moment any human reads any text, they receive extra information to help them understand that information.
I agree with the sibling. You need to say what it means to "have an idea what the meaning of text is" if you're going to use this as an argument that neural nets don't understand language.
I think what it means to understand language is to be able to generate and react to language to accomplish a wide range of goals.
Neural nets are clearly not capable of understanding language as well as humans by this definition, but they're somewhere on the spectrum between rocks and humans, and getting closer to the human side every day.
I can't help but think that arguments that algorithms simply don't or can't understand language at all are appealing to some kind of Cartesian dualism that separates entities that have mind from those are merely mechanical. If that's your metaphysics, then you're going to continue to find that no particular mechanical system really understands language, all the way up to (and maybe beyond) the point where mechanical systems can use and react to language in fully all situations that humans can.
>These algorithms have no idea what the meaning of any text is
And yet, if you ask them what the meaning of a given text is, they’re getting disturbingly close.
What they don’t have yet is self-awareness, a model of their own repeated interfacing with the world. And an understanding of language in that context.
Yes this is cool. But c’mon — these scientists should know better (correlation vs. causation, yada yada).
Humans have always believed that their current state of technology reflects how the brain works — clay pots and waterworks, electric circuits, cogs, etc. Just because there’s an “I” in AI and a “neural” in front of “network” doesn’t mean you’re not just mimicking something as opposed to replicating how it actually works…
So then just assume that this is the next step towards the unattainable (during which endeavour we'll discover something better(?)) and go with it? :)
I am drawing conclusions here but it seems to me you're implying that simulating the human mind is unattainable, while I'm arguing that you're possibly right, but that in the process we'll find something that could be even better (and/or cause our own doom).
> As each model was presented with a string of words, the researchers measured the activity of the nodes that make up the network. They then compared these patterns to activity in the human brain, measured in subjects performing three language tasks: listening to stories, reading sentences one at a time, and reading sentences in which one word is revealed at a time.
It seems obvious that computers can emulate how humans process information. That's the tradition of the word "computer", which used to be a human occupation. The big question is regarding the sources of information. Humans seem to have access to more information than artificial computers.
Advances in AI are of two overlapping categories 1) catching up to human information processing, 2) capturing information from humans / from human information sources (many of which are not sensory).
The big question is whether artificial computers can 1) process information the same way we can and 2) achieve the same access to information that we have. We are making a lot of progress in 1), significantly less progress in 2).
21 comments
[ 4.8 ms ] story [ 62.1 ms ] threadWhat is understanding? Are we sure human understanding is not just patterns and combinations of patterns in our huge volumes of experience?
In this case, these models have vastly different architectures, learn language is vastly different format, and are given massive corpora of words written from millions of random perspectives instead of receiving words from one perspective via targeted communications.
On what basis is there to expect such a system to magically develop the expressive intentional characteristics of human language when we can't even find another animal who seems to have this ability?
There are a number of facts suggestive of this viewpoint:
For one, the apparent generality of the algorithms, resulting in a great deal of neuroplasticity and data-agnosticism. For example, you can wire visual input into auditory cortex and that auditory cortex will learn to see. [1]
For two, the similarity of Gabor filters learned automatically by statistical inference algorithms and the response properties of visual neurons. [2]
For three, the fantastic success, relative to other techniques at least, of such algorithms on natural language tasks such as machine translation.
Even its deficits are suggestive, such as the fact that the neocortex, like most statistical inference algorithms, can't easily do one-shot learning — for that you need a hippocampus.
Present-day machines obviously lack many abilities which humans have. Until we have a fully-functional AGI, it's hard to know for sure that there isn't a major missing piece which is nothing like the algorithms now under development. However, I'm not aware of any other algorithmic hypotheses of how the mind works which seem to so naturally fit the neuroscientific facts.
[1] Sur et al (1988). Experimentally Induced Visual Projections into Auditory Thalamus and Cortex. https://doi.org/10.1126/science.2462279
[2] Olshausen and Field (1996). Emergence of simple-cell receptive field properties by learning a sparse code for natural images. https://doi.org/10.1038/381607a0
The meaning of human language is to do with having an effect on the world, having a perception of the world, and having motivations related to the world - such as conquest, subjugation, hearing lamentations etc.
It seems possible to have the information content of language without the above extension in the world. Information is "just" combinations of patterns.
Humans need surprisingly little data to learn language. If we can discover our internal presumptions (a priori probabilities), AI mightn't need so much data either.
To use a crude metaphor, language could be just an API schema. Is there content accessible via the schema that is more meaningful than the schema itself? For APIs, yes. I think so for language too.
Not to mention the possibility of side channels. That, at the same moment any human reads any text, they receive extra information to help them understand that information.
I think what it means to understand language is to be able to generate and react to language to accomplish a wide range of goals.
Neural nets are clearly not capable of understanding language as well as humans by this definition, but they're somewhere on the spectrum between rocks and humans, and getting closer to the human side every day.
I can't help but think that arguments that algorithms simply don't or can't understand language at all are appealing to some kind of Cartesian dualism that separates entities that have mind from those are merely mechanical. If that's your metaphysics, then you're going to continue to find that no particular mechanical system really understands language, all the way up to (and maybe beyond) the point where mechanical systems can use and react to language in fully all situations that humans can.
Sorry but can you be a bit more specific here? How exactly has NLU split from NLP?
And yet, if you ask them what the meaning of a given text is, they’re getting disturbingly close.
What they don’t have yet is self-awareness, a model of their own repeated interfacing with the world. And an understanding of language in that context.
But it’s coming.
Humans have always believed that their current state of technology reflects how the brain works — clay pots and waterworks, electric circuits, cogs, etc. Just because there’s an “I” in AI and a “neural” in front of “network” doesn’t mean you’re not just mimicking something as opposed to replicating how it actually works…
I am drawing conclusions here but it seems to me you're implying that simulating the human mind is unattainable, while I'm arguing that you're possibly right, but that in the process we'll find something that could be even better (and/or cause our own doom).
"If I'd known what it can cost to ..., I'd have tried harder to ... it."
Context is everything. There are a lot of ways to finish the sentence. If it appears in an unfamiliar context, the right answer is NP hard.
It seems obvious that computers can emulate how humans process information. That's the tradition of the word "computer", which used to be a human occupation. The big question is regarding the sources of information. Humans seem to have access to more information than artificial computers.
Advances in AI are of two overlapping categories 1) catching up to human information processing, 2) capturing information from humans / from human information sources (many of which are not sensory).
The big question is whether artificial computers can 1) process information the same way we can and 2) achieve the same access to information that we have. We are making a lot of progress in 1), significantly less progress in 2).