The visible gorilla.

A staple of my lectures in the 1990s was showing the ‘invisible gorilla’ video, in which viewers were asked to count the number of times that students with white shirts passed a basket ball. After the start of the game a student in a gorilla costume walks slowly through the group, pauses in the middle to wave and moves off screen to the left. Most viewers who are busy counting the ball passes don’t report seeing the gorilla. Here's the video:

 

Wallish et al. now update this experiment on inattentional blindness in an article titled "The visible gorilla: Unexpected fast—not physically salient—Objects are noticeable." Here are their summaries:  

Significance

Inattentional blindness, the inability to notice unexpected objects if attention is focused on a task, is one of the most striking phenomena in cognitive psychology. It is particularly surprising, in light of the research on attentional capture and motion perception, that human observers suffer from this effect even when the unexpected object is moving. Inattentional blindness is commonly interpreted as an inevitable cognitive deficit—the flip side of task focusing. We show that this interpretation is incomplete, as observers can balance the need to focus on task demands with the need to hedge for unexpected but potentially important objects by redeploying attention in response to fast motion. This finding is consistent with the perspective of a fundamentally competent agent who effectively operates in an uncertain world.

Abstract

It is widely believed that observers can fail to notice clearly visible unattended objects, even if they are moving. Here, we created parametric tasks to test this belief and report the results of three high-powered experiments (total n = 4,493) indicating that this effect is strongly modulated by the speed of the unattended object. Specifically, fast—but not slow—objects are readily noticeable, whether they are attended or not. These results suggest that fast motion serves as a potent exogenous cue that overrides task-focused attention, showing that fast speeds, not long exposure duration or physical salience, strongly diminish inattentional blindness effects.

Unnarratability -The Tower of Babel redux - where have all the common narratives gone?

I pass on some clips from Venkatesh Rao's recent Ribbonfarm Studio posting.. Perspectives like his make me feel that one's most effective self preservation stance might be to assume that we are on the dawn of a new dark age, a period during which only power matters, and community, cooperation, and kindness are diminished - a period like the early middle ages in Europe which did permit under the sheltered circumstances of the church a privileged few to a life of contemplation.    

Strongly Narratable Conditions

The 1985-2015 period, arguably, was strongly narratable, and unsurprisingly witnessed the appearance of many strong global grand narratives. These mostly hewed to the logic of the there-is-no-alternative (TINA) platform narrative of neoliberalism, even when opposed to it...From Francis Fukuyama and Thomas Friedman in the early years, to Thomas Piketty, Yuval Noah Harari, and David Graeber in the final years, many could, and did, peddle coherent (if not always compelling) Big Histories. Narrative performance venues like TED flourished. The TINA platform narrative supplied the worldwinds for all narratives.

Weakly Narratable Conditions

The 2007-2020 period, arguably, was such a period (the long overlap of 8 years, 2007-15, was a period with uneven weak/strong narratability). In such conditions, a situation is messed-up and contentious, but in a way that lends itself to the flourishing of a pluralist, polycentric narrative landscape, where there are multiple contending accounts of a shared situation, Rashomon style, but the situation is merely ambiguous, not incoherent.

While weakly narratable conditions lack platform narratives, you could argue that there is something of a prevailing narrative protocol during weakly narratable times - an emergent lawful pattern of narrative conflict that cannot be codified into a legible set of consensus rules of narrative engagement, but produces slow noisy progress anyway, does not devolve into confused chaos, and sustains a learnable narrative literacy.

This is what it meant to be “very online” in 2007-20. It meant you had acquired a certain literacy around the prevailing narrative protocol. Perhaps nobody could make sense of what was going on overall, beyond their private, solipsistic account of events, and it was perhaps not possible to play to win, but there was enough coherence in the situation that you could at least play to not lose.

Unnarratable Conditions

The pandemic hit, and we got to what I think of as unnarratable conditions...While the specific story of the pandemic itself was narratable, the story of the wider post-Weirding world, thrown into tumult by the pandemic, was essentially unnarratable.

Unnarratable times are fundamentally incoherent melees of contending historical forces. Times when there isn’t even a narrative protocol you can acquire a reliable literacy in, let alone a platform narrative upon which to rest your sense-making efforts. Where the environmental entropy is so high, people struggle to put together any kind of narrative, even solipsistic private ones that harbor no ambitions of inDuencing others. There is no privileged class (comparable to the “Very Online” before 2020) that can plausibly claim a greater narrative literacy than other classes.

Those who claim to possess satisfying grand narratives are barely able to persuade even close allies to share it, let alone induce narrative protocols through them, or install them as platform narratives. The result: a collective retreat to a warren of cozy cultural redoubts, usually governed by comforting reactionary or nostalgic local narratives, and a derelict public discourse.

We have been in such a condition at least since 2022, and arguably since 2020. If you set aside the narrow liminal story of the pandemic, the world has been nearly unnarratable for years now.

A MindBlog paragraph on non-dual awareness massaged by Bard and ChatGPT-4

This post is using MindBlog to archive a further example example of the sort shown in Monday’s post ...of GPT bots editing  a paragraph I wrote, this one being on non-dual awareness.  The paragraph:

The fact that  there is no there there means it. Empty is empty. The ground state is just the mute body doing its thing. From its depths storylines rise and fall, bonbons to savor or ignore. The body can know that it has found a new ground state, and can be no longer willing to just play nice if the upstairs word salad is dinking around with or threatening its peace. It can withdraw from the fray. It can rest as a calm non-dual self centered within a cosmic perspective of time and space that views the rise and fall of human civilizations during the anthropocene as minor dances in the music of time.

After incorporating some changes suggested by Google Bard and GPT4:

When we say something is empty, we mean just that: it contains nothing. Our natural baseline or 'ground state' is the body performing its automatic functions without our conscious intervention. From the body's inherent depths, various thoughts or experiences emerge and recede, like bonbons we can savor or ignore. The body can grow to know that it has found a new ground state in the present moment, a ground state of awareness that no longer needs to be caught up in word salads of disruptive thoughts that can threaten its peace.  It can withdraw from the fray and can rest as a calm non-dual state of being centered within a cosmic perspective of time and space that can view the rise and fall of human civilizations during the anthropocene as minor dances in the music of time.

The evolution of transhuman forms - a MindBlog paragraph edited by GPT implementations

In this post I am documenting the final result of passing a paragraph I wrote on future human evolution through a number of different GPT (generative pre-trained transformers). This is to archive the results of this exercise for myself, and in case it might be of interest to a few MindBlog readers. The bots were prompted by a request to re-write or edit the paragraph to make it more accessible to a reader with a college education.

The bottom line is that the edited versions with the most depth and coherence were obtained using Open AI’s ChatGPT-4.  Bard.google.com, Google Workspace Labs, and Microsoft’s Bing, respectively, provided increasingly less useful versions.   Here is my original paragraph:

The linguistic capabilities of generative pre-trained transformers based on silicon  - the same capabilities that have enabled human civilization  - will soon exceed those of humans. We have brought our understanding of what human beings as they now exist are about to a plateau level at which further increments of understanding will become increasingly small. The cream has been skimmed, the best cherries have been picked, and it seems appropriate to turn our  attention now to studying the evolution of trans human forms utilizing LLMs that are better able to manage themselves and the planet.  The action is in the the pretraining P of GPT, as the models move beyond their initial human management to the stage of evolving themselves, hopefully into benign versions of Darwin’s  “forms most beautiful and most wonderful.” What is the point of humans other than being self replicating machines that evolve into further forms, hopefully before they extinguish themselves by destroying the ecological environment required for their sustenance? Can they evolve into trans human minds (or mind, singular)?

Here is the result of including some of the edits suggested by ChatGPT4’s regenerated response, a  vers. 3 of the paragraph above:

The linguistic capabilities of generative pre-trained transformers based on silicon - the same capabilities that have enabled human civilization - will soon surpass those of humans.

Our understanding of human nature as it currently exists has reached a plateau level, at which further increments of understanding are becoming increasingly small. The cream has been skimmed, the best cherries have been picked. Consequently it seems the right time to shift our focus toward studying the evolution of 'transhuman' forms. These are beings that could potentially leverage advanced language models (LLMs) to better self-regulate themselves and manage the planet.

The real crux of the matter lies in the pretraining phase of these transformers (symbolized by 'P' in GPT). This is when the models are initially developed and fine-tuned. As they evolve beyond their human programming the optimistic vision is that they might transform into benign versions of what Charles Darwin called “forms most beautiful and most wonderful.”

From a philosophical perspective, one might ask: What is the ultimate purpose of human beings, if not to serve as self-replicating entities that evolve into more advanced forms - hopefully before they extinguish themselves by destroying the ecological environment required for their sustenance? Is it possible for humans to evolve into a collective of transhuman minds or even a singular, transcendent mind?  These are questions worth exploring as we stand on the brink of an AI-enhanced future.

Proxy Failure is an Inherent Risk in Goal-Oriented Systems

I will pass on the title and abstract of another article to appear in Behavioral and Brain Science for which reviewers comments are being solicited. MindBlog readers can email me to request a PDF of the target article. 

Dead rats, dopamine, performance metrics, and peacock tails: proxy failure is an inherent risk in goal- oriented systems 

Authors: Yohan J. John, Leigh Caldwell, Dakota E. McCoy, and Oliver Braganza 

Abstract: When a measure becomes a target, it ceases to be a good measure. For example, when standardized test scores in education become targets, teachers may start 'teaching to the test', leading to breakdown of the relationship between the measure--test performance--and the underlying goal--quality education. Similar phenomena have been named and described across a broad range of contexts, such as economics, academia, machine-learning, and ecology. Yet it remains unclear whether these phenomena bear only superficial similarities, or if they derive from some fundamental unifying mechanism. Here, we propose such a unifying mechanism, which we label proxy failure. We first review illustrative examples and their labels, such as the 'Cobra effect', 'Goodhart's law', and 'Campbell's law'. Second, we identify central prerequisites and constraints of proxy failure, noting that it is often only a partial failure or divergence. We argue that whenever incentivization or selection is based on an imperfect proxy measure of the underlying goal, a pressure arises which tends to make the proxy a worse approximation of the goal. Third, we develop this perspective for three concrete contexts, namely neuroscience, economics and ecology, highlighting similarities and differences. Fourth, we outline consequences of proxy failure, suggesting it is key to understanding the structure and evolution of goal-oriented systems. Our account draws on a broad range of disciplines, but we can only scratch the surface within each. We thus hope the present account elicits a collaborative enterprise, entailing both critical discussion as well as extensions in contexts we have missed.

MindBlog's reading list.

I've decided to pass on links to articles I have found worthwhile reading , realizing that I am not going to have time to frame their ideas into longer posts because I'm speading more time now at my Steinway B's keyboard than at my computer's keyboard. If you encounter a paywall with any of the links, you might try entering the URL at https://archive.is/.

An installment of Venkatesh Rao’s newsletter: The permaweird narrative 

Jaron Lanier “There is no A.I.” in The New Yorker  

‘Human Beings Are Soon Going to Be Eclipsed’ David Brooks in The New York Times commenting on Douglas Hofstadter's  recent ideas.  

Marc Andreessen offers a horrific commentary titled "Fighting" on Elon Musk challenging Mark Zuckerberg to a cage fight.  

Learning from history. Archeological evidence that early hierarchical or authoritarian cultures didn't persist as long as more cooperative eqalitarian ones.  

Arthur Brooks on "The illusion of explanatory depth", an installment in his series "How to build a life.""  

Potential anti-aging therapy.  One sample of the effusive outpouring of new ideas and widgets offered by New Atlas.

 

 

 

‘Adversarial’ search for neural basis of consciousness yields first results

Finkel does a summary of the first round of results of an 'adversarial colloboration' funded by the Templeton World Charity Foundation in which

...both sides of the consciousness debate agreed on experiments to be conducted by “theory-neutral” labs with no stake in the outcome. It pits integrated information theory (IIT), the sensory network hypothesis that proposes a posterior “hot zone” as the site of consciousness, against the global neuronal workspace theory (GNWT), which likens networks of neurons in the front of the brain to a clipboard where sensory signals, thoughts, and memories combine before being broadcast across the brain.

The results corroborate IIT’s overall claim that posterior cortical areas are sufficient for consciousness, and neither the involvement of [the prefrontal cortex] nor global broadcasting are necessary,”

The article describes how the debate continues, with advocates of the prefrontal view suggesting this first experimental round had limitations, and that further experiments will support the role of the prefrontal cortex.

The True Threat of Artificial Intelligence

I would recommend having a read through Evgeny Morozov's piece in the NYTimes as an antidote to Marc Adreessen's optimistic vision of AI that was the subject of MindBlog's June 23 post. Here is a small clip from the article, followed by the titles describing different problem areas he sees:

Discussions of A.G.I. are rife with such apocalyptic scenarios. Yet a nascent A.G.I. lobby of academics, investors and entrepreneurs counter that, once made safe, A.G.I. would be a boon to civilization...This is why, for all the hand-wringing, so many smart people in the tech industry are toiling to build this controversial technology: not using it to save the world seems immoral.

They are beholden to an ideology that views this new technology as inevitable and, in a safe version, as universally beneficial. Its proponents can think of no better alternatives for fixing humanity and expanding its intelligence.

But this ideology — call it A.G.I.-ism — is mistaken. The real risks of A.G.I. are political and won’t be fixed by taming rebellious robots. The safest of A.G.I.s would not deliver the progressive panacea promised by its lobby. And in presenting its emergence as all but inevitable, A.G.I.-ism distracts from finding better ways to augment intelligence.

Unbeknown to its proponents, A.G.I.-ism is just a bastard child of a much grander ideology, one preaching that, as Margaret Thatcher memorably put it, there is no alternative, not to the market.

Rather than breaking capitalism, as Mr. Altman has hinted it could do, A.G.I. — or at least the rush to build it — is more likely to create a powerful (and much hipper) ally for capitalism’s most destructive creed: neoliberalism.

Fascinated with privatization, competition and free trade, the architects of neoliberalism wanted to dynamize and transform a stagnant and labor-friendly economy through markets and deregulation.

Some of these transformations worked, but they came at an immense cost. Over the years, neoliberalism drew many, many critics, who blamed it for the Great Recession and financial crisis, Trumpism, Brexit and much else.

It is not surprising, then, that the Biden administration has distanced itself from the ideology, acknowledging that markets sometimes get it wrong. Foundations, think tanks and academics have even dared to imagine a post-neoliberal future.

Yet neoliberalism is far from dead. Worse, it has found an ally in A.G.I.-ism, which stands to reinforce and replicate its main biases: that private actors outperform public ones (the market bias), that adapting to reality beats transforming it (the adaptation bias) and that efficiency trumps social concerns (the efficiency bias).

These biases turn the alluring promise behind A.G.I. on its head: Instead of saving the world, the quest to build it will make things only worse. Here is how.

A.G.I. will never overcome the market’s demands for profit.

A.G.I. will dull the pain of our thorniest problems without fixing them.

A.G.I. undermines civic virtues and amplifies trends we already dislike.

Depending on how (and if) the robot rebellion unfolds, A.G.I. may or may not prove an existential threat. But with its antisocial bent and its neoliberal biases, A.G.I.-ism already is: We don’t need to wait for the magic Roombas to question its tenets.

Inheritance of social status - stability in England from 1600 to 2022.

From historical records Clarks demonstrates (open source) strong persistence of social status across family trees over 400 years, in spite of large increases in general levels of education and social mobility.  

Significance

There is widespread belief across the social sciences in the ability of social interventions and social institutions to significantly influence rates of social mobility. In England, 1600 to 2022, we see considerable change in social institutions across time. Half the population was illiterate in 1,800, and not until 1,880 was compulsory primary education introduced. Progressively after this, educational provision and other social supports for poorer families expanded greatly. The paper shows, however, that these interventions did not change in any measurable way the strong familial persistence of social status across generations.

Abstract

A lineage of 422,374 English people (1600 to 2022) contains correlations in social outcomes among relatives as distant as 4th cousins. These correlations show striking patterns. The first is the strong persistence of social status across family trees. Correlations decline by a factor of only 0.79 across each generation. Even fourth cousins, with a common ancestor only five generations earlier, show significant status correlations. The second remarkable feature is that the decline in correlation with genetic distance in the lineage is unchanged from 1600 to 2022. Vast social changes in England between 1600 and 2022 would have been expected to increase social mobility. Yet people in 2022 remain correlated in outcomes with their lineage relatives in exactly the same way as in preindustrial England. The third surprising feature is that the correlations parallel those of a simple model of additive genetic determination of status, with a genetic correlation in marriage of 0.57.

A meta-analysis questions the cognitive benefits of physical activity.

I give up. If anything was supposed to have been proven I would have thought it would be that exercise has a beneficial effect on brain health and cognition. Now Ciria et al. offer the following in Nature Human Biology:

Extensive research links regular physical exercise to an overall enhancement of cognitive function across the lifespan. Here we assess the causal evidence supporting this relationship in the healthy population, using an umbrella review of meta-analyses limited to randomized controlled trials (RCTs). Despite most of the 24 reviewed meta-analyses reporting a positive overall effect, our assessment reveals evidence of low statistical power in the primary RCTs, selective inclusion of studies, publication bias and large variation in combinations of pre-processing and analytic decisions. In addition, our meta-analysis of all the primary RCTs included in the revised meta-analyses shows small exercise-related benefits (d = 0.22, 95% confidence interval 0.16 to 0.28) that became substantially smaller after accounting for key moderators (that is, active control and baseline differences; d = 0.13, 95% confidence interval 0.07 to 0.20), and negligible after correcting for publication bias (d = 0.05, 95% confidence interval −0.09 to 0.14). These findings suggest caution in claims and recommendations linking regular physical exercise to cognitive benefits in the healthy human population until more reliable causal evidence accumulates.

I can not offer an informed opinion on this abstract because my usual access to journals through the University of Wisconsin library does not work with Nature Human Behavior. However, I can point you to an excellent commentary by Claudia Lopez Lloreda that discusses the meta-analysis done by Ciria et al. and gives a summary of several recent studies on exercise and brain health.

Why music training slows cognitive aging

A team of Chinese collaborators has reported experiments in the Oxford academic journal Cerebral Cortex titled "Functional gradients in prefrontal regions and somatomotor networks reflect the effect of music training experience on cognitive aging" which are stated to show that music training enhances the functional separation between regions across prefrontal and somatomotor networks, delaying deterioration in working memory performance and prefrontal suppression of prominant but irrelevant information. I'm passing on the abstract and a clip from the paper's conclusion, and can send interested readers the whole article. I think it is an important article but I find it is rendered almost unintelligble by Chinese to English translation issues. I'm surprised the journal let this article appear without further editing.

Studies showed that the top-down control of the prefrontal cortex (PFC) on sensory/motor cortices changes during cognitive aging. Although music training has demonstrated efficacy on cognitive aging, its brain mechanism is still far from clear. Current music intervention studies have paid insufficient attention to the relationship between PFC and sensory regions. Functional gradient provides a new perspective that allows researchers to understand network spatial relationships, which helps study the mechanism of music training that affects cognitive aging. In this work, we estimated the functional gradients in four groups, young musicians, young control, older musicians, and older control. We found that cognitive aging leads to gradient compression. Compared with young subjects, older subjects presented lower and higher principal gradient scores in the right dorsal and medial prefrontal and the bilateral somatomotor regions, respectively. Meanwhile, by comparing older control and musicians, we found a mitigating effect of music training on gradient compression. Furthermore, we revealed that the connectivity transitions between prefrontal and somatomotor regions at short functional distances are a potential mechanism for music to intervene in cognitive aging. This work contributes to understanding the neuroplasticity of music training on cognitive aging.

From the conclusion paragraph:

In a nutshell, we demonstrate the top-down control of prefrontal regions to the somatomotor network, which is associated with inhibitory function and represents a potential marker of cognitive aging, and reveal that music training may work by affecting the connectivity between the two regions. Although this work has investigated the neuroplasticity of music on cognitive aging by recruiting subjects of different age spans, the present study did not include the study of longitudinal changes of the same group. Further studies should include longitudinal follow-up of the same groups over time to more accurately evaluate the effect of music intervention on the process of cognitive aging.

What Babies Know from zero to 1 year - core systems of knowledge

The journal Behavioral and Brain Sciences has sent out to reviewers the précis of a book, "What Babies Know" by Elizabeth S. Spelke, Harvard Psychology Dept. The abstract of her précis:

Where does human knowledge begin? Research on human infants, children, adults, and non- human animals, using diverse methods from the cognitive, brain, and computational sciences, provides evidence for six early emerging, domain-specific systems of core knowledge. These automatic, unconscious systems are situated between perceptual systems and systems of explicit concepts and beliefs. They emerge early in infancy, guide children’s learning, and function throughout life.

Spelke lists domain-specific core systems that are ancient, emerge early in life, and are invariant over later development. These deal with vision, objects, places, number, core knowledge, agents, social cognition, and language. Figures in the précis illustrate basic experiments characterizing the core systems. Motivated readers can obtain a PDF of the precis by emailing me.