The data is there

by liberal japonicus

I'm sure everyone is getting a surfeit of doomscrolling from a lot of other places. Here at ObWi, we like to wrap our observations about the coming judgement day in wrapping paper that says 'ain't that amazing'.

“Recognizing and segmenting visual objects in images, as well as environmental sounds and spoken words in audio recordings, are each difficult problems in their own right. Historically researchers have relied upon expensive, human-provided annotations in order to train machine learning models to accomplish these tasks,” says David Harwath, assistant professor in computer science at the University of Texas at Austin who was not involved in the work. “DenseAV makes significant progress towards developing methods that can learn to solve these tasks simultaneously by simply observing the world through sight and sound — based on the insight that the things we see and interact with often make sound, and we also use spoken language to talk about them. This model also makes no assumptions about the specific language that is being spoken, and could therefore in principle learn from data in any language. It would be exciting to see what DenseAV could learn by scaling it up to thousands or millions of hours of video data across a multitude of languages.”

Discuss.

Comments

[JanieM]

It would be exciting to see what DenseAV could learn by scaling it up to thousands or millions of hours of video data across a multitude of languages.

All of our home videos ever posted via the web....

From here (no idea if it's a reliable site but I've heard these kinds of numbers for years):

As of June 2022, there are more than 500 hours of new videos uploaded per minute on the platform [Youtube]. This equates to approximately 300,000 hours of new content per hour.

I wonder what the AI would make of online porn.

[Hartmut]

I wonder what the AI would make of online porn.

Given what problems it has with hands, this could yield some strange anatomical variations. And there would be a market for it.
Most of it tends to be non-verbal or with a very limited vocabulary, so the machine learning to speak pornish would produce an interesting accent but - depending on source - restrict the topics to mainly theology, groceries and a bit of vernacular zoology.

[wj]

And alternative approach to training an AI language model:
https://www.sciencenews.org/article/babies-ai-learn-words-model

Not, be it noted, a Large Language Model. But rather, an approach the attempts to mimic the way a human infant actually learns language.

Side note: It appears that the reasearch is based on an infant learning English. One obvious next step would be to try and replicate the experiment using other, preferably non-European, languages.

[Michael Cain]

TTBOMK, all of the current AI excitement is over large neural networks in a variety of configurations, trained using some sort of error function and stochastic gradient descent, and the networks limited to those where back propagation and the chain rule can be used. All of the math is 30 years old, or older. (Leibniz mentioned the chain rule in 1676.) We've now developed hardware sufficient to apply the math to networks large enough to be useful.

There's lots of research to be done on what different network configurations excel at, and what can be done with different sorts of error functions. My intuition is that this will eventually stall, and the next real breakthrough will be on the mathematics side.

Every time I read about how our brains work, it seems to include "it's more complicated than we thought."

[JanieM]

Every time I read about how our brains DNA works, it seems to include "it's more complicated than we thought."

Every time I read about how our brains anything works, it seems to include "it's more complicated than we thought."

Someone once told me about an image from Neil DeGrasse Tyson, of all knowledge being enclosed in a circle, with the circumference of the circle expanding as more knowledge was added. But the circumference getting bigger suggests that there's also ever more to discover.

I feel very mortal. Going to miss a lot by dying....

[Snarki, child of Loki]

Yeah, it's just so sad how people who died a few decades ago missed out on all the wonders of the Internet.

Okay, bad example.

[Michael Cain]

Every time I read about how... DNA works, it seems to include "it's more complicated than we thought."

The expert opinion seems to be shifting from "most of DNA is just junk" to "most of what we thought was junk codes mRNA and other things that modulate/trigger/suppress expression of the protein-coding parts of the DNA."

I feel very mortal. Going to miss a lot by dying....

So much to learn, so little time. When I tell my children that when the time comes, I'll be one of those hanging on by my fingernails, saying, "No! I'm not done yet!" they smile and nod.

[Donald]

Larry Moran thinks most of our DNA is junk.

https://sandwalk.blogspot.com/2024/06/tom-cech-writes-about-dark-matter-of.html

Part of the reason for thinking this is the c value paradox.

https://www.cell.com/current-biology/pdf/S0960-9822(12)01154-2.pdf

Not my field but I have read Moran off and on for years.

[wj]

Larry Moran thinks most of our DNA is junk.

Most of everything is junk. Right up until somebody identifies something useful about it.

[CharlesWT]

Some researchers view DNA not as a blueprint but more as an operations manual for a manufacturing plant, the cell.

Some researchers are moving away from gene-centric to cell and organism-centric research.

"The article titled "Evolution May Be Purposeful And It’s Freaking Scientists Out" by Andréa Morris delves into a controversial perspective on evolution proposed by Denis Noble, a biophysicist and a fellow of the Royal Society. Noble challenges the gene-centric model of evolution, which has dominated the field since the publication of Richard Dawkins' "The Selfish Gene" in 1976, advocating instead for a paradigm where purpose and intentionality play crucial roles in evolutionary processes." (Summary)
Evolution May Be Purposeful And It’s Freaking Scientists Out

[CharlesWT]

Further out on the edge.

"The article, "Every Single Cell in Your Body Could Be Conscious, Scientists Say. That Could Rewrite Everything We Know About Human Evolution," discusses a revolutionary idea proposed by evolutionary biologist William B. Miller and his colleagues. They suggest that the roughly 37 trillion cells in the human body possess a form of consciousness, a concept that could fundamentally alter our understanding of biology and evolution." (Summary)
Every Single Cell in Your Body Could Be Conscious, Scientists Say. That Could Rewrite Everything We Know About Human Evolution: If trillions of tiny bits of consciousness are floating around inside you, it could change how we think about life.

[CharlesWT]

And finally Greg Bear's 1985 novel:

"Renegade biotechnologist Vergil Ulam creates simple biological computers based on his own lymphocytes. Faced with orders from his nervous employer to destroy his work, he injects them into his own body, intending to smuggle the "noocytes" (as he calls them) out of the company and work on them elsewhere. Inside Ulam's body, the noocytes multiply and evolve rapidly, altering their own genetic material and quickly becoming self-aware. The nanoscale civilization they construct soon begins to transform Ulam, then others. The people who are infected start to find that genetic faults such as myopia and high blood pressure are fixed. Ulam's eyesight, posture, strength, and intelligence are all improved. The infected can even have conversations with their noocytes, some reporting that the cells seem to sing."
Blood Music

[Pete]

Breakthrough AI algorithm suffers from esteem issues, describes itself as 'Dense'.

[Donald]

“ Most of everything is junk. Right up until somebody identifies something useful about it.”

Um, no. You are assuming you know the answer to a scientific question when a large number of population geneticists apparently disagree with you. You can read the links or not and I am not going to act like I know the answer to the question when I am not a molecular biologist or population geneticist, but it is a real scientific debate that can’t be settled with internet one- liners. I do know the junk side has good arguments. I mean, try reading about the level of selection issue in evolutionary theory. One expects that there should be junk DNA— selection is not infinitely powerful.

Charles WT—. That’s a whole other issue. Being a Christian, I would be fine if it turned out that final causes are a real thing and science went too far down the path of mechanistic materialism. But it would take a heck of a lot of evidence and debate to shift that paradigm

[Donald]

Moran would hate what Charles said and also wj’s comment and my second paragraph. His complaint and that of others is that science journalism has not done a good job conveying thei complexity of the issue, and I think that is right.

Anyway, pulling out now, because of I am not an expert and people can browse Moran’s blog if they want.

[wj]

“ Most of everything is junk. Right up until somebody identifies something useful about it.”

My point, apparently badly expressed, is that the phrase "junk DNA" as initially used included a bunch of stuff which has subsequently been found to have actual impact on the development of the organism. Which suggests, at least to me, that some wariness is in order with applying the phrase to those parts of DNA for which no function is currently known.

I'm not disputing that there is likely some DNA which will eventually prove to not have a function for a particular organism. Just that our understanding of DNA is still a work in progress.

[Hartmut]

The solution should be easy (ethics concerns aside): Just take a human fertilized egg, make a copy and remove parts of the perceived junk in one. Then implant both and compare the results. Should not take more than let's say 2 decades per try. Repeat for next piece of junk starting from the same base cell. Doing enough in parallel and stretched out over let's say a century and we should have sorted out what is junk and what is not. Failed experiments can run for office on a GOP ticket.

[CharlesWT]

Since this article was written AI has increased the ability to reverse engineer biology.

"So what does that actually mean? To reverse engineer biology means applying the engineering concept of taking apart a process or mechanism in order to understand it and re-engineer it (perhaps in a new way)—and applying it to the biological world. That requires an intersection of two very different mindsets, which in some ways feel like fundamentally opposed approaches."
How to Reverse Engineer Biology

[ral]

[wj] I'm not disputing that there is likely some DNA which will eventually prove to not have a function for a particular organism. Just that our understanding of DNA is still a work in progress.

If by "work in progress" you will allow "just scratching the surface" I agree.

[wj]

Progress must start somewhere. :-)

[CaseyL]

"I can't figure out what it does, therefore, it doesn't do anything important."

The world is, unfortunately, full of people who operate along that line of thinking.

[JanieM]

This isn't specifically an open thread, but maybe this DKos piece on AI is close enough to the topic to fit in. After all, LLMs were mentioned a few comments back.

I am beyond delighted to see my own (rather surprising to me) boredom with AI transformed into so much entertainment.

[Hartmut]

As Rushdie put it the world runs on P2C2E.

[nous]

JanieM - That DKos piece really needs a dose of “The Revolution Will Not Be Served With Fries.” It’s a student favorite in my science fiction class.

https://www.lightspeedmagazine.com/fiction/the-revolution-will-not-be-served-with-fries/

Taco Bell won’t know what hit it after POS141 is through.

[novakant]

Chomsky died.

[novakant]

https://www.newstatesman.com/ideas/2024/06/the-noam-chomsky-i-knew

[ral]

There's more to you, Hartmut, than meets the blinking eye.

[JanieM]

novakant: New Statesman must have taken down that piece. Google gives the link, but the link goes to an error page. Various outlets say it hasn't been confirmed that he's dead...... One of them dated June 19.....

[nous]

Waiting for confirmation on the Chomsky news. Seems there is some question about the reports.

[novakant]

Wow, no kidding. Sorry.

I checked the Wikipedia page earlier and it gave a date of death and the phrasing was 'Chomsky was...'. Now, it's been changed back again.

[Donald]

The debate about function echoes the debate about the neutral theory of molecular evolution that started in the 60’s— the argument was that most of the genome is selectively neutral and evolution occurs in those sites at a rate equal to the mutation rate, whereas the part of the genome which was important evolved much more slowly, because mutations in those regions are likely to be deleterious and only rarely are beneficial.

The neutral theory won over most molecular evolutionists. If it is correct, it is hard to see how most of the genome could have a function, unless the function doesn’t depend on the sequence of nucleotides. Which is a little like saying that a book chapter has a function but one which doesn’t depend on the sequence of letters it contains.

https://www.nature.com/scitable/topicpage/neutral-theory-the-null-hypothesis-of-molecular-839/

[nous]

Just finished another term of grading, and of trying to balance fairness and optimism in grading. Was happy to have been able to get a few conscientious students over the bar to the next grade level for their final, only to have a few grade grubbers come in and beg for undeserved grade increases.

But since I am grumbling about grade grubbers, I should also say that GPA is a bullshit measure, made worse by GPA cutoffs for admissions and financial aid. I have a student from last quarter who is 0.1 beneath the threshold for transfer to another school at which they would get lower tuition and financial aid, and the only reason they had a lower GPA was because they were taking too many credits and stretched too thin for financial reasons.

If it were about education, the best grading system would be a simple non-passing/passing/pass-with-distinction grade. Anything else leads to misplaced priorities and grade inflation.

And this is why teachers eventually burn out.

[Snarki, child of Loki]

The business about Chomsky reminds me: a couple of decades back, an an acquaintance was watching a news report of a plane mishap in Taiwan, and thought they saw me among the passenger casualties. Asked about it on a group mailing list.

It was highly satisfying to reply that "reports of my demise are greatly exaggerated".

[wj]

the best grading system would be a simple non-passing/passing/pass-with-distinction grade. Anything else leads to misplaced priorities and grade inflation.

As far as I could ever tell, the main purpose of grades was to make it easier for administrators to make decisions about admissions. And, to a lesser degree, financial aid. Their relationship to actual learning is pretty coincidental. And, with grade inflation, they no longer even satisfy that administrative function