Artificial Intelligence & Robotics

[Mel81x+]

23 minutes ago, nudge said:

Fascinating. The idea of an omniscient director AI is very cool.  I also love how the Alien unlocks its behaviour trees based on the player's behaviourial patterns and then adjusts its own actions accordingly. Very neat.

Also I would poop my pants if I had to play it  

I played it and loved it. I highly recommend it to anyone to either watch or play. It's in the upper echelons of survival-gaming imo. The way that thing shows up out of nowhere to make you run and hide is unreal.

I am trying to find the RE2:REMAKE cut video on how Mr. X teleports and goes searching for you as well. That's a bit more straightforward but what Capcom did to make him chase you through rooms is just unreal. Shame they didnt' bring it to the more recent RE3:REMAKE.

Edited April 14, 2020 by Mel81x

[nudge]

Didn't know where else to put it, but this is quite an interesting little guide on how to read and analyse DNA on a computer using Python. https://biospace.xyz/posts/reading-dna-on-computers/

On a similar topic, I recently discovered Rosalind, a platform to learn bioinformatics, algorithms and programming through problem solving. Quite fun. http://rosalind.info/

[nudge]

https://www.technologyreview.com/2020/02/05/349131/an-algorithm-that-can-spot-cause-and-effect-could-supercharge-medical-ai/

@Mel81x The title is a bit misleading, but it's pretty interesting nonetheless...

[Mel81x+]

4 minutes ago, nudge said:

https://www.technologyreview.com/2020/02/05/349131/an-algorithm-that-can-spot-cause-and-effect-could-supercharge-medical-ai/

@Mel81x The title is a bit misleading, but it's pretty interesting nonetheless...

It is interesting but I don't know how much I buy into the casual linkage of data even if you use quantum computing to treat the data as conversations versus you know, actual data. The idea then that a chat-bot can actually diagnose (and with accuracy) symptoms and then offer a remedy course is definitely something worth looking into but I still think so much data is missing even in that sector as is pointed out when peer-reviews applauded the concept but said it could also go horribly wrong. 

What I do like about all this is how far we've come in analyzing data and more importantly how inflection points are gathered from information sources that are totally dissonant from each other. Used to be you'd be asked to create data repositories and they needed linkage but these days you dont even need that.

[Mel81x+]

https://techcrunch.com/2020/07/03/we-need-a-new-field-of-ai-to-combat-racial-bias/

@nudge and @Eco this is a good read for you'll as I know both of you'll are working with Python and learning data/security. If you're interested in reading the NIST publication its very good as well.

https://www.nist.gov/news-events/news/2019/12/nist-study-evaluates-effects-race-age-sex-face-recognition-software

Additionally, take a look at this and how algorithms started discriminating against women when they started running. 

https://slate.com/business/2018/10/amazon-artificial-intelligence-hiring-discrimination-women.html

I know this is probably not the place to post this but its all AI based and I suppose we can all take a learning away that the wrong data handling will eventually lead to some pretty undesirable results.

[nudge]

Thanks for the links @Mel81x, will properly read it after all the racing today. 

I remember reading Karen Hao's interactive feature on MIT Technology Review that was related to the topic a while ago, there they had a predictive algorithm called COMPAS, which is used as a risk assessment tool in US courtrooms to decide whether the defendant should be released from jail before their trial or not. The feature had you tweak the algorithm based on certain scenarios to make it "fairer", in order to demonstrate the issues with it and the biases involved; that was an interesting experiment. 

https://www.technologyreview.com/2019/10/17/75285/ai-fairer-than-judge-criminal-risk-assessment-algorithm/

[Mel81x+]

1 hour ago, nudge said:

Thanks for the links @Mel81x, will properly read it after all the racing today. 

I remember reading Karen Hao's interactive feature on MIT Technology Review that was related to the topic a while ago, there they had a predictive algorithm called COMPAS, which is used as a risk assessment tool in US courtrooms to decide whether the defendant should be released from jail before their trial or not. The feature had you tweak the algorithm based on certain scenarios to make it "fairer", in order to demonstrate the issues with it and the biases involved; that was an interesting experiment. 

https://www.technologyreview.com/2019/10/17/75285/ai-fairer-than-judge-criminal-risk-assessment-algorithm/

I dont know how much I trust an algorithm that isn't given all the data it needs to make decisions like the one presented by Karen (sorry I spent 20 seconds laughing at the name and its connotation with what we're reading but moving on). The problem with the algorithm and its premise is that the system used for judging doesn't account for other factors like how the defendants got to the crime, were they in groups, etc and that is also a consideration for how the release/jailing system works. When it comes to the other issue of credit scores thats a far more clearer one. Someone who can't maintain the basic income for recovery over a period shouldn't be allowed credit but there also there aren't enough inflection points in the data to help make the decision. They account for things like inflation and market rates but the algorithms are mostly aligned towards risk avoidance rather than credit assignment so that might be a failure we need to consider.

[nudge]

On 04/07/2020 at 08:13, Mel81x said:

https://techcrunch.com/2020/07/03/we-need-a-new-field-of-ai-to-combat-racial-bias/

@nudge and @Eco this is a good read for you'll as I know both of you'll are working with Python and learning data/security. If you're interested in reading the NIST publication its very good as well.

https://www.nist.gov/news-events/news/2019/12/nist-study-evaluates-effects-race-age-sex-face-recognition-software

Additionally, take a look at this and how algorithms started discriminating against women when they started running. 

https://slate.com/business/2018/10/amazon-artificial-intelligence-hiring-discrimination-women.html

I know this is probably not the place to post this but its all AI based and I suppose we can all take a learning away that the wrong data handling will eventually lead to some pretty undesirable results.

That was a good read. So in essence, what has been happening so far in many cases is that the AI didn't remove the bias, but rather automated it   

On a more serious note, when it comes to facial recognition, it seems like a combination of the quality (or lack thereof) of algorithm training data and physical properties of the universe. I think it's safe to assume that facial recognition is likely to be more accurate on people with lighter skin tones, because more reflected light means more data allowing the software to make a match more accurately, whereas it is more error-prone on people with darker skin? 

Also regarding Amazon, it's not surprising their recruitment AI tool turned out to be crap. I mean, even their regular internal search algorithm is fucking horrible when you are trying to find some product on their website! 

[CaaC (John)]

I was not sure if this belongs in here as I was tossing up putting it into Science & Environment forum which the BBC link I got it from classed it as the latter or post this in the space forum as NASA was mentioned so feel free @Eco to move it into the appropriate forum if you must.  

Science & Environment

Robotic scientists will 'speed up discovery'

Scientists at the University of Liverpool have unveiled a robotic colleague that has been working non-stop in their lab throughout lockdown.

The £100,000 programmable researcher learns from its results to refine its experiments.

"It can work autonomously, so I can run experiments from home," explained Benjamin Burger, one of the developers.

Such technology could make scientific discovery "a thousand times faster", scientists say.

A new report by the Royal Society of Chemistry lays out a "post-Covid national research strategy", using robotics, artificial intelligence and advanced computing as part of a suite of technologies that "must be urgently embraced" to help socially distancing scientists continue their search for solutions to global challenges.

• Nasa humanoid starts work in UK lab
• Will coronavirus speed up the use of robots?

Robo-chemist

The robotic scientist is currently embarking on a series of tests to find a catalyst that could speed up the reaction that takes place inside solar cells.

But it could, according to Prof Andy Cooper, the materials scientist who has put the robot to work in his lab, be used in the fight against Covid-19.

"We've had a lot of interest [in the robot] from labs that are doing Covid research," he told BBC News.

"Covid, climate change - there are lots of problems that really need international co-operation. So our vision is we might have robots like this all across the world connected by a centralised brain which can be anywhere. We haven't done that yet - this is the first example - but that's absolutely what we'd like to do."

Socially distant science

Today, in a world where scientists also need to limit their time in the lab and maintain social distance from each other, the robo-scientist has come into its own.

"It doesn't get bored, doesn't get tired, works around the clock and doesn't need holidays," Dr Burger joked.

On a more serious note, he said that the robot had transformed the speed at which he could carry out research. "It can easily go through thousands of samples," he said, "so it frees up my time to focus on innovation and new solutions."

Like robotics designed for research in Space, machines like this could also take on riskier experiments - in harsher laboratory environments or using more toxic substances.

That, according to Deirdre Black, head of research and innovation at the Royal Society of Chemistry, is why UK science needs to build new technologies into its infrastructure.

"This is about human beings harnessing all of these digital technologies so that they can go faster - discover and innovate faster and explore bigger and more complex problems, like decarbonisation, preventing and treating disease, and making our air cleaner," she told BBC News.

So does this mean that while many scientists have been in lockdown, the machines have come to take their jobs?

"Absolutely not," said Dr Black. "Science will always need people".

https://www.bbc.co.uk/news/science-environment-53029854

 

Edited July 6, 2020 by CaaC (John)

[Mel81x+]

On 06/07/2020 at 02:23, nudge said:

That was a good read. So in essence, what has been happening so far in many cases is that the AI didn't remove the bias, but rather automated it   

On a more serious note, when it comes to facial recognition, it seems like a combination of the quality (or lack thereof) of algorithm training data and physical properties of the universe. I think it's safe to assume that facial recognition is likely to be more accurate on people with lighter skin tones, because more reflected light means more data allowing the software to make a match more accurately, whereas it is more error-prone on people with darker skin? 

Also regarding Amazon, it's not surprising their recruitment AI tool turned out to be crap. I mean, even their regular internal search algorithm is fucking horrible when you are trying to find some product on their website! 

I think the AI pushed the bias narrative because it figured it offered the most balancing. Hard to believe they didn't let it try and learn more to adapt to the way feedback may have corrected it? I mean if people can offer feedback to it then maybe it gets better which is the whole point of an AI anyways. 

[nudge]

Thought this was quite an interesting read about the key impediments to building machines that understand meaning, what meaning would look like for artificial intelligence, and how far understanding is necessary for artificially intelligent machines to approach human-level abilities in language, perception, and reasoning. Nothing ground breaking, but a nice if somewhat philosophical discourse.

https://bdtechtalks.com/2020/07/13/ai-barrier-meaning-understanding/

@Mel81x @Eco

[nudge]

@Mel81x, what are your thoughts on GPT-3? I spent a good chunk of my weekend exploring the available samples of its capabilities, and must say I find it very impressive for what it is... The bloody thing can write workable code, do machine translation, answer questions, one-shot learn novel words, write passable poetry, folk tales, generate news articles and pretty much anything else, in a much, much better, more human-like quality than its predecessor - which is of course not surprising, given the vast amount on data it was trained on (half a trillion words) and the number of parameters used that's even hard to comprehend (175 billion!!!), but nevertheless, the results are mind blowing; especially its ability to meta-learn, generalise to other tasks with barely any or none finetuning and solve tasks it has never encountered. One area where it impresses me in particular is the AI Dungeon "Dragon" model upgrade; the level of coherence in the world, story and dynamic characters generated by the GPT-3 are nothing short of amazing...

 

Also this is some next level Westworld inspired shite:

I'm getting serious Dolores vibes here... Have you ever seen anything so full of splendor?  

[Mel81x+]

9 hours ago, nudge said:

@Mel81x, what are your thoughts on GPT-3? I spent a good chunk of my weekend exploring the available samples of its capabilities, and must say I find it very impressive for what it is... The bloody thing can write workable code, do machine translation, answer questions, one-shot learn novel words, write passable poetry, folk tales, generate news articles and pretty much anything else, in a much, much better, more human-like quality than its predecessor - which is of course not surprising, given the vast amount on data it was trained on (half a trillion words) and the number of parameters used that's even hard to comprehend (175 billion!!!), but nevertheless, the results are mind blowing; especially its ability to meta-learn, generalise to other tasks with barely any or none finetuning and solve tasks it has never encountered. One area where it impresses me in particular is the AI Dungeon "Dragon" model upgrade; the level of coherence in the world, story and dynamic characters generated by the GPT-3 are nothing short of amazing...

 

Also this is some next level Westworld inspired shite:

I'm getting serious Dolores vibes here... Have you ever seen anything so full of splendor?  

I get HAL vibes from it and I also get the feeling it thinks I am in its way and would put me in a box and suffocate me till it finds a more subservient entity to do its bidding haha.

Putting all that aside I think its wonderful what the folks at OpenAI are achieving and more importantly the parts where it is able to do machine language translations on its own with some time and not so complex issues. The demos were very simple and the big push for NLP style questions but its working to a certain degree and it will only get better. As developers we may be out of work pretty soon if this keeps up but I figure the market has been shifting for years now and we're at a stage where you really can build something on the internet with no knowledge of computer languages or the platform you're building it on (some website builders do this today anyways). The next logical step here isn't pondering our existence but telling it jokes and seeing how it comprehends them in the context of the joke being told and then we should be both worried and very excited.

[CaaC (John)]

Could the internet become conscious?

An old belief system called ‘panpsychism’ suggests that something resembling a mind or consciousness exists in all physical things. In recent years, researchers such as neuroscientist Christof Koch have tried to update these ideas, saying that if there are enough connections between elements – like synapses in the brain – then consciousness may naturally start to form.

The average human brain has 86 billion neurons, with some 100 trillion connections between them – does this mean that if something else has the same number of connections, it will also become conscious? Some believe so.

Like our brains, the internet is a massively connected entity, made up of computers, mobile devices and the ‘Internet of Things’ (IoT) – smart, networked objects including household appliances, wearable gadgets, vehicles and even entire factories. According to one estimate, the number of internet-connected devices will reach 125 billion worldwide by 2030. So the connectivity of the internet seems to be fast approaching that of our brain.

But how would we know if the internet had become conscious? An internet that’s as connected as the brain might process information as quickly as we do, but that doesn’t mean it’s going to ‘wake up’ or become self-aware. There are big differences between the two systems: our sense of consciousness is created by brains that have evolved over millions of years, whereas the internet is a human-designed network that’s been around for a few decades.

Ultimately, consciousness is still such a poorly understood subject that we won’t have a definitive answer to your question until we figure out how to measure consciousness in the first place.

Science Focus

[AMG]

 

[nudge]

@Mel81x Going to a robotics and futurology exhibition next week. Basically a demonstration of the newest robots, Tesla coils/Faraday cage show, virtual reality, holograms, bionics, etc. Was given a short demo by Boston Dynamics Spot the Robot Dog today, fucking impressive. Really looking forward to it...

[Mel81x+]

1 hour ago, nudge said:

@Mel81x Going to a robotics and futurology exhibition next week. Basically a demonstration of the newest robots, Tesla coils/Faraday cage show, virtual reality, holograms, bionics, etc. Was given a short demo by Boston Dynamics Spot the Robot Dog today, fucking impressive. Really looking forward to it...

If you're allowed to take pics I'd like to see whats up and coming.

[nudge]

5 minutes ago, Mel81x said:

If you're allowed to take pics I'd like to see whats up and coming.

Will do! 

[Mel81x+]

2 minutes ago, nudge said:

Will do! 

Oh and no camera phones please. Standards must be maintained. 

[nudge]

1 minute ago, Mel81x said:

Oh and no camera phones please. Standards must be maintained. 

Haha I planned to just take my Pentax ME Super and shoot on film, but now that you asked, I'll bring my DSLR too  

[Mel81x+]

3 minutes ago, nudge said:

Haha I planned to just take my Pentax ME Super and shoot on film, but now that you asked, I'll bring my DSLR too  

Nah enjoy the show and if you're shooting on film I'd still like to see the results.

[CaaC (John)]

If this is in the wrong forum please be welcome to move it Admin/Mods 

Robots and magnetic soap: scientists rethink oil spill clean-ups

© Photograph: Laura Morosoli/EPA Community-made floating barriers of straw and fabric to contain oil washing ashore off the south-east coast of Mauritius.

Special sponges, magnetic soap and autonomous robots are among the latest wave of inventions aimed at tackling oil spills.

Incidents such as the tanker stranding in Mauritius in August can devastate the environment and threaten communities who rely on the sea or tourism for their livelihoods. They often take months or years to clean up.

Prof Vinayak Dravid and Vikas Nandwana, a PhD student, believe the humble sponge could be the key to fighting oil spillages in the oceans.

“As long as fossil fuels are still in circulation, there will always be oil spills,” said Dravid. “We wanted to create a technology which can make cleaning oil spillages much easier and, more importantly, much cleaner and safer for the environment.”

Dravid and Nanwana, both at Northwestern University, Illinois, have developed a sponge capable of selectively soaking up oil spills found in ocean water.

The sponge has a coating of magnetic nanostructures and a carbon-based surface that attracts oil and resists water. It binds to the oil molecules, capturing and storing the oil until it is squeezed out, and can absorb more than 30 times its weight in oil.

To mimic natural waves, the team put the sponge on a shaker submerged in water. Even after vigorous shaking, the sponge released less than 1% of the oil it had absorbed back into the water.

“We are confident this sponge could help save the fauna and flora on Mauritius’s shores,” said Dravid. “Its biggest advantage is how the sponge can be made into balls to be left in an oil-laden sand or soil pit. The oil will get absorbed in days, and the sponge can be reused.”

Last month three sailors died and one went missing after their tugboat capsized while clearing the Mauritius oil spill.

In June, Russia declared a state of emergency after 20,000 tonnes of diesel fuel spilt from a power plant in Norilsk into the Ambarnaya River, a main connector to the Arctic Ocean. –

A decade on from the catastrophic Deepwater Horizon spill in the US, oil production has now outstripped pre-accident levels, raising safety concerns within the Barking Spider administration.

Nandwana likened the sponge to a Swiss army knife of the future. “This is a nanoscale solution to a gigaton problem,” he said. “The oil recovered from our sponges can be sold back to who is responsible for the spillage. We hope after an oil recovery the sponge can be recycled and burned into soot for use as an electrode for lithium-ion batteries.”

Magnetic soap is another unusual option for cleaning up oil spills. Scientists from Bristol University have created a soap composed of dissolvable iron-rich salts, which responds to magnetic fields when placed in solution. It is hoped the soap could one day revolutionise industrial cleaning operations and environmental clean-up procedures.

Prof Julian Eastoe, part of the team who developed the magnetic soap, hopes one day the soap may move into commercial usage. “The potential applications of magnetic surfactants are huge,” he said. “Their responsiveness to external stimuli allows a range of properties, such as their electrical conductivity, melting point, the size and shape, and how readily it dissolves in water, to be altered by a simple magnetic on and off switch.”

Self-driving navigating robots may be another solution. Carlo Ratti, the director of the MIT Senseable City Lab, pioneered the Sea Swarm robot in response to the Deepwater Horizon oil spill.

“This technology was conceived to be deployed anywhere it was needed –– in oceans, rivers, or seas. It’s mostly useful in deltas and zigzagging coastlines where most of the other technologies fail,” Ratti said. “We all need to be accountable for the environment. Some accidents are still bound to occur, so we still need to develop mitigation or cleaning strategies.”

The robot can hold up to 20 times its weight in oil. It works as a fleet or “swarm” of vehicles communicating their location through GPS. Each robot consists of a head covered by a layer of photovoltaic cells and a conveyor belt covered with tiny wires. The cells generate enough electricity to propel the vehicles forward for several weeks at a time.

“As the head moves through the water, the conveyor belt constantly rotates and sucks up pollution,” Ratti said. “The nanowire-covered belt is then compressed to remove the oil. As the clean part of the belt comes out of the head, it immediately begins absorbing oil, making the collection process seamless.

After each use, the fabric can be heated to remove the oil. Once the oil is removed, the nanowire mesh can be recycled again and again. In terms of end of life, they are trackable, so they can be collected and disposed of.”

https://www.msn.com/en-gb/news/world/robots-and-magnetic-soap-scientists-rethink-oil-spill-clean-ups/ar-BB19wxTR?li=AAnZ9Ug&ocid=mailsignout

[nudge]

@Mel81x thought you might enjoy this: https://towardsdatascience.com/how-to-create-an-agent-based-simulation-model-37bd7b4b0da7

This probably isn't the right thread, wasn't sure where else to post it though; basically one fella created a computer model to simulate passenger happiness based on the conditions on the Snowpiercer. Nothing groundbreaking, but an interesting idea and experiment  

[Mel81x+]

5 hours ago, nudge said:

@Mel81x thought you might enjoy this: https://towardsdatascience.com/how-to-create-an-agent-based-simulation-model-37bd7b4b0da7

This probably isn't the right thread, wasn't sure where else to post it though; basically one fella created a computer model to simulate passenger happiness based on the conditions on the Snowpiercer. Nothing groundbreaking, but an interesting idea and experiment  

Love it. I'll trade you an equally entertaining piece of literature on AI and in particular a case-study in law for the UK.

https://www.pinsentmasons.com/out-law/news/uk-patent-applications-ai-inventors

[Mel81x+]

For @Eco and @nudge

https://arstechnica.com/science/2020/11/when-ai-sees-a-man-it-thinks-official-a-woman-smile/

Fascinating read on how the AI used biased-data sets to build model cards for pictures it was presented.