Nosimpler

Nature, Published online: 15 May 2019; doi:10.1038/d41586-019-01461-7

Prostate cancer contains nerve cells that are linked to disease progression, but their source was unknown. A mouse study reveals that cells from the brain invade prostate tumours and give rise to this nerve-cell population.

Modern recording techniques enable large-scale measurements of neural activity in a variety of model organisms. The dynamics of neural activity shed light on how organisms process sensory information and generate motor behavior. Here, we study these dynamics using optical recordings of neural activity in the nematode C. elegans. To understand these data, we develop state space models that decompose neural time-series into segments with simple, linear dynamics. We incorporate these models into a hierarchical framework that combines partial recordings from many worms to learn shared structure, while still allowing for individual variability. This framework reveals latent states of population neural activity, along with the discrete behavioral states that govern dynamics in this state space. We find stochastic transition patterns between discrete states and see that transition probabilities are determined by both current brain activity and sensory cues. Our methods automatically recover transition times that closely match manual labels of different behaviors, such as forward crawling, reversals, and turns. Finally, the resulting model can simulate neural data, faithfully capturing salient patterns of whole brain dynamics seen in real data.

Focusing attention on one speaker on the background of other irrelevant speech can be a challenging feat. A longstanding question in attention research is whether and how frequently individuals shift their attention towards task-irrelevant speech, arguably leading to occasional detection of words in a so-called unattended message. However, this has been difficult to gauge empirically, particularly when participants attend to continuous natural speech, due to the lack of appropriate metrics for detecting shifts in internal attention. Here we introduce a new experimental platform for studying the dynamic deployment of attention among concurrent speakers, utilizing a unique combination of Virtual Reality and Eye-Tracking technology. We created a Virtual Cafe in which participants sit across from and attend to the narrative of a target speaker. We manipulate the number and location of distractor speakers, manifest as additional patrons throughout the Virtual Cafe. By monitoring participants eye-gaze dynamics, we studied the patterns of overt shifts of attention among the concurrent speakers as well as the consequences of these shifts on speech comprehension. Our results reveal important individual differences in the gaze-pattern displayed during selective attention to speech. While some participants stayed fixated on a target speaker throughout the entire experiment, approximately 30% of participants frequently shifted their gaze toward distractor speakers or other locations in the environment, regardless of the severity of audiovisual distraction. Critically, the tendency for frequent gaze-shifts negatively impacted comprehension of the target speaker. We also found that gaze-shifts occurred primarily during gaps in the acoustic input, suggesting they are prompted by momentary unmasking of the competing audio, in line with glimpsing theories of processing speech in noise. These results open a new window into understanding the dynamics of attention as they wax and wane over time, and the different listening patterns employed for dealing with the influx of sensory input in multisensory environments. Moreover, the novel approach developed here for tracking the locus of momentary attention in a naturalistic virtual-reality environment holds high promise for extending the study of human behavior and cognition and bridging the gap between the laboratory and real-life.

From the always-interesting Atlas Obscura:

Massimo Rapella, a 48-year-old chicken farmer from northern Italy, is helping chickens rediscover their wild side. Since 2009, Rapella and his wife Elisabetta have been keeping an estimated 2,100 hens in a patch of pristine Alpine forest near Sondrio, in the heart of the Valtellina valley...

Shortly after relocating, Rapella and his wife started keeping a few chickens to provide eggs for their own consumption. But soon enough they noticed some unexpected behavior from their flock. “Our chickens liked roaming around the nearby woods,” Rapella explains. “So I encouraged them to venture out and lay eggs in the wild.”

A few months later, Rapella saw that the birds looked healthier—with shiny feathers and bright-colored wattles—and that their eggs had a fuller taste. “I started wondering if I could take on more chickens and create an ‘Alpine egg’ to sell in local markets,” he says. Today, he sells his uovo di selva, or egg of the woods, to about 400 direct consumers and 40 restaurants...

Most domestic chickens today would not find themselves at home in a forest: at least, not immediately. “The first large batch of chickens I took in looked very lost,” Rapella says. “They had never seen a tree nor a bug in their life, and they were scared of snow.”..

“White birds really stand out to predators,” Clauer says. Rapella keeps two different breeds of chicken: Hy-Line brown hens and the easy-to-spot white Leghorns. While he once lost the occasional chicken, now he relies on a double fence and two trained Maremma sheepdogs to keep badgers, martens (a weasel-like carnivore), foxes, and buzzards at bay.

Rapella’s chickens lay eggs almost every day, like any domesticated chicken, but they do so in the woods. “They like natural nests offered by tree roots or branches,” he says. “Usually when you spot a cranny with some leaves, you know there could be eggs.” Once a hen finds her favorite nesting spot, she goes back to it for each subsequent laying, making Rapella’s egg-hunting easier. Together with two employees, he gathers an estimated 1,000 eggs every morning.

His uovo di selva tastes like egg, but concentrated. There’s more flavor to it, and also more protein, due to the bug-filled diet of the chickens. As a result, when chefs whip the whites from Rapella’s protein-rich eggs, they get three times the volume. The egg yolk can even change with the seasons...

More at the link.

Via Digg

An unarmed mentally ill man was shot at by police.  Some of the bullets missed and struck bystanders in the street.  The DA decided to charge the mentally ill man with first-degree assault because of the injuries to the bystanders.

Police later said that after a cat-and-mouse game, Broadnax had reached into his pants pocket and removed an object, briefly “pointing” it at the responding officers. They thought it was a gun, and fired three rounds. After the shots, and a tussle with cops, Broadnax was hit with a Taser and arrested.

When the dust settled, a few things became clear. First, Broadnax was unarmed; the object police had thought was a gun was, in fact, a wallet. Second, Broadnax seemed to be in the midst of a mental health crisis. He told investigators immediately after his arrest that he was having auditory hallucinations, hearing the “voices of his dead relatives.”

The third fact that emerged was that the officers who opened fire had missed their target. Broadnax was unscathed. The bullets meant for him had instead struck Khoshakhlagh and a second victim, 59-year-old Theodora Ray, who was getting dinner at a food cart on 42nd Street, as she did almost every night...

Then, near the end of October, Khoshakhlagh’s attorney got a phone call from the office of Manhattan District Attorney Cyrus Vance Jr. Prosecutors had changed their minds about the case: They now planned to prosecute Broadnax for the bullet wounds sustained by Khoshakhlagh and Ray. Under an unusual theory of law, the prosecution claimed Broadnax’s actions that night in Times Square were so irresponsible, it was as if he himself had pulled the trigger.

The top charge alone — assault in the first degree — was enough to land him in prison for the next twenty-five years...

As Binder says, when police are involved in incidents like this — high-profile and potentially embarrassing mishaps — there can be pressure to ensure that someone, anyone, catches an indictment. “It’s not surprising that Mr. Broadnax was charged,” Binder says. “This is something prosecutors do when police behave irresponsibly.

More details in a longread in The Village Voice.


Actually in The Onion:

Scientists Announce Discovery Of Dry Ice On Mars Means Planet May One Day Be Suitable For Halloween Party

Tourist In White House Gift Shop Browses Rack Of Security Clearances

 Purdue Pharma Reports Opioid Deaths Falling Short Of Quarterly Goals

When monkeys observe and learn from each other’s choices, neurons in the amygdala spontaneously encode decision computations to simulate the social partner’s choices.

“A measure of the flexibility of excellence is that it allows the inclusion of reputation as one category among others in a ranking which is in fact definitive of reputation. The metalepsis that allows reputation to be 20 percent of itself is permitted by the intense flexibility of excellence; it allows a category mistake to masquerade as scientific objectivity.”

Bill Readings, The University in Ruins.

 

Check out the video of Christian Williams’’s talk in the Applied Category Theory Seminar here at U. C. Riverside. It was nicely edited by Paola Fernandez and uploaded by Joe Moeller.

Abstract. Historically, code represents a sequence of instructions for a single machine. Each computer is its own world, and only interacts with others by sending and receiving data through external ports. As society becomes more interconnected, this paradigm becomes more inadequate – these virtually isolated nodes tend to form networks of great bottleneck and opacity. Communication is a fundamental and integral part of computing, and needs to be incorporated in the theory of computation.

To describe systems of interacting agents with dynamic interconnection, in 1980 Robin Milner invented the pi calculus: a formal language in which a term represents an open, evolving system of processes (or agents) which communicate over names (or channels). Because a computer is itself such a system, the pi calculus can be seen as a generalization of traditional computing languages; there is an embedding of lambda into pi – but there is an important change in focus: programming is less like controlling a machine and more like designing an ecosystem of autonomous organisms.

We review the basics of the pi calculus, and explore a variety of examples which demonstrate this new approach to programming. We will discuss some of the history of these ideas, called “process algebra”, and see exciting modern applications in blockchain and biology.

“… as we seriously address the problem of modelling mobile communicating systems we get a sense of completing a model which was previously incomplete; for we can now begin to describe what goes on outside a computer in the same terms as what goes on inside – i.e. in terms of interaction. Turning this observation inside-out, we may say that we inhabit a global computer, an informatic world which demands to be understood just as fundamentally as physicists understand the material world.” — Robin Milner

The talks slides are here.

Reading material:

• Robin Milner, The polyadic pi calculus: a tutorial.

• Robin Milner, Communicating and Mobile Systems.

• Joachim Parrow, An introduction to the pi calculus.

Last week, we had Peter Scholze give an interesting distinguished lecture series here at UCLA on “Prismatic Cohomology”, which is a new type of cohomology theory worked out by Scholze and Bhargav Bhatt. (Video of the talks will be available shortly; for now we have some notes taken by two note–takers in the audience on that web page.) My understanding of this (speaking as someone that is rather far removed from this area) is that it is progress towards the “motivic” dream of being able to define cohomology for varieties (or similar objects) defined over arbitrary commutative rings , and with coefficients in another arbitrary commutative ring . Currently, we have various flavours of cohomology that only work for certain types of domain rings and coefficient rings :

• Singular cohomology, which roughly speaking works when the domain ring is a characteristic zero field such as or , but can allow for arbitrary coefficients ;
• de Rham cohomology, which roughly speaking works as long as the coefficient ring is the same as the domain ring (or a homomorphic image thereof), as one can only talk about -valued differential forms if the underlying space is also defined over ;
• -adic cohomology, which is a remarkably powerful application of étale cohomology, but only works well when the coefficient ring is localised around a prime that is different from the characteristic of the domain ring ; and
• Crystalline cohomology, in which the domain ring is a field of some finite characteristic , but the coefficient ring can be a slight deformation of , such as the ring of Witt vectors of .

There are various relationships between the cohomology theories, for instance de Rham cohomology coincides with singular cohomology for smooth varieties in the limiting case . The following picture Scholze drew in his first lecture captures these sorts of relationships nicely:

The new prismatic cohomology of Bhatt and Scholze unifies many of these cohomologies in the “neighbourhood” of the point in the above diagram, in which the domain ring and the coefficient ring are both thought of as being “close to characteristic ” in some sense, so that the dilates of these rings is either zero, or “small”. For instance, the -adic ring is technically of characteristic , but is a “small” ideal of (it consists of those elements of of -adic valuation at most ), so one can think of as being “close to characteristic ” in some sense. Scholze drew a “zoomed in” version of the previous diagram to informally describe the types of rings for which prismatic cohomology is effective:

To define prismatic cohomology rings one needs a “prism”: a ring homomorphism from to equipped with a “Frobenius-like” endomorphism on obeying some axioms. By tuning these homomorphisms one can recover existing cohomology theories like crystalline or de Rham cohomology as special cases of prismatic cohomology. These specialisations are analogous to how a prism splits white light into various individual colours, giving rise to the terminology “prismatic”, and depicted by this further diagram of Scholze:

(And yes, Peter confirmed that he and Bhargav were inspired by the Dark Side of the Moon album cover in selecting the terminology.)

There was an abstract definition of prismatic cohomology (as being the essentially unique cohomology arising from prisms that obeyed certain natural axioms), but there was also a more concrete way to view them in terms of coordinates, as a “-deformation” of de Rham cohomology. Whereas in de Rham cohomology one worked with derivative operators that for instance applied to monomials by the usual formula

prismatic cohomology in coordinates can be computed using a “-derivative” operator that for instance applies to monomials by the formula

where

is the “-analogue” of (a polynomial in that equals in the limit ). (The -analogues become more complicated for more general forms than these.) In this more concrete setting, the fact that prismatic cohomology is independent of the choice of coordinates apparently becomes quite a non-trivial theorem.

 

Author(s): Eric Chitambar and Gilad Gour

This review introduces a new development in theoretical quantum physics, the “resource-theoretic” point of view. The approach aims to be closely linked to experiment, and to state exactly what result you can hope to achieve for what expenditure of effort in the laboratory. This development is an extension of the principles of thermodynamics to quantum problems; but there are resources that would never have been considered previously in thermodynamics, such as shared knowledge of a frame of reference. Many additional examples and new quantifications of resources are provided.

[Rev. Mod. Phys. 91, 025001] Published Thu Apr 04, 2019

I just finished reading The Shape of a Life, which is the great geometer Shing-Tung Yau’s autobiography, co-authored with Steve Nadis. It’s quite fascinating, and an essential read for anyone interested in the history of modern mathematics. Yau has been for a long time a central figure in the field of geometric analysis, so this is in some ways as much an autobiography of the subject as well as of the man.

Back in 2010 I wrote here about an earlier volume by Yau and Nadis, The Shape of Inner Space. What I really liked about that book (and discussed in some detail there) was the autobiographical material about Yau. Much of the book though was devoted to topics like string theory attempts to get physics out of Calabi-Yaus, with a discussion that was detailed and accurate, but to my mind often not of great interest (since these attempts don’t work…).

The new book seems to have been written specifically to appeal to me, greatly expanding the autobiographical material of the earlier book, while limiting the discussion of dubious speculative physics. There is still a fair amount about physics, but this time more focused on another of Yau’s interests, the mathematical theory of general relativity.

The book begins with the story of Yau’s early years in Hong Kong, how he managed to survive an impoverished childhood, avoid becoming a duck farmer, and ultimately find a way to get to the US and graduate study in mathematics at Berkeley. It’s a compelling story of that period and those places. It’s also about the best example I can think of to show how bringing someone with undeveloped talent into the environment of a first-rate research university can change their life, liberating them to accomplish great things, with dramatic impact on their intellectual development as well as that of a whole field.

Yau has always had a deep interest in the history of mathematics, and the story he tells of his intellectual development explains in detail how his own work and ideas grew out of earlier strands of thought. Even as a graduate student, he had started to develop the point of view that has been so fruitful in geometric analysis, using the study of non-linear partial differential equations to prove theorems about geometry and topology. Besides his proof of the Calabi conjecture, this ultimately led to the proof of the Poincare conjecture, a story Yau explains in detail.

Over the years Yau has been involved in various controversies over priority for mathematical results. In this book he doesn’t shy away from discussing these, but generally gives a measured explanation of his point of view on what happened. There’s also a fair number of often amusing stories about mathematicians and the math community that liven up the history. For one sort of example, there are Yau’s descriptions of his culture clash with the long-haired, pot-smoking Berkeley of 1969. For another, here’s a story about Richard Hamilton (of whom Yau has a very high opinion) and his 1982 lectures at the IAS:

Hamilton, who had come from Cornell, stayed for a week in an IAS apartment. At the end of his stay, the chief math secretary was livid because Hamilton had made a huge mess of the apartment, and it took a long time to clean up the place. On the other hand, he had given some wonderful talks, and collaborations between Hamilton, my students, and me picked up from that time forward. So, on balance, his visit would have to be called a great success. Hamilton may have posed some challenges to the cleaning and janitorial staff, but he had posed even more consequential challenges to the mathematics community, some of which were taken up by members of my group.

Yau is generally considered a major figure not just for his research, but also as a politician of the mathematics community, deeply involved for many years in efforts to build or expand research centers, here and in China. A recent example is the creation of the CMSA at Harvard. He has a lot to say about the stories of these efforts, and he definitely does not do so with the style of the politician careful to offend no one. In this book you get Yau’s honest, unvarnished version of what happened, as well as his analysis of some general problems, and I won’t be surprised if some people take offense at this material.

One thing there’s perhaps a bit too much of in the book are the references to his conflicts with his advisor Shiing-Shen Chern (which I’d somehow never heard about before). A major touching theme though throughout the book is that of fathers, sons and traditions of filial piety. There’s a lot about Yau’s father (who Yau very much looked up to) and quite a bit about his sons. On the mathematical side, there’s a lot about his numerous students, many of whom have gone on to important academic careers. As his academic father, Chern also fits into this theme, although not so felicitously. At the end of the book, Yau looks forward to his own future as, like Chern before him, the grand old man of the field. He’s planning more teaching and less research, and taking pleasure in his mathematical legacy and progeny.

I noticed today that Cambridge University Press has recently published Why Trust a Theory?, a volume of articles based on a December 2015 conference held in Munich. The book is available online here (if your university is paying for it…), and preprint versions of many of the contributions are on the arXiv.

The conference had its origins in a piece published a year earlier in Nature by George Ellis and Joe Silk, entitled Scientific method: Defend the integrity of physics. Ellis and Silk made a forceful case that widely advertised but inherently untestable string theory and multiverse research does damage to the public understanding of science and is a threat to the credibility of science at a time it is under attack. The piece suggested:

A conference should be convened next year to take the first steps. People from both sides of the testability debate must be involved.

Looking through the proceedings volume, there’s lots of abstract discussion of philosophy of science and some diversity of points of view on the multiverse. When it comes to string theory though, the organizers interpreted “people on both sides” to mean bringing in one person willing to point out that there is a problem with string theory, and an army of string theorists to defend the theory. On the issue of the problems of string theory, the volume contains nearly 100 pages of pro-string theory hype, from Polchinski (two contributions), Silverstein, Kane and Quevedo. As usual with Kane, there’s a string theory “prediction” of the gluino mass (1.5 TeV +/- 10-15%) which has already been falsified. All I could find on the side of substantive criticism of string theory was in Carlo Rovelli’s contribution (preprint version here), and mainly in a single paragraph:

String theory is a living proof of the dangers of excessive reliance on non-empirical arguments. It raised great expectations thirty years ago, promising to compute all the parameters of the Standard Model from first principles, to derive from first principles its symmetry group SU(3)×SU(2)×U(1) and the existence of its three families of elementary particles, to predict the sign and the value of the cosmological constant, to predict novel observable physics, to understand the ultimate fate of black holes, and to offer a unique, well-founded unified theory of everything. Nothing of this has come true. String theorists, instead, have predicted a negative cosmological constant, deviations from Newton’s 1/r^2 law at sub-millimeters scale, black holes at the European Organization for Nuclear Research(CERN), low-energy super-symmetric particles, and more. All this was false. Still, Joe Polchinski, a prominent string theorist, writes [7] that he evaluates the Bayesian probability of string to be correct at 98.5% (!). This is clearly nonsense.

I won’t spend more time here discussing the conference and the articles in this volume, mainly because I’ve already written a lot about this in previous posts. For a contemporaneous discussion of the conference and Polchinski’s String Theory to the Rescue paper, see here and here. There are also interesting blog posts about the conference from Massimo Pigliucci, see here, here and here, and a Quanta piece by Natalie Wolchover here. For a discussion of Sean Carroll’s Beyond Falsifiability contribution, see here (and discussion here and here). For a discussion of Eva Silverstein’s contribution, see here.

Update: A few more links to material about the Munich conference: Jim Baggott here and here, Andrew Gelman here, Davide Castelvecchi here, and the conference website (with videos) here.

Update: Looking at the Preface, I notice that the editors claim:

Additional contributions were solicited by the editors with the aim of ensuring as full and balanced presentation as possible of the various positions in the debate.

With regards to string theory, the one additional contribution in the volume is from string theorist Eva Silverstein, so evidently the editors felt that balance required yet more on the pro-string theory side….

Update: I mischaracterized Polchinski’s calculation of the probability that string theory is correct as 98.5%. More accurately, he claims that the probability is “over 3 sigma” (i.e. over 99.73%).

Update: I finally got around to watching the videos of the panel discussions at the workshop (all videos available here). What most struck me about these discussions was the heavily dominant role of David Gross, who was on two of three panels, participating from the audience in the third. On the panels he was on, Gross was speaking far more than anyone else, and rarely if at all would anyone disagree with him. Gross’s point of view is that there is a testability problem with the multiverse, but all is well with string theory (although probably not at Polchinski’s “over 99.73% sure to be true” level). He’s a powerful intellect and a forceful speaker, so it’s not surprising that no one would take him on. But on the topic of string theory I think there are very serious problems with many of the claims he makes (for his arguments of 15 years ago, see the first substantive post of this blog), and the organizers should have found someone willing to challenge him on those.

In Part 1 of this series, I gave an abstract description of one of the main problems in Machine Learning, the Generalization Problem, in which one uses the values of a function at a finite number of points to infer the entire function. The typical approach to this problem is to choose a finite-dimensional subset of the space of all possible functions, then choose the function from this family that minimizes something called cost function, defined by how accurate each function is on the sampled points. In this post, I will describe how the regression example from the last post generalizes to a family of models called Neural Networks, then describe how I recently used some fairly basic topology to demonstrate restrictions on the types of functions certain neural networks can produce.

First lets recall the setup: We have a finite set of points and we want to find a function such that for each , the distance from to in is “as small as possible”. We do this by choosing a map , then picking the point in whose image minimizes a previously chosen cost function. In the last post, we also had a second set that we used to evaluate how well we did, but we won’t need that in this post.

The basic example of this is linear regression where , so and are both one-dimensional, and is spanned by the variables . This can be generalized to higher-dimensional and by replacing the scalars by a matrix and a vector, respectively. This is higher-dimensional linear regression.

Neural networks define a very general framework for defining other families of functions. One goal of this post is to describe a large (but not complete) chunk of this framework. But before we do that, we need to go from linear regression to logistic regression.

Logistic regression is used for prediction problems where rather than predicting a value associated with a data point, we want to predict the probability that a data point is in a given class or meets some condition. So we want a function whose range is the interval rather than all real numbers.

Logistic regression accomplishes this by composing the family of functions from linear regression with the logistic function a scaled and translated version of hyperbolic tangent that maps the line to the interval . So if is one-dimensional, then we will still have but a point will map to the function .

In the training set for a logistic regression problem, the -value of each datapoint will be either 0 or 1. The cost function for each data point is the negative log of the difference between the predicted and actual values. This logarithm is not arbitrary – it’s the result of calculating the likelihood (in the statistical sense) of the function (interpreted as a probability distribution) and applying a logarithm to turn the multiplication in the definition of likelihood into the addition required for the cost function. Details left to the reader.

As with linear regression, we can increase the dimension of or in logistic regression. For , it’s simply a matter of changing to a vector, since we can still apply the logistic function to the one-dimensional output. To increase the dimension of , we further promote to a matrix and to a vector, then apply the logistic function independently to each output dimension. We can interpret this as each dimension of predicting a different class/condition. So logistic regression is not basis-independent in the way that linear regression is.

In the realm of machine learning, the logistic function in logistic regression is called an activation function. The idea is that each dimension represents a neuron and the output value represents whether or not the neuron is firing. So the activation function determines how the linear combination of values feeding into the neuron determines whether or not it should fire. (In theory, the output of the activation function should be a boolean, but then we couldn’t do gradient descent.) Another popular activation function is the Rectified Linear Unit (ReLU) .

Now that we know what logistic regression is and what activation functions are, we can define a large family of neural networks by simply composing a chain of (higher-dimensional) logistic regressions. Each regression is called a layer. The outputs dimensions have to match up to the input dimensions, and the overall parameter space of the composition is the direct product of the parameter spaces of the layers. We can use the same activation functions between layers, or different ones.

That’s it.

The reason this is called an (artificial) neural network is because we can think of the output dimensions of each layer as being neurons that are connected to neurons in the next layer. The linear part of each logistic regression defines the input to each neuron as a weighted sum of the previous neurons’ outputs. For this reason, the matrix is often called a weight matrix and the values are called weights. Every layer except the last one is called a hidden layer.

Note that the weights, which make up the parameters of the function family, only modify the linear steps in the neural network, while the activation functions remain fixed. However, without the non-linear activation functions the neural network would just be a composition of linear functions which would just produce a single linear function.

As I noted, this is a large family of neural networks, but the concept is much more general. There are neural networks where there are connections between non-consecutive layers, or where the neurons aren’t in layers at all. There are networks where certain weights are constrained to be the same as each other (such as convolutional neural networks) and networks that take a sequence of inputs (Recurrent Neural Networks and Long Short-Term Memory Machines).

But the family I defined above is where one typically starts, and that’s what I’m going to focus on for the last few paragraphs of this blog post.

As with linear or logistic regression, a neural network defines a finite-dimensional family of functions. But unlike with regression, there isn’t an easy way to characterize these functions. It was proved in 1989, by three independent groups, that if you have one hidden layer, but allow it to have arbitrarily high dimension, you can approximate any continuous function to an arbitrarily small epsilon on any compact subset.

However, there’s an open question of whether something similar is true if you restrict the dimension but allow arbitrarily many hidden layers. The preprint I mentioned in the last post proves that for neural networks with a one-dimensional output, if the hidden layers are only allowed to have dimension less than or equal to the input dimension, there are certain functions that can’t be approximated, regardless of the number of layers. In particular, every component of every level set in such a function must be unbounded.

The main observation in the proof is that for a neural network with a one-dimensional output, if the hidden layers are all the same dimension, the weight matrices are all non-singular and the activation function is one-to-one, then the composition of all the functions up to right before the last linear function will be one-to-one too. That last linear function is just a projection onto the line, so the preimage of any point into the last hidden layer is a hyperplane. Since the activation functions may not be onto, the preimage of this hyperplane in the one-to-one map may not be a topological hyperplane, but it will be an unbounded subset of the domain.

But the Theorem also applies to networks with lower-dimensional hidden layers, and doesn’t make assumptions about the weight matrix. That’s because these functions are all in the closure of the previous set so there’s a limit argument to show that their level sets also have to be unbounded. In fact you can also apply the limit argument if the activation function is a uniform limit of one-to-one functions, like the ReLU. The proof is the type of argument you might see in an undergraduate topology class.

So as with many examples of applied math, the mathematics of this result isn’t particularly complex. What makes it interesting is the connection to ideas that are studied elsewhere. It also suggests a new set of problems that could lead to more interesting math, namely the question of how to characterize finite-dimensional families of functions such as those defined by neural networks.

Yay! David Spivak and Brendan Fong are teaching a course on applied category theory based on their book, and the lectures are on YouTube! Here are the first two videos:

Their book is free here:

• Brendan Fong and David Spivak, Seven Sketches in Compositionality: An Invitation to Applied Category Theory.

If you’re in Boston you can actually go to the course. It’s at MIT January 14 – Feb 1, Monday-Friday, 14:00-15:00 in room 4-237.

They taught it last year too, and last year’s YouTube videos are on the same YouTube channel.

Also, I taught a course based on the first 4 chapters of their book, and you can read my “lectures”, see discussions and do problems here:

• Applied category theory course.

So, there’s no excuse not to start applying category theory in your everday life!

According to Veda, the game was created by the 13th century poet saint Gyandev.

In the original game square 12 was faith, 51 was Reliability, 57 was Generosity, 76 was Knowledge, and 78 was Asceticism. These were the squares where the ladder was found. Square 41 was for Disobedience, 44 for Arrogance, 49 for Vulgarity, 52 for Theft, 58 for Lying, 62 for Drunkenness, 69 for Debt, 84 for Anger, 92 for Greed, 95 for Pride, 73 for Murder and 99 for Lust. These were the squares where the snake was found. The Square 100 represented Nirvana or Moksha.

More info:

Snakes and Ladders originated in India as part of a family of dice board games that included Gyan chauper and pachisi (present-day Ludo and Parcheesi). The game made its way to England and was sold as "Snakes and Ladders", then the basic concept was introduced in the United States as Chutes and Ladders by game pioneer Milton Bradley in 1943.

The game was popular in ancient India by the name Moksha Patam. It was also associated with traditional Hindu philosophy contrasting karma and kama, or destiny and desire. It emphasized destiny, as opposed to games such as pachisi, which focused on life as a mixture of skill (free will) and luck. The underlying ideals of the game inspired a version introduced in Victorian England in 1892. The game has also been interpreted and used as a tool for teaching the effects of good deeds versus bad. The board was covered with symbolic images, the top featuring gods, angels, and majestic beings, while the rest of the board was covered with pictures of animals, flowers and people.

The ladders represented virtues such as generosity, faith, and humility, while the snakes represented vices such as lust, anger, murder, and theft. The morality lesson of the game was that a person can attain salvation (Moksha) through doing good, whereas by doing evil one will inherit rebirth to lower forms of life. The number of ladders was less than the number of snakes as a reminder that a path of good is much more difficult to tread than a path of sins. Presumably, reaching the last square (number 100) represented the attainment of Moksha (spiritual liberation).

When the game was brought to England, the Indian virtues and vices were replaced by English ones in hopes of better reflecting Victorian doctrines of morality. Squares of Fulfillment, Grace and Success were accessible by ladders of Thrift, Penitence and Industry and snakes of Indulgence, Disobedience and Indolence caused one to end up in Illness, Disgrace and Poverty. While the Indian version of the game had snakes outnumbering ladders, the English counterpart was more forgiving as it contained each in the same amount. This concept of equality signifies the cultural ideal that for every sin one commits, there exists another chance at redemption.

Interesting that success in the game as originally designed depended entirely on luck (roll of dice) with no apparent skills or strategy involved; perhaps that's part of the karma lesson.  AFAIK, the American version didn't incorporate any virtues or sins - it was more like random good and bad luck.  I may be misremembering.  But I certainly didn't know it was an ancient game.

What if we abandoned photosynthesis as the means of producing food, and released most of the world’s surface from agriculture?

By George Monbiot, published in the Guardian 31^st October 2018

 

It’s not about “them”, it’s about us. The horrific rate of biological annihilation reported this week – 60% of the Earth’s vertebrate wildlife gone since 1970 – is driven primarily by the food industry. Farming and fishing are the major causes of the collapse of both marine and terrestrial ecosystems. Meat – consumed in greater quantities by the rich than by the poor – is the strongest cause of all. We might shake our heads in horror at the clearance of forests, the drainage of wetlands, the slaughter of predators and the massacre of sharks and turtles by fishing fleets, but it is done at our behest.

As the Guardian’s recent report from Argentina reveals, the huge forests of the Gran Chaco are heading towards extermination, as they are replaced by deserts of soya beans, almost all of which are used to produce animal feed, particularly for Europe. With Jair Bolsonaro in power in Brazil, deforestation in the Amazon is likely to accelerate, much of it driven by the beef lobby that helped bring him to power. The great forests of Indonesia and West Papua are being felled and burnt for oil palm at devastating speed.

The most important environmental action we can take is to reduce the area of land and sea used by farming and fishing. This means, above all, switching to a plant-based diet: research published in the journal Science shows that cutting out animal products would reduce the global requirement for farmland by 76%. It would also give us a fair chance of feeding the world. Grazing is no answer to the ecocide caused by grain-fed livestock: it is an astonishingly wasteful use of vast tracts of land that would otherwise support wildlife and wild ecosystems.

The same action is essential to prevent climate breakdown. Because governments, bowing to the demands of capital, have left it so late, it is almost impossible to see how we can stop more than 1.5° of global warming without drawing carbon dioxide out of the atmosphere. The only way of doing it that has been demonstrated at scale is to allow trees to return to deforested land.

But could we go beyond even a plant-based diet? Could we go beyond agriculture itself? What if, instead of producing food from soil, we were to produce it from air? What if, instead of basing our nutrition on photosynthesis, we were to use electricity, to fuel a process whose conversion of sunlight into food is ten times more efficient?

This sounds like science fiction, but it is already approaching commercialisation. For the past year, a group of Finnish researchers has been producing food without either animals or plants. Their only ingredients are hydrogen-oxidising bacteria, electricity from solar panels, a small amount of water, carbon dioxide drawn from the air, nitrogen and trace quantities of minerals such as calcium, sodium, potassium and zinc. The food they have produced is 50 to 60% protein, the rest is carbohydrate and fat. They have started a company (Solar Foods), which seeks to open its first factory in 2021. This week it was selected as an incubation project by the European Space Agency.

They use electricity from solar panels to electrolyse water, producing hydrogen, that feeds bacteria (which turn it back into water). Unlike other forms of microbial protein (such as Quorn), it requires no carbohydrate feedstock – in other words, no plants.

Perhaps you are horrified by this prospect. Certainly, there’s nothing beautiful about it. It would be hard to write a pastoral poem about bacteria grazing on hydrogen. But this is part of the problem. We have allowed a mythical aesthetic to blind us to the ugly realities of industrial agriculture. Instilled with an image of farming that begins in infancy, as about half the books for very small children involve a rosy-cheeked farmer with one cow, one horse, one pig and one chicken, living in bucolic harmony, we fail to see the amazing cruelty of large-scale animal farming, the blood and gore, filth and pollution. We fail to apprehend the mass clearance of land required to feed us, the Insectageddon caused by pesticides, the drying up of rivers, the loss of soil, the reduction of the magnificent diversity of life on Earth to a homogenous grey waste.

The compound the Finnish researchers have produced from air, water and electricity is most likely to be used as a bulk ingredient in processed food. But (though this goes well beyond the company’s current plans) is there any reason why, with modifications of the process, it could not start to deliver the proteins required to make cultured meat, or the oils that could render palm plantations redundant? Is there any reason why it should not eventually replace much of what we eat?

According to the researchers’ estimates, 20,000 times less land is required for their factories than to produce the same amount of food by growing soya. Cultivating all the protein the world now eats with their technique would require an area smaller than Ohio. The best places to do it are deserts, where solar energy is most abundant. When electricity can be generated at €15 per megawatt hour (a few years hence), their process becomes cost-competitive with the cheapest source of soya.

Could a similar technique also be used to produce cellulose and lignin, eventually replacing the need for commercial forestry? Is there any inherent reason why the hydrogen pathway could not create as many products as photosynthesis does today? Could it help to change our entire relationship with the natural world, reducing our footprint to a fraction of its current size?

There are plenty of questions to be answered, plenty of possible hurdles and constraints. But think of the possibilities. Agricultural commodities, currently using almost all the Earth’s fertile land area, could be shrunk into a few small pockets of infertile land. The potential for ecological restoration is astonishing. The potential for feeding the world, a question that has literally been keeping me awake at night, is just as electrifying.

None of this means we can afford to relax and wait for an infant technology to save us. In the meantime, as urgent intermediate steps, we should switch to a plant-based diet and mobilise against the destruction of the living planet. You could start by joining the Extinction Rebellion that launches today [Wednesday].

But if this works, it could help, alongside political mobilisation, to change almost everything. Places which have become agricultural deserts, trashed by giant corporations, could be reforested, drawing carbon dioxide from the air on a vast scale. The ecosystems of land and sea could recover, not just in pockets but across great tracts of the planet. A new age of global hunger becomes less likely.

Crude and destructive technologies got us into this mess. Refined technologies can help get us out of it. The struggle to save every possible species and ecosystem from the current wave of destruction is worthwhile. One day, perhaps within our lifetimes, they could repopulate a thriving world.

www.monbiot.com

Reconstructing ancestral characters and traits along a phylogenetic tree is central to evolutionary biology. It is the key to understanding morphology changes among species, inferring ancestral biochemical properties of life, and recovering migration routes in phylogeography. The goal is twofold: to reconstruct the character state at the tree root (e.g. the region of origin of some species), and to understand the process of state changes along the tree (e.g. species flow between countries). Although each goal can be achieved with high accuracy individually, we use mathematics and simulations to demonstrate that it is generally impossible to accurately estimate both the root state and the rates of state changes along the tree branches from the observed data at the tips of the tree. This inherent Darwinian uncertainty principle concerning the simultaneous estimation of pattern and process governs ancestral reconstructions in biology. Increasing the number of tips improves the joint estimation accuracy for certain tree shapes that arise in evolutionary models, however, for other trees shapes it does not.

The action potential propagating in a nerve fibre generates accompanying mechanical and thermal effects. The whole signal is therefore an ensemble which includes primary and secondary components. The primary components of a signal are the action potential itself and longitudinal mechanical waves in axoplasm and surrounding biomembrane. These components are characterized by corresponding velocities. The secondary components of a signal are derived from primary components and include transverse displacement of a biomembrane and the temperature -- these have no independent velocities but have been measured in several experiments. A robust mathematical model is presented based on differential equations describing the signal primary components which are coupled into a system by coupling forces. The model includes also mathematical formulation for establishing the secondary components following the ideas from experimental studies.

We study whether a depth two neural network can learn another depth two network using gradient descent. Assuming a linear output node, we show that the question of whether gradient descent converges to the target function is equivalent to the following question in electrodynamics: Given $k$ fixed protons in $\mathbb{R}^d,$ and $k$ electrons, each moving due to the attractive force from the protons and repulsive force from the remaining electrons, whether at equilibrium all the electrons will be matched up with the protons, up to a permutation. Under the standard electrical force, this follows from the classic Earnshaw's theorem. In our setting, the force is determined by the activation function and the input distribution. Building on this equivalence, we prove the existence of an activation function such that gradient descent learns at least one of the hidden nodes in the target network. Iterating, we show that gradient descent can be used to learn the entire network one node at a time.

Non-nociceptive roles of opioids in the CNS: opioids’ effects on neurogenesis, learning, memory and affect

Non-nociceptive roles of opioids in the CNS: opioids’ effects on neurogenesis, learning, memory and affect, Published online: 05 December 2018; doi:10.1038/s41583-018-0092-2

Maladaptive modulation of learning, memory and affect by opioids is linked to dysfunctional neurogenesis. In this Review, Kibaly and colleagues discuss this link and how strategies that target neurogenesis to rescue opioid-dependent learning, memory and affect impairments constitute future directions for anti-addiction therapies.

Emergence of three-dimensional order and structure in growing biofilms

Emergence of three-dimensional order and structure in growing biofilms, Published online: 26 November 2018; doi:10.1038/s41567-018-0356-9

Single-cell tracking of up to 10,000 bacteria reveals the structure and dynamics of 3D biofilms—providing evidence to suggest that both local ordering and global biofilm architecture emerge from mechanical interactions.

Status epilepticus (SE) is defined as a state of unrelenting seizure activity. It is associated with a rapidly rising mortality rate, and thus constitutes a medical emergency. Benzodiazepines, which act as positive modulators of chloride (Cl-) permeable GABAA receptors, are indicated as the first line of treatment, but this is ineffective in many cases. We found that 48% of children presenting with SE were unresponsive to benzodiazepine treatment, and critically, that the duration of SE at the time of treatment is an important predictor of non-responsiveness. We therefore investigated the cellular mechanisms that underlie acquired benzodiazepine resistance, using rodent organotypic brain slices. Removing Mg2+ ions leads to an evolving pattern of epileptiform activity, and eventually to a persistent state of repetitive discharges that strongly resembles clinical EEG recordings of SE. We show that the persistent SE-like activity is associated with a reduction in GABAA receptor conductance and Cl- extrusion capability. We explored the effect on intraneuronal Cl- using both gramicidin, perforated-patch clamp recordings and Cl- imaging. This showed that during SE-like activity, reduced Cl- extrusion capacity was further exacerbated by activity-dependent Cl- loading, resulting in a persistently high intraneuronal Cl-. Consistent with these results, we found that optogenetic stimulation of GABAergic interneurons in the SE-like state, actually enhanced epileptiform activity in a GABAAR dependent manner. Together our findings describe a novel potential mechanism underlying benzodiazepine-resistant SE, with relevance to how this life-threatening condition should be managed in the clinic.

To improve their metabolic performance, cells have to realise good compromises between large production fluxes, low enzyme investments, and well-adapted metabolite levels. In models, this idea can be formulated in the form of optimality principles that trade a high metabolic benefit against low enzyme cost. However, different modelling approaches are often incompatible. I propose a unified theory, called metabolic economics, to bridge the gap between different optimality-based cell models that exploits hidden equivalences between these approaches. Metabolic economics introduces new variables on the network, called economic variables, which represent the cost and benefit of metabolites, fluxes, and enzymes, and can be defined by Lagrange multipliers, auxiliary variables that are commonly used to handle constraints in optimality problems. Metabolic economics translates optimality conditions into local balance equations between these variables. The economic potentials and loads describe the value of metabolite production and metabolite concentrations. As proxy variables, they can describe indirect fitness effects, arising elsewhere in the network, as if they arose locally in a reaction of interest. Here I derive these variables and their balance equations for three types of optimality problems: for direct optimisation of enzyme levels in kinetic models; for flux cost minimisation (FCM), a minimisation of enzyme cost, with flux and metabolite profiles as the variables to be optimised; and for optimal protein allocation in whole-cell models, where growth rate or other whole-cell objectives are maximised. The economic balance equations add a new layer of description to mechanistic models, a description in terms of beneficial cell functions and associated costs, and can seen as economic laws of metabolism. Metabolic economics provides concepts for comparing and combining metabolic optimality problems, employing different modelling paradigms or different levels of detail, which can be useful for semi-automatic or modular modelling.

I have predicted for a while that as the issue of climate change becomes ever more urgent, the public attitude regarding geoengineering will at some point undergo a phase transition. For a long time it seems the general attitude has been that deliberately interfering with the Earth’s climate on a large scale is “unthinkable”: beyond the pale. I predict that at some point this will flip and the general attitude will become: “how soon can we do it?”

The danger then is that we rush headlong into something untested that we’ll regret.

For a while I’ve been advocating research in geoengineering, to prevent a big mistake like this. Those who consider it “unthinkable” often object to such research, but I think preventing research is not a good long-term policy. I think it actually makes it more likely that at some point, when enough people become really desperate about climate change, we will do something rash without enough information about the possible effects.

Anyway, one can argue about this all day: I can see the arguments for both sides. But here is some news: scientists will soon study how calcium carbonate disperses when you dump a little into the atmosphere:

• First sun-dimming experiment will test a way to cool Earth, Nature, 27 November 2018.

It’s a good article—read it! Here’s the key idea:

If all goes as planned, the Harvard team will be the first in the world to move solar geoengineering out of the lab and into the stratosphere, with a project called the Stratospheric Controlled Perturbation Experiment (SCoPEx). The first phase — a US$3-million test involving two flights of a steerable balloon 20 kilometres above the southwest United States — could launch as early as the first half of 2019. Once in place, the experiment would release small plumes of calcium carbonate, each of around 100 grams, roughly equivalent to the amount found in an average bottle of off-the-shelf antacid. The balloon would then turn around to observe how the particles disperse.

The test itself is extremely modest. Dai, whose doctoral work over the past four years has involved building a tabletop device to simulate and measure chemical reactions in the stratosphere in advance of the experiment, does not stress about concerns over such research. “I’m studying a chemical substance,” she says. “It’s not like it’s a nuclear bomb.”

Nevertheless, the experiment will be the first to fly under the banner of solar geoengineering. And so it is under intense scrutiny, including from some environmental groups, who say such efforts are a dangerous distraction from addressing the only permanent solution to climate change: reducing greenhouse-gas emissions. The scientific outcome of SCoPEx doesn’t really matter, says Jim Thomas, co-executive director of the ETC Group, an environmental advocacy organization in Val-David, near Montreal, Canada, that opposes geoengineering: “This is as much an experiment in changing social norms and crossing a line as it is a science experiment.”

Aware of this attention, the team is moving slowly and is working to set up clear oversight for the experiment, in the form of an external advisory committee to review the project. Some say that such a framework, which could pave the way for future experiments, is even more important than the results of this one test. “SCoPEx is the first out of the gate, and it is triggering an important conversation about what independent guidance, advice and oversight should look like,” says Peter Frumhoff, chief climate scientist at the Union of Concerned Scientists in Cambridge, Massachusetts, and a member of an independent panel that has been charged with selecting the head of the advisory committee. “Getting it done right is far more important than getting it done quickly.”

For more on SCoPEx, including a FAQ, go here:

• Stratospheric Controlled Perturbation Experiment (SCoPEx), Keutsch Group, Harvard.

The cornerstone of statistical mechanics of complex networks is the idea that the links, and not the nodes, are the effective particles of the system. Here we formulate a mapping between weighted networks and lattice gasses, making the conceptual step forward of interpreting weighted links as particles with a generalised coordinate. This leads to the definition of the grand canonical ensemble of weighted complex networks. We derive exact expressions for the partition function and thermodynamic quantities, both in the cases of global and local (i.e., node-specific) constraints on density and mean energy of particles. We further show that, when modelling real cases of networks, the binary and weighted statistics of the ensemble can be disentangled, leading to a simplified framework for a range of practical applications.

Physical models of biological systems can become difficult to interpret when they have a large number of parameters. But the models themselves actually depend on (i.e. are sensitive to) only a subset of those parameters. Rigorously identifying this subset of "stiff" parameters has been made possible by the development of parameter space compression (PSC). However, PSC has only been applied to analytically-solvable physical models. We have generalized this powerful method by developing a numerical approach to PSC that can be applied to any computational model. We validated our method against analytically-solvable models of random walk with drift and protein production and degradation. We then applied our method to an active area of biophysics research, namely to a simple computational model of microtubule dynamic instability. Such models have become increasingly complex, perhaps unnecessarily. By adding two new parameters that account for prominent structural features of microtubules, we identify one that can be "compressed away" (the "seam" in the microtubule) and another that is essential to model performance (the "tapering" of microtubule ends). Furthermore, we show that the microtubule model has an underlying, low-dimensional structure that explains the vast majority of our experimental data. We argue that numerical PSC can identify the low-dimensional structure of any computational model in biophysics. The low-dimensional structure of a model is easier to interpret and identifies the mechanisms and experiments that best characterize the system.

Understanding how gene expression translates to and affects human behaviour is one of the ultimate aims of neuroscience. In this paper, we present a pipeline based on Mapper, a topological simplification tool, to produce and analyze genes co-expression data. We first validate the method by reproducing key results from the literature on the Allen Human Brain Atlas, and the correlations between resting-state fMRI and gene co-expression maps. We then analyze a dopamine-related gene-set and find that co-expression networks produced by Mapper returned a structure that matches the well-known anatomy of the dopaminergic pathway. Our results suggest that topological network descriptions can be a powerful tool to explore the relationships between genetic pathways and their association with brain function and its perturbation due to illness and/or pharmacological challenge.

Parvalbumin (PV) positive interneurons modulate cortical activity through highly specialized connectivity patterns onto excitatory pyramidal neurons (PNs) and other inhibitory cells. PV cells are auto-connected through powerful autapses, but the contribution of this form of fast disinhibition to cortical function is unknown. We found that autaptic transmission represents the most powerful input of PV cells in neocortical Layer V. Autaptic strength was greater than synaptic strength onto PNs as result of a larger quantal size, whereas autaptic and heterosynaptic PV-PV synapses differed in the number of release sites. Overall, single-axon autaptic transmission contributed to ~40% of the total perisomatic inhibition that PV interneurons received. The strength of autaptic transmission modulated the coupling of PV-cell firing with optogenetically-induced {gamma}-oscillations preventing high frequency bursts of spikes. Autaptic self-inhibition represents an exceptionally large and fast disinhibitory mechanism to synchronize the output of PV cells during cognitive-relevant cortical network activity.

An early (Oldowan) chopping tool.

I have received a letter from a reader who goes by the handle jgkess. Under the title the origin of communicative intent in the use of hominem proto-language he (or maybe she) writes: “The idea was to get another to Do something, (or not do something) either proximally or distally (in a temporal sense), by way of getting him to think or feel in an intended way. There was no "generic" intent just to "inform" another---that would be insufficiently motivating, and communicative behaviour is, after all, motivated behaviour. In the pragmatics of hominem proto-linguistic communication, I think, lie the seeds of the evolution of our kind of general intelligence---this is a kind of take on Dan Sperber's work.”

Seventy years ago, Norbert Wiener published a book entitled Cybernetics or Control and Communication in  the Animal and the Machine. It publicized secret wartime achievements in getting machines to control one-another by communicating (i.e., by exchanging information). Back in the early 1970s I finally read the book, which describes communication entirely in terms of making something separate from the communicator act in certain way. Cells within an organism control one another, ants control one another, computers in a network control one another, Employees in a military organization control one another. So there seems to be much in favor of this idea of control, but while the book was eye-opening and powerful it did not persuade me that control is the main function of language.

As an English major, I immediately protested that humans also can recite Keats’ Ode on a Grecian Urn, or talk about what they did the previous weekend, or argue over politics, report a piece of news, or teach a course in advanced calculus. None of these tasks have a cybernetic function. The difference becomes obvious when you compare a high-level computer language like C++ with any natural language. C++ tells a machine what to do. It cannot be used for any of the purposes I listed at the top of this paragraph. Meanwhile, telling people what to do (writing procedures) is a special skill  that earns technical writers a livelihood. One of the guiding principles of this blog is that natural languages and computer languages are different things and it is a category error to refer to one and draw conclusions about the other (e.g., it is a mistake to argue that computer languages are not ambiguous, therefore natural languages should not have to be ambiguous either).

A natural question for this blog is why don’t apes talk? They seem smart enough and probably have a higher IQ than some people who do talk.  Forty years ago  there were a series of experiments in which apes were taught sign languages, proving they were smart enough to use some language. The problem was that apes could not get  beyond cybernetic motives. They used signs to signal their wants to humans and also answered questions creatively (e.g., Human: what’s that (pointing to a swan), Ape: water bird.) So apes are smart and creatuve, but they only volunteered one kind of statement: requests. They told humans they wanted a hug, or an apple, or even that  their tooth needed fixing. So they could signal their wants. Intriguingly, apes that knew how to sign did not start chatting with one another. I assume that was because apes already had ways of making requests of one another and found no advantage in signalling to their fellows they wanted an apple. Get it yourself, one can imagine Ape¹ telling Ape².

 

In his book, Origins of Communication, Michael Tomasello notes that among wild chimpanzees it is common for a youngster to lose sight of its mother and to begin to howl in anxiety. Other chimpanzees probably know what  the fuss is about and could point mama out to the upset toddler, but they never do that. They never use language or signaling to share their knowledge with others. They do not have the motivation to come to the youngster’s rescue.

But humans pitch in to inform others, even strangers, all the time. It is common for strangers in an area to ask for directions and receive them. jgkess denies that originally there was a  “‘generic’ intent just to ‘inform’ another---that would be insufficiently motivating” but it seems that the motivation has come along somehow since informing others is a routine part of daily, human existence,

It is common for two-year-olds to shout out the names of things they see. A toddler shouts doggie and a mother glances toward a TV screen and says, Yes, that’s a big dog. That is a fairly clear example of a human using language to inform another human for no good  reason beyond the drive to express what the human knows. Human communication is distinctive in function as well as structure from other known communication systems. Babies get adults to do things for them by using a communication system older than language: they cry.

On  this blog, I have insisted for years that the sine qua non of language is the speech triangle: speaker and listener paying joint attention to a topic. Not every utterance is defined by the speech triangle (e.g., Stop in the name of the law) but a communication system that cannot form a speech triangle is not a language.

If you want to imagine hominems first using speech to tell each other what to do, I cannot stop you or prove you wrong, but only when the hominems started using a speech triangle did they begin to use even a proto-language. We can see that almost 2 million years ago, the Homo lineage was passing along knowledge -- specifically, they taught new generations how to make Oldowan tools, We cannot prove they used language to  teach the tool making. They may have just shown students how to smash rocks together to get a cutting edge. The knowledge was passed along for a million years or more and spread over a  wide area. Nor did passing along knowledge stop. The tools eventually became more complex and required more teaching, This steady tradition of passing along knowledge is possible because of humanity’s unusual communal nature.

We depend on one another to become who we are. We become members of whatever community raises us, speaking its language, sharing its  tastes and customs, assuming its assumptions. About the only instinct we have left is the instinct to be like those around us (especially,  those who are raising us). Presumably, Homo habilis was not so dependent on its culture to make its members who  they were, but we have been heading in our current direction for a very long time. At some hazy patch along the way, we introduced language as an especially powerful tool for getting us to share information and thoughts, organizing human communities so that anybody’s genius could be shared. Sharing information, not controlling others, is and has been  the secret of the Homo lineage’s success.