Building artificial intelligence that can build artificial intelligence

This gets scarier and scarier. Clips from an article by Cade Metz:

...perhaps a nightmare for highly skilled computer programmers: artificially intelligent machines that can build other artificially intelligent machines...the Google project called AutoML...a machine-learning algorithm that learns to build other machine-learning algorithms.

The tech industry is promising everything from smartphone apps that can recognize faces to cars that can drive on their own. But by some estimates, only 10,000 people worldwide have the education, experience and talent needed to build the complex and sometimes mysterious mathematical algorithms that will drive this new breed of artificial intelligence.

Neural networks are rapidly accelerating the development of A.I. Rather than building an image-recognition service or a language translation app by hand, one line of code at a time, engineers can much more quickly build an algorithm that learns tasks on its own.

In building a neural network, researchers run dozens or even hundreds of experiments across a vast network of machines, testing how well an algorithm can learn a task like recognizing an image or translating from one language to another. Then they adjust particular parts of the algorithm over and over again, until they settle on something that works. Some call it a “dark art,” just because researchers find it difficult to explain why they make particular adjustments.

But with AutoML, Google is trying to automate this process. It is building algorithms that analyze the development of other algorithms, learning which methods are successful and which are not. Eventually, they learn to build more effective machine learning. Google said AutoML could now build algorithms that, in some cases, identified objects in photos more accurately than services built solely by human experts.

This is not always an easy thing to wrap your head around. But it is part of a significant trend in A.I. research. Experts call it “learning to learn” or “meta-learning.”...“Computers are going to invent the algorithms for us, essentially,” said a Berkeley professor, Pieter Abbeel. “Algorithms invented by computers can solve many, many problems very quickly — at least that is the hope.”

Buddhism is more Western than you think.

Robert Wright does a review of Adam Gopnik’s review (in the New Yorker) of Wright’s book “Why Buddhism Is True.” The whole piece is very clearly written and worth reading, and I want to pass on a few clips:

In recent decades, important aspects of the Buddhist concept of not-self have gotten support from psychology. In particular, psychology has bolstered Buddhism’s doubts about our intuition of what you might call the “C.E.O. self” — our sense that the conscious “self” is the initiator of thought and action…There is a paradox that can surface if you pursue the logic of not-self through meditation. Namely: recognizing that “you” are not in control, that you are not a C.E.O., can help give “you” more control. Or, at least, you can behave more like a C.E.O. is expected to behave: more rationally, more wisely, more reflectively; less emotionally, less rashly, less reactively.

Here’s how it can work. Suppose that, via mindfulness meditation, you observe a feeling like anxiety or anger and, rather than let it draw you into a whole train of anxious or angry thoughts, you let it pass away. Though you experience the feeling — and in a sense experience it more fully than usual — you experience it with “non-attachment” and so evade its grip. And you now see the thoughts that accompanied it in a new light — they no longer seem like trustworthy emanations from some “I” but rather as transient notions accompanying transient feelings.

There’s a broader and deeper sense in which Buddhist thought is more “Western” than stereotype suggests. What, after all, is more Western than science’s emphasis on causality, on figuring out what causes what, and hoping to thus explain why all things do the things they do? Well, in a sense, the Buddhist idea of “not-self” grows out of the belief undergirding this mission — that the world is pervasively governed by causal laws. The reason there is no “abiding core” within us is that the ever-changing forces that impinge on us — the sights, the sounds, the smells, the tastes — are constantly setting off chain reactions inside of us.

Indeed, this constant causal interaction with our environment raises doubts not only about how firm the core of the “self” is but, in a sense, how firm the bounds of the self are. Buddhism’s doubts about the distinctness and solidity of the “self” — and of other things, for that matter — rests on a recognition of the sense in which pervasive causality means pervasive fluidity.

…psychology has lately started to let go of its once-sharp distinction between “cognitive” and “affective” parts of the mind; it has started to see that feelings are so finely intertwined with thoughts as to be part of their very coloration. This wouldn’t qualify as breaking news in Buddhist circles. A sutra attributed to the Buddha says that a “mind object” — a category that includes thoughts — is just like a taste or a smell: whether a person is “tasting a flavor with the tongue” or “smelling an odor with the nose” or “cognizing a mind object with the mind,” the person “lusts after it if it is pleasing” and “dislikes it if it is unpleasing.”

Brain-scan studies have produced tentative evidence that this lusting and disliking — embracing thoughts that feel good and rejecting thoughts that feel bad — lies near the heart of certain “cognitive biases.” If such evidence continues to accumulate, the Buddhist assertion that a clear view of the world involves letting go of these lusts and dislikes will have drawn a measure of support from modern science.

The emotional intelligence of one- to four-year-olds

Interesting work from Wu et al. showing young children connect diverse positive emotional vocalizations to their probable causes, showing more sophisticated emotion understanding than previously realized:

The ability to understand why others feel the way they do is critical to human relationships. Here, we show that emotion understanding in early childhood is more sophisticated than previously believed, extending well beyond the ability to distinguish basic emotions or draw different inferences from positively and negatively valenced emotions. In a forced-choice task, 2- to 4-year-olds successfully identified probable causes of five distinct positive emotional vocalizations elicited by what adults would consider funny, delicious, exciting, sympathetic, and adorable stimuli (Experiment 1). Similar results were obtained in a preferential looking paradigm with 12- to 23-month-olds, a direct replication with 18- to 23-month-olds (Experiment 2), and a simplified design with 12- to 17-month-olds (Experiment 3). Moreover, 12- to 17-month-olds selectively explored, given improbable causes of different positive emotional reactions (Experiments 4 and 5). The results suggest that by the second year of life, children make sophisticated and subtle distinctions among a wide range of positive emotions and reason about the probable causes of others’ emotional reactions. These abilities may play a critical role in developing theory of mind, social cognition, and early relationships.

America is facing an epistemic crisis

The Chinese curse "May you live in interesting times" is certainly taking its toll on all of us who don't hide from the current news. I don't recall ever seeing so many interesting and incisive opinion essays in newspapers and magazines. I’ve taken the tile of this post from an article by David Roberts in Vox, and want to pass on a few clips:

Roberts asks "what if Mueller proves his case and it doesn't matter?":

Say Mueller reveals hard proof that the Trump campaign knowingly colluded with Russia, strategically using leaked emails to hurt Clinton’s campaign. Say the president — backed by the Wall Street Journal editorial page, Fox News, Breitbart, most of the US Cabinet, half the panelists on CNN, most of the radio talk show hosts in the country, and an enormous network of Russian-paid hackers and volunteer shitposters working through social media — rejects the evidence.

They might say Mueller is compromised. It’s a Hillary/Deep State plot. There’s nothing wrong with colluding with Russia in this particular way. Dems did it first. All of the above. Whatever.

Say the entire right-wing media machine kicks to life and dismisses the whole thing as a scam — and conservatives believe them. The conservative base remains committed to Trump, politicians remain scared to cross the base, and US politics remains stuck in partisan paralysis, unable to act on what Mueller discovers.

In short, what if Mueller proves the case and it’s not enough? What if there is no longer any evidentiary standard that could overcome the influence of right-wing media?

The US is experiencing a deep epistemic breach, a split not just in what we value or want, but in who we trust, how we come to know things, and what we believe we know — what we believe exists, is true, has happened and is happening. (Epistemology is the branch of philosophy having to do with how we know things and what it means for something to be true or false, accurate or inaccurate.)

As long as the base is convinced that Mueller is an agent of the deep state (or whatever), it will punish any Republican politician that strays from the pack and criticizes Trump. For a GOP officeholder, standing up for democratic integrity could mean sacrificing reelection in 2018 or 2020.

As long as Republican politicians are frightened by the base, the base is frightened by scary conspiracies in right-wing media, and right-wing media makes more money the more frightened everyone is, Trump appears to be safe. As long as the incentives are aligned in that direction, there will be no substantial movement to censure, restrain, or remove him from office.

Improving brain function by shocking it.

This post points to three recent articles on non-invasive electrical brain stimulation of various types that enhance brain brain function.

Krause et al. show that Transcranial Direct Current Stimulates associative learning and alters functional connectivity in the macaque monkey brain:

Highlights

• tDCS improves animals’ behavior on an associative learning task 

• Stimulation has local effects on LFP power and coherence. 

• It also causes frequency-specific changes in connectivity between brain areas 

• Inter-area coherence in gamma frequencies is linked to behavioral improvement 

Summary

There has been growing interest in transcranial direct current stimulation (tDCS), a non-invasive technique purported to modulate neural activity via weak, externally applied electric fields. Although some promising preliminary data have been reported for applications ranging from stroke rehabilitation to cognitive enhancement, little is known about how tDCS affects the human brain, and some studies have concluded that it may have no effect at all. Here, we describe a macaque model of tDCS that allows us to simultaneously examine the effects of tDCS on brain activity and behavior. We find that applying tDCS to right prefrontal cortex improves monkeys’ performance on an associative learning task. While firing rates do not change within the targeted area, tDCS does induce large low-frequency oscillations in the underlying tissue. These oscillations alter functional connectivity, both locally and between distant brain areas, and these long-range changes correlate with tDCS’s effects on behavior. Together, these results are consistent with the idea that tDCS leads to widespread changes in brain activity and suggest that it may be a valuable method for cheaply and non-invasively altering functional connectivity in humans.

Grossman et al. (Open Access) describe the use of multiple external high frequency electric fields to generate electric field envelopes inside the brain that can stimulate neurons. This could potentially substitute for current stimulation therapies for Parkinson’s disease, depression, and obsessive-compulsive disorder that require implanting electrodes in the brain.

And, an opinion article by Diana et al. discusses rehabilitation of the addicted brain with transcranial magnetic stimulation.

How linguistic metaphor scaffolds reasoning

Continuing the line of inquiry pioneered by Lakoff and Johnson's 1980 book, "Metaphors We Live by", Thibodeau et al. provide further examples of how the use of metaphor shapes our thoughts. I'm passing on their summary, and also two figures.

Abstract

Language helps people communicate and think. Precise and accurate language would seem best suited to achieve these goals. But a close look at the way people actually talk reveals an abundance of apparent imprecision in the form of metaphor: ideas are ‘light bulbs’, crime is a ‘virus’, and cancer is an ‘enemy’ in a ‘war’. In this article, we review recent evidence that metaphoric language can facilitate communication and shape thinking even though it is literally false. We first discuss recent experiments showing that linguistic metaphor can guide thought and behavior. Then we explore the conditions under which metaphors are most influential. Throughout, we highlight theoretical and practical implications, as well as key challenges and opportunities for future research.

Trends

Metaphors pervade discussions of abstract concepts and complex issues: ideas are ‘light bulbs’, crime is a ‘virus’, and cancer is an ‘enemy’ in a ‘war’.

At a process level, metaphors, like analogies, involve structure mapping, in which relational structure from the source domain is leveraged for thinking about the target domain.

Metaphors influence how people think about the topics they describe by shaping how people attend to, remember, and process information.

The effects of metaphor on reasoning are not simply the result of lexical priming.

Metaphors can covertly influence how people think. That is, people are not always aware that they have been influenced by a metaphor.

Two figures (click to enlarge):

Arousal versus relaxation in meditative practices.

I am grateful to Robert Ruhloff for his comment on MindBlog's Oct. 25th post on Mindfulness, in which he pointed to a reference whose abstract I would like to pass on here:

Based on evidence of parasympathetic activation, early studies defined meditation as a relaxation response. Later research attempted to categorize meditation as either involving focused or distributed attentional systems. Neither of these hypotheses received strong empirical support, and most of the studies investigated Theravada style meditative practices. In this study, we compared neurophysiological (EEG, EKG) and cognitive correlates of meditative practices that are thought to utilize either focused or distributed attention, from both Theravada and Vajrayana traditions. The results of Study 1 show that both focused (Shamatha) and distributed (Vipassana) attention meditations of the Theravada tradition produced enhanced parasympathetic activation indicative of a relaxation response. In contrast, both focused (Deity) and distributed (Rig-pa) meditations of the Vajrayana tradition produced sympathetic activation, indicative of arousal. Additionally, the results of Study 2 demonstrated an immediate dramatic increase in performance on cognitive tasks following only Vajrayana styles of meditation, indicating enhanced phasic alertness due to arousal. Furthermore, our EEG results showed qualitatively different patterns of activation between Theravada and Vajrayana meditations, albeit highly similar activity between meditations within the same tradition. In conclusion, consistent with Tibetan scriptures that described Shamatha and Vipassana techniques as those that calm and relax the mind, and Vajrayana techniques as those that require ‘an awake quality’ of the mind, we show that Theravada and Vajrayana meditations are based on different neurophysiological mechanisms, which give rise to either a relaxation or arousal response. Hence, it may be more appropriate to categorize meditations in terms of relaxation vs. arousal, whereas classification methods that rely on the focused vs. distributed attention dichotomy may need to be reexamined.

Self as object: Trends in Self Research

The current issue of Trends in Neurosciences has a nice open source article reviewing different aspects of assessing what our 'self' is, considering 'self as object' in neural terms:

Self representation is fundamental to mental functions. While the self has mostly been studied in traditional psychophilosophical terms (‘self as subject’), recent laboratory work suggests that the self can be measured quantitatively by assessing biases towards self-associated stimuli (‘self as object’). Here, we summarize new quantitative paradigms for assessing the self, drawn from psychology, neuroeconomics, embodied cognition, and social neuroscience. We then propose a neural model of the self as an emerging property of interactions between a core ‘self network’ (e.g., medial prefrontal cortex; mPFC), a cognitive control network [e.g., dorsolateral (dl)PFC], and a salience network (e.g., insula). This framework not only represents a step forward in self research, but also has important clinical significance, resonating recent efforts in computational psychiatry.

For your brain's sake, keep moving.

Gretchen Reynolds points to work by van Praag and collaborators showing that a week of activity rather than inactivity (in adult male rats) increases both the formation of new nerve cells in the hippocampus and the richness of their interactions. The new cells had more and longer dendrites, more of which led to spatial memory areas.

Where our brains perceive social interactions.

From Isik et al.:

Significance

Humans spend a large percentage of their time perceiving the appearance, actions, and intentions of others, and extensive previous research has identified multiple brain regions engaged in these functions. However, social life depends on the ability to understand not just individuals, but also groups and their interactions. Here we show that a specific region of the posterior superior temporal sulcus responds strongly and selectively when viewing social interactions between two other agents. This region also contains information about whether the interaction is positive (helping) or negative (hindering), and may underlie our ability to perceive, understand, and navigate within our social world.

Abstract

Primates are highly attuned not just to social characteristics of individual agents, but also to social interactions between multiple agents. Here we report a neural correlate of the representation of social interactions in the human brain. Specifically, we observe a strong univariate response in the posterior superior temporal sulcus (pSTS) to stimuli depicting social interactions between two agents, compared with (i) pairs of agents not interacting with each other, (ii) physical interactions between inanimate objects, and (iii) individual animate agents pursuing goals and interacting with inanimate objects. We further show that this region contains information about the nature of the social interaction—specifically, whether one agent is helping or hindering the other. This sensitivity to social interactions is strongest in a specific subregion of the pSTS but extends to a lesser extent into nearby regions previously implicated in theory of mind and dynamic face perception. This sensitivity to the presence and nature of social interactions is not easily explainable in terms of low-level visual features, attention, or the animacy, actions, or goals of individual agents. This region may underlie our ability to understand the structure of our social world and navigate within it.

Modern flimflam men? - The Flow Genome Project

(Note added Aug.16, 2019.  Please see this update: https://mindblog.dericbownds.net/2019/08/a-schism-in-flow-land-flow-genome.html. Kolter and Wheal have parted ways and have competing websites. Kolter's is the Flow Research Collective and Wheal's is the link given below to Flow Genome Project.)

Cleaning up my queue of articles on which a MindBlog post might be based, I came across this piece by Casey Schwartz titled “How to Hack your Brain (for $5,000)," which immediately triggered my bullshit detector. It uncritically describes what seems to me a circus act devised by internet age flimflam men, Jamie Wheal and Steven Kotler, whose company (the Flow Genome Project, based in my own new hometown of Austin Texas!) is “dedicated to gathering the latest science behind flow states. It’s board of advisers includes neuroscientists, filmmakers and a kiteboarder.” The result seems to be this kind of gibble-gabble of hand waving about various neurotransmitters. From Schwartz's article:

“Flow,” they write, is associated with six neurotransmitters: dopamine, serotonin, oxytocin, norepinephrine, anandamide and endorphins. Knowing the neurochemical profile of flow means, in theory, people can devise ways of achieving it more often, more reliably and more quickly.

One tries in vain to find anything of substance on their website, such as a list of the neuroscientists, or references to work on the neurotransmitters mentioned. Their core video presents the two bright-eyed and bushy tailed entrepreneurs engaging you with their personal stories and lots of kewl graphics of spinning brains and neurons. Since I'm being so negative, I felt obliged to buy the Kindle version of "Stealing Fire" by Kolter and Wheal.

The bottom line is that it is an creative, wide ranging, everything but the kitchen sink, whacked out, exuberant, off the wall advertisement for Flow Genome which doesn't offer much substance. It has detailed references and what looks on the surface like a very respectable bibliography. I can't even begin to describe the confusion and chaos that lies below this veneer if one simply begins to follow through on any of the reference threads. Clicking on footnotes that purport to be supportive of the 'science' gets you a mishmash of review articles. There are several references to "The knobs and levers being tweaked in the brain: See www.flowgenomeproject.com/stealingfiretools." This link does not work. Or, "And if you’re interested in helping further this research, visit: www.stealingfirebook.com/research/". This link does not work.

I have great respect for  Mihaly Csikszentmihalyi’s original writings on the state of ‘flow,’ which clearly lays out Kolter and Wheal's' "four signature characteristics underneath: Selflessness, Timelessness, Effortlessness, and Richness, or STER for short." However, my 75 year old curmudgeonly brain is not sympathetic to the package offered by the Flow Genome Project, whose claimed positive outcome I suspect might best be described as a mass placebo effect induced by a pseudoscientific charismatic religious act...If you believe it works, it works!

Focus on present predicts enhanced life satisfaction but not happiness

Another study by Felsman et al., in the vein of the work described in MindBlog's Oct. 25 post. That study claimed a correlation between present-moment attention and increased positive affect, this study suggests a correlation with life satisfaction but not happiness:

Mindfulness theorists suggest that people spend most of their time focusing on the past or future rather than the present. Despite the prevalence of this assumption, no research that we are aware of has evaluated whether it is true or what the implications of focusing on the present are for subjective well-being. We addressed this issue by using experience sampling to examine how frequently people focus on the present throughout the day over the course of a week and whether focusing on the present predicts improvements in the 2 components of subjective well-being over time—how people feel and how satisfied they are with their lives. Results indicated that participants were present-focused the majority of the time (66%). Moreover, focusing on the present predicted improvements in life satisfaction (but not happiness) over time by reducing negative rumination. These findings advance our understanding of how temporal orientation and well-being relate.

Cognitive reflection in men impaired by single testosterone dose

An interesting bit from Nave et al.:

In nonhumans, the sex steroid testosterone regulates reproductive behaviors such as fighting between males and mating. In humans, correlational studies have linked testosterone with aggression and disorders associated with poor impulse control, but the neuropsychological processes at work are poorly understood. Building on a dual-process framework, we propose a mechanism underlying testosterone’s behavioral effects in humans: reduction in cognitive reflection. In the largest study of behavioral effects of testosterone administration to date, 243 men received either testosterone or placebo and took the Cognitive Reflection Test (CRT), which estimates the capacity to override incorrect intuitive judgments with deliberate correct responses. Testosterone administration reduced CRT scores. The effect remained after we controlled for age, mood, math skills, whether participants believed they had received the placebo or testosterone, and the effects of 14 additional hormones, and it held for each of the CRT questions in isolation. Our findings suggest a mechanism underlying testosterone’s diverse effects on humans’ judgments and decision making and provide novel, clear, and testable predictions.

The split brain revisited - a new perspective

I pass on the suggestion by Pinto et al. that classical experiments on subjects whose corpus callosum communicating between the two hemispheres has been severed, on detailed analysis, do not imply that those subjects now have split consciousness, or two different 'selves':

Summary

The split-brain phenomenon is caused by the surgical severing of the corpus callosum, the main route of communication between the cerebral hemispheres. The classical view of this syndrome asserts that conscious unity is abolished. The left hemisphere consciously experiences and functions independently of the right hemisphere. This view is a cornerstone of current consciousness research. In this review, we first discuss the evidence for the classical view. We then propose an alternative, the ‘conscious unity, split perception’ model. This model asserts that a split brain produces one conscious agent who experiences two parallel, unintegrated streams of information. In addition to changing our view of the split-brain phenomenon, this new model also poses a serious challenge for current dominant theories of consciousness.

Trends

Five hallmarks characterize the split-brain syndrome: a response × visual field interaction, strong hemispheric specialization, confabulations after left-hand actions, split attention, and the inability to compare stimuli across the midline.

These hallmarks underlie the classical notion that split brain implies split consciousness. This notion suggests that massive interhemispheric communication is necessary for conscious unity.

Closer examination challenges the classical notion. Either the hallmark also occurs in healthy adults or the hallmark does not hold up for all split-brain patients.

A re-evaluation of the split-brain data suggests a new model that might better account for the data. This model asserts that a split-brain patient is one conscious agent with unintegrated visual perception.

This new model challenges prominent theories of consciousness, since it implies that massive communication is not needed for conscious unity.