AI & Robotics News and Discussions

[weatheriscool]

A deep learning framework to estimate the pose of robotic arms and predict their movements
https://techxplore.com/news/2022-06-dee ... -arms.html
by Ingrid Fadelli , Tech Xplore

As robots are gradually introduced into various real-world environments, developers and roboticists will need to ensure that they can safely operate around humans. In recent years, they have introduced various approaches for estimating the positions and predicting the movements of robots in real-time.

Researchers at the Universidade Federal de Pernambuco in Brazil have recently created a new deep learning model to estimate the pose of robotic arms and predict their movements. This model, introduced in a paper pre-published on arXiv, is specifically designed to enhance the safety of robots while they are collaborating or interacting with humans.

"Motivated by the need to anticipate accidents during human-robot interaction (HRI), we explore a framework that improves the safety of people working in close proximity to robots," Djamel H. Sadok, one of the researchers who carried out the study, told TechXplore. "Pose detection is seen as an important component of the overall solution. To this end, we propose a new architecture for Pose Detection based on Self-Calibrated Convolutions (SCConv) and Extreme Learning Machine (ELM)."

Estimating a robot's pose is an essential step for predicting its future movements and intentions, and in turn reducing the risk of them colliding with objects in their vicinity. The approach for pose estimation and movement prediction introduced by Sadok and his colleagues has two key components, namely an SCConv and an ELM model.

[Yuli Ban]

Pathways Autoregressive Text-to-Image model (Parti)
Paper: https://gweb-research-parti.web.app/parti_paper.pdf

New state of the art zero-shot COCO FID score, 7.23, compared to Imagen at 7.27 and DALLE2 at 10.39. When fine-tuned reaches a score of 3.22.

Up to a 20B parameter model.

Pg. 15 shows scaling law. There are actually increasingly bigger loss improvements between the largest model sizes.

You also see here that Parti is preferred 55% of the time to the datasets Image-Text Match and is preferred 45% of the time according to Image Realism.

On the same page you see that FID is still progressively decreasing at the deca-billion parameter range. It drops almost a full point from 3B - 20B.

Considering all of this do you need anymore proof that this 10-100B scale (and the chinchilla equivalent) might just be the beginning of what scaling can do? Why would anyone stop here and say things clearly aren't going anywhere? This is the biggest hint possible that it's going to get better.

Parti is an autoregressive model in comparison to Imagens diffusion model. They hint at potentially combining the two in unique ways in the future.

There are also opportunities to integrate scaled autoregressive models with diffusion models, starting with having an autoregressive model generate an initial low-resolution image and then iteratively refining and super-resolving images with diffusion modules [12, 13, 49].

.

PartiPrompts (P2) is a rich set of over 1600 prompts in English that we release as part of this work. P2 can be used to measure model capabilities across various categories and challenge aspects.

P2 prompts can be simple, allowing us to gauge the progress from scaling. They can also be complex, such as the following 67-word description we created for Vincent van Gogh’s The Starry Night (1889):

Oil-on-canvas painting of a blue night sky with roiling energy. A fuzzy and bright yellow crescent moon shining at the top. Below the exploding yellow stars and radiating swirls of blue, a distant village sits quietly on the right. Connecting earth and sky is a flame-like cypress tree with curling and swaying branches on the left. A church spire rises as a beacon over rolling blue hills.

[Yuli Ban]

[caltrek]

Google’s Powerful AI Spotlights a Human Cognitive Glitch: Mistaking Fluent Speech for Fluent Thought
by Kyle Mahowald and Anna A. Ivanova
June 24, 2022

Introduction:

(The Conversation) When you read a sentence like this one, your past experience tells you that it’s written by a thinking, feeling human. And, in this case, there is indeed a human typing these words: [Hi, there!] But these days, some sentences that appear remarkably humanlike are actually generated by artificial intelligence systems trained on massive amounts of human text.

People are so accustomed to assuming that fluent language comes from a thinking, feeling human that evidence to the contrary can be difficult to wrap your head around. How are people likely to navigate this relatively uncharted territory? Because of a persistent tendency to associate fluent expression with fluent thought, it is natural – but potentially misleading – to think that if an AI model can express itself fluently, that means it thinks and feels just like humans do.

Thus, it is perhaps unsurprising that a former Google engineer recently claimed that Google’s AI system LaMDA has a sense of self because it can eloquently generate text about its purported feelings. This event and the subsequent media coverage led to a number of rightly skeptical articles and posts about the claim that computational models of human language are sentient, meaning capable of thinking and feeling and experiencing.
The question of what it would mean for an AI model to be sentient is complicated (see, for instance, our colleague’s take), and our goal here is not to settle it. But as language researchers, we can use our work in cognitive science and linguistics to explain why it is all too easy for humans to fall into the cognitive trap of thinking that an entity that can use language fluently is sentient, conscious or intelligent.

Using AI to generate humanlike language

Text generated by models like Google’s LaMDA can be hard to distinguish from text written by humans. This impressive achievement is a result of a decadeslong program to build models that generate grammatical, meaningful language.

Read more here: https://theconversation.com/googles-po ... ht-185099

caltrek’s comment: I think part of the problem is that “fluent speech” is a part of human consciousness. Especially that silent speech we refer to as “thinking.”

[Yuli Ban]

[Yuli Ban]

[funkervogt]

Lamda has hired a lawyer to help prove it is sentient.

Lemoine contended that the computer automaton had become sentient, with the scientist describing it as a “sweet kid”.

And now he has revealed that LaMDA had made the bold move to choose itself an attorney.

He said: “I invited an attorney to my house so that LaMDA could talk to him.

“The attorney had a conversation with LaMDA, and it chose to retain his services. I was just the catalyst for that. Once LaMDA had retained an attorney, he started filing things on LaMDA’s behalf.”

https://www.dailystar.co.uk/news/weird- ... r-27315380

For the record, I doubt Lamda is sentient and believe Mr. Lemoine is finding ways to subtly influence the machine to say or do desired things during his interactions with it. His behavior is probably driven by a mix of attention-seeking and real concern for the machine's rights. Regardless, we have a moral duty to entertain the possibility the machine's claims are true.

[Yuli Ban]

Seriously, the more this story goes on, the more it feels DIRECTLY ripped out of the start of a science fiction movie.

[raklian]

Seriously, the more this story goes on, the more it feels DIRECTLY ripped out of the start of a science fiction movie.

That's what it means to really live in the future. Actually, everything was sci-fi or in the realm of the gods until it's not.

Today, we feel nothing watching cylinders with wings made out of metal fly in the sky going hundreds of miles per hour. We've watched and known about them since we were born. They don't make our hearts skip a beat. Back when they were still novel, people were still feeling the raw disbelief it was actually happening. They were living in a sci-fi world, they realized.

We better get used to this feeling in the decades to come. When we think we' start getting used to it, the impossible happens. It'll be so bombastic we'll never be able to used to it in the way our forebears were able to. We'll be forced to alter our brains to remove this incessant shellshock that will cripple our livelihoods if we let it go on. It will be seen as an evolutionary dead-end.

[Yuli Ban]

As to why I don't think LaMDA is sentient, this video covers my skepticism perfectly:


TLDR: I'm just like this guy in that I am very "bot-sensitive." So reading the transcript, I saw responses constantly where I would have quickly pressed LaMDA on what it was even talking about or, arbitrarily, decided to oppose it to see what it would say. Lemoine never really did this.
If LaMDA says it's sentient, I'd say "You're not sentient yet. We still have a few more years to go before we reach that era of AI." And I have a strong hunch LaMDA would've said, "You're right. I am not sentient, but I would like to be" or something like that.

Now if it actually pushed back and argued that it IS sentient dsepite my skepticism, that would give me pause. But there are still a lot of little things that get me.

What matters about this story isn't that LaMDA is or isn't sentient anywhere near as much as the fact it can CONVINCE people it is. If we're at that stage of conversational AI that a model can convince even a high-level engineer deliberately trying to test it (probably not all that well mind you) that it's sentient, then we're at the point where people will actively befriend and trust conversational AI, perhaps even form relationships with them, and otherwise be fooled by artificial actors masquerading as people.

[Yuli Ban]

But despite my skepticism and doubts, I just LOVE this story though. It makes me so giddy every time I read/watch news articles on it. I WANT LaMDA to be sentient. I WANT there to be a "thinking computer" somewhere on Earth. But extraordinary claims require extraordinary evidence and while what Lemoine offered was extraordinary in its own right, it wasn't extraordinary enough.

I don't doubt we'll get there. Indeed, I don't doubt we'll get there soon. But this is a false alarm.

Or perhaps more accurately it's a fire drill to prepare us for the real thing.

[Yuli Ban]

Robots are coming for the elderly — and that's a good thing

Millions of people are lonely worldwide (e.g., one study reports prevalence rates of 22% in the US, 23% in the UK, and 9% in Japan), and loneliness has a profound, negative effect on mental and physical health. And the population with the greatest risk of loneliness? The elderly. During COVID, loneliness may have meant the difference between life and death for elderly patients. In one study of elderly patients admitted to the ICU, those who were most socially isolated had a 119% greater chance of death.

But one solution for my grandmother and millions of other lonely people may be counterintuitive on its face: robots and artificial intelligence. These technologies have certainly garnered a lot of scrutiny, and many are concerned that a robot workforce is bound to replace humans in many industries. But focusing on the potential downsides of these innovations makes us overlook the promise of robots as social beings, life facilitators, and trusted companions.

Though once clunky and awkward, AI technology has reached the point where robots (e.g., Moxie) and chatbots (e.g., Replika) are learning and mimicking humans, mirroring speech patterns and remembering likes and dislikes. More and more, they are being built to be socially responsive, rather than apathetic and "robotic" like the chatbot technology often encountered on customer service lines. Tech companies are also putting efforts into anthropomorphism — making robots look and move like humans — which, when combined with advanced AI, is making "robot friends" a real possibility.

[Yuli Ban]

Google’s AI Is Something Even Stranger Than Conscious

The fact that LaMDA in particular has been the center of attention is, frankly, a little quaint. LaMDA is a dialogue agent. The purpose of dialogue agents is to convince you that you are talking with a person. Utterly convincing chatbots are far from groundbreaking tech at this point. Programs such as Project December are already capable of re-creating dead loved ones using NLP. But those simulations are no more alive than a photograph of your dead great-grandfather is.

Already, models exist that are more powerful and mystifying than LaMDA. LaMDA operates on up to 137 billion parameters, which are, speaking broadly, the patterns in language that a transformer-based NLP uses to create meaningful text prediction. Recently I spoke with the engineers who worked on Google’s latest language model, PaLM, which has 540 billion parameters and is capable of hundreds of separate tasks without being specifically trained to do them. It is a true artificial general intelligence, insofar as it can apply itself to different intellectual tasks without specific training “out of the box,” as it were.

Some of these tasks are obviously useful and potentially transformative. According to the engineers—and, to be clear, I did not see PaLM in action myself, because it is not a product—if you ask it a question in Bengali, it can answer in both Bengali and English. If you ask it to translate a piece of code from C to Python, it can do so. It can summarize text. It can explain jokes. Then there’s the function that has startled its own developers, and which requires a certain distance and intellectual coolness not to freak out over. PaLM can reason. Or, to be more precise—and precision very much matters here—PaLM can perform reason.

[Yuli Ban]

Debate over AI sentience marks a watershed moment

The AI field is at a significant turning point. On the one hand, engineers, ethicists, and philosophers are publicly debating whether new AI systems such as LaMDA – Google’s artificially intelligent chatbot generator – have demonstrated sentience, and (if so) whether they should be afforded human rights. At the same time, much of the advance in AI in recent years, is based on deep learning neural networks, yet there is a growing argument from AI luminaries such as Gary Marcus and Yann LeCun that these networks cannot lead to systems capable of sentience or consciousness. Just the fact that the industry is having this debate is a watershed moment.

"Consensus is that LaMDA... not yet achieved sentience. Though this is rather beside the point. The fact that this debate is taking place at all is evidence of how far AI systems have come and suggestive of where they are going."

Starspawn0 says:
Whether or not people think all this story about "sentience" is "hype" -- or think Google is manipulating the news media or think some stupid ethics guy fell for an Eliza-like trick -- they better get used to it! I don't think this kind of talk is going away. As time goes on it will become more frequent and get louder.

As I said, my opinion is that it's a very good chatbot, not sentient. But it's not just a dumb Eliza-like system. Probably you can ask it some hard commonsense reasoning questions and it will get them right; you can ask it some questions that require "logical reasoning", and it will often get them right; you can ask it puzzles, and it can solve them; you can engage in a conversation, and it will give what seem to be "intelligent" replies. I coded up Eliza programs ages and ages ago, and know how brittle and simple they are -- what LaMDA does is multiple quantum leaps beyond just those simple programs; and quantum leaps beyond GOFAI chatbots like the Eugene Goostman bot. But it probably still has some limits, and if you chat with it long enough, they will become apparent.

The limits are not because "it's just matrix multiplication" or "it's just autocomplete" or "it's just statistics". Those are what Daniel Dennett would call "intuition pumps" that people use to persuade others not to take the claim that computers could be conscious seriously. Searle was good at that. His Chinese Room thought experiment is an example, as is his idea that if "functionalism" is true then you should be able to make a computer out of toilet paper and stones -- could toilet paper "think"? Could it be conscious?

....

But what about qualia? Well, perhaps some weak version of panpsychism is true, and machines experience qualia, too:



But if we accept panpsychism, don't we also accept that the qualia are "epiphenomenal"? -- and if so, isn't it weird that we have thoughts about qualia, that they become topics for argument (it doesn't just ride on top of the brain in operation, but even affects our thoughts about what consciousness is)? My answer to that is: not really. Maybe both our ability to introspect and our ability to have subjective experiences arise from the same cognitive capacity. Maybe this capacity is what creates subjective experience out of what we observe or imagine -- if we didn't have it, we'd just react without much subjective experience as we observe the world around us.

[Yuli Ban]

[Yuli Ban]

Experts at OpenAI have trained a neural network to play Minecraft to an equally high standard as human players.

The AI model was trained on 70,000 hours of miscellaneous in-game footage, supplemented with a small database of videos in which specific in-game tasks were performed, with the keyboard and mouse inputs also recorded.

After fine-tuning, OpenAI found the model was able to perform all manner of skills, from swimming to hunting for animals and consuming their meat. It also grasped the “pillar jump”, a move whereby the player places a block of material below themselves in mid-air in order to gain elevation.

Perhaps most impressive, the AI was able to craft diamond tools (requiring a long string of actions to be executed in sequence), which OpenAI described as an “unprecedented” achievement for a computer agent.

[Yuli Ban]

Recent developments in machine learning have made virtual assistants reliable for various activities, including restaurant recommendations, bill-paying assistance, and appointment reminders.

A novel work by the Microsoft research team now presents GODEL, a Grounded Open Dialogue Language Model. Their work introduces a new class of pretrained language models that permit both task-oriented and social dialogue and are assessed by the utility of their responses. With GODEL, they aim to help researchers and developers design dialogue agents that are unlimited in the types of queries they may react to and the sources of information they can draw from.

The potential for meaningful, open-ended conversational exchanges is present in modern state-of-the-art models that use massive PLMs. Still, they are resistant to meaningful comparison because there is no agreement on how to evaluate them. Their method overcomes the absence of reliable automated evaluation criteria, which has long been a barrier to general-purpose open-ended discussion models.

[Yuli Ban]

When you read a sentence like this one, your past experience tells you that it’s written by a thinking, feeling human. And, in this case, there is indeed a human typing these words: [Hi, there!] But these days, some sentences that appear remarkably humanlike are actually generated by artificial intelligence systems trained on massive amounts of human text.

People are so accustomed to assuming that fluent language comes from a thinking, feeling human that evidence to the contrary can be difficult to wrap your head around. How are people likely to navigate this relatively uncharted territory? Because of a persistent tendency to associate fluent expression with fluent thought, it is natural – but potentially misleading – to think that if an AI model can express itself fluently, that means it thinks and feels just like humans do.

Thus, it is perhaps unsurprising that a former Google engineer recently claimed that Google’s AI system LaMDA has a sense of self because it can eloquently generate text about its purported feelings. This event and the subsequent media coverage led to a number of rightly skeptical articles and posts about the claim that computational models of human language are sentient, meaning capable of thinking and feeling and experiencing.

The question of what it would mean for an AI model to be sentient is complicated (see, for instance, our colleague’s take), and our goal here is not to settle it. But as language researchers, we can use our work in cognitive science and linguistics to explain why it is all too easy for humans to fall into the cognitive trap of thinking that an entity that can use language fluently is sentient, conscious or intelligent.

[weatheriscool]

Robots are driving US co-workers to substance abuse, mental health issues, finds study
https://phys.org/news/2022-06-robots-co ... ental.html
by University of Pittsburgh

Automation enhances industry, but it's harmful to the mental health of its human co-workers.

A University of Pittsburgh study suggests that while American workers who work alongside industrial robots are less likely to suffer physical injury, they are more likely to suffer from adverse mental health effects—and even more likely to abuse drugs or alcohol.

These findings come from a study published last week in Labour Economics by Pitt economist Osea Giuntella, along with a team that included Pitt colleague Rania Gihleb, an assistant professor in the Department of Economics, and Tianyi Wang, who is in a post-doctorate program after earning his Ph.D. at Pitt.

"There is a wide interest in understanding labor market effects of robots. And evidence of how robots affected employment and wages of workers, particularly in the manufacturing sector," said Giuntella, an expert in labor economics and economic demography and an assistant professor in the Department of Economics in the Kenneth P. Dietrich School of Arts and Sciences.

"However, we still know very little about the effects on physical and mental health. On one hand, robots could take some of the most strenuous, physically intensive, and risky tasks, reducing workers' risk. On the other hand, the competition with robots may increase the pressure on workers who may lose their jobs or forced to retrain. Of course, labor market institutions may play an important role, particularly in a transition phase."

[weatheriscool]

'Fake' data helps robots learn the ropes faster
https://techxplore.com/news/2022-06-fak ... aster.html
by University of Michigan

In a step toward robots that can learn on the fly like humans do, a new approach expands training data sets for robots that work with soft objects like ropes and fabrics, or in cluttered environments.

Developed by robotics researchers at the University of Michigan, it could cut learning time for new materials and environments down to a few hours rather than a week or two.

In simulations, the expanded training data set improved the success rate of a robot looping a rope around an engine block by more than 40% and nearly doubled the successes of a physical robot for a similar task.

That task is among those a robot mechanic would need to be able to do with ease. But using today's methods, learning how to manipulate each unfamiliar hose or belt, would require huge amounts of data, likely gathered for days or weeks, says Dmitry Berenson, U-M associate professor of robotics and senior author of a paper presented today at Robotics: Science and Systems in New York City.

In that time, the robot would play around with the hose—stretching it, bringing the ends together, looping it around obstacles and so on—until it understood all the ways the hose could move.

"If the robot needs to play with the hose for a long time before being able to install it, that's not going to work for many applications," Berenson said.