Donate to Remove ads

Got a credit card? use our Credit Card & Finance Calculators

Thanks to johnstevens77,Bhoddhisatva,scotia,Anonymous,Cornytiv34, for Donating to support the site

AI endeavours

The Big Picture Place
odysseus2000
Lemon Half
Posts: 6358
Joined: November 8th, 2016, 11:33 pm
Has thanked: 1527 times
Been thanked: 958 times

Re: AI endeavours

#138440

Postby odysseus2000 » May 11th, 2018, 8:40 pm

Atlas looks very human in the countryside:

https://www.youtube.com/watch?v=vjSohj-Iclc

Regards,

odysseus2000
Lemon Half
Posts: 6358
Joined: November 8th, 2016, 11:33 pm
Has thanked: 1527 times
Been thanked: 958 times

Re: AI endeavours

#146574

Postby odysseus2000 » June 19th, 2018, 6:28 am

AI debates with humans & kind of interesting how the AI seems to need a human synthetic face & body to be more effective, but even so it drew with its human opponents:

https://www.theguardian.com/technology/ ... ct-debater

Regards,

odysseus2000
Lemon Half
Posts: 6358
Joined: November 8th, 2016, 11:33 pm
Has thanked: 1527 times
Been thanked: 958 times

Re: AI endeavours

#164017

Postby odysseus2000 » September 3rd, 2018, 11:25 pm

Short video on how Chinese government is using AI against people who do anti-social things like jay walk etc. The technology can identify folk and the police have glasses that link into the government data base of individuals. No idea how well all of this tech works, but get caught too many times and you get sent to a correction centre, go there too many times and it becomes your permanent address:

https://twitter.com/CNET/status/1035970546251362305

Big Brother 2018!

Regards,

odysseus2000
Lemon Half
Posts: 6358
Joined: November 8th, 2016, 11:33 pm
Has thanked: 1527 times
Been thanked: 958 times

Re: AI endeavours

#194406

Postby odysseus2000 » January 18th, 2019, 12:09 am

AI is better at gambling than professional gamblers:

https://www.engadget.com/2019/01/17/ai- ... -gamblers/

hat tip to Elon Musk (twitter) for link

Regards,

odysseus2000
Lemon Half
Posts: 6358
Joined: November 8th, 2016, 11:33 pm
Has thanked: 1527 times
Been thanked: 958 times

Re: AI endeavours

#197195

Postby odysseus2000 » January 29th, 2019, 12:22 am

This is an interesting article on AI and there processors needed:

https://arstechnica.com/gadgets/2018/07 ... arms-race/

if this is right then NVDIA who have lead the race are now producing chips that are too general whereas lots of competitors are working on chips that are tailored to the needs of AI.

If any of this is true, I am minded to believe it, then we can expect significant gains in AI hardware as folk switch to chips specifically optimised for AI rather than just general function.

Regards,

odysseus2000
Lemon Half
Posts: 6358
Joined: November 8th, 2016, 11:33 pm
Has thanked: 1527 times
Been thanked: 958 times

Re: AI endeavours

#218313

Postby odysseus2000 » April 30th, 2019, 5:23 am

This is the most amazing technology & imho very dangerous even now & getting better:

https://youtu.be/T76bK2t2r8g

How do you defend against weaponisation of this technology?

Regards,

odysseus2000
Lemon Half
Posts: 6358
Joined: November 8th, 2016, 11:33 pm
Has thanked: 1527 times
Been thanked: 958 times

Re: AI endeavours

#371021

Postby odysseus2000 » December 29th, 2020, 10:50 pm

Interesting to see Boston Dynamics have developed their AI robots. The rate of improvement from machines that needed to be quadrupeds to stand to the first stumbling bipeds has now reached the ability for bipeds to dance:

https://twitter.com/elonmusk/status/134 ... 80930?s=20

Given this rate of improvement it seems likely that robots will soon be able to outperform all Olympic athletes and all soldiers.

Hyundai took a big stake in Boston dynamics, so perhaps they will rival Tesla for robotic driving.

Regards,

onthemove
Lemon Slice
Posts: 540
Joined: June 24th, 2017, 4:03 pm
Has thanked: 722 times
Been thanked: 471 times

Re: AI endeavours

#371051

Postby onthemove » December 30th, 2020, 12:59 am

odysseus2000 wrote:Interesting to see Boston Dynamics have developed their AI robots. The rate of improvement from machines that needed to be quadrupeds to stand to the first stumbling bipeds has now reached the ability for bipeds to dance:

https://twitter.com/elonmusk/status/134 ... 80930?s=20

Given this rate of improvement it seems likely that robots will soon be able to outperform all Olympic athletes and all soldiers.

Hyundai took a big stake in Boston dynamics, so perhaps they will rival Tesla for robotic driving.

Regards,


That's really impressive kinematics, etc.

But, imv, not as impressive on the 'AI' side as some of their earlier demonstrations where they had the robots running around outside in very challenging conditions and terrain.

In terms of their routine, this is clearly pre-programmed, and the robots are dancing on a perfectly flat floor with the only obstacles being the other robots, and the routine's choreography will already have ensured they don't collide with each other. You could see the reflection of the cameraman and his assistant making sure they kept out of the area where the knew the robots would be dancing. So the robots weren't having to avoid them.

So it isn't demonstrating any particularly clever 'AI' in terms of sensing the world and interacting with it.

Don't get me wrong, from a robotics point of view, that motion control - being able to maintain their balance, etc - is very, very impressive, and yes there probably is some AI involved in doing that. But AI for controlling dynamically unstable elements isn't that new... e.g. https://www.researchgate.net/publicatio ... _Real_Time

Admittedly, these robots are now doing it coordinating multiple axes with (I presume) all the processing self contained within the robots, so it is still very impressive advance in technology ... but rather than necessarily being a demonstration of any amazingly new technique or AI per se, I see it as more a very impressive _implementation_ of technology... the impressiveness coming from the care, attention and effort that has gone into the implementation - the refinement of the components, the efficiency of the components, etc, and the care that will have gone into the implementation of the software controlling it, to make that as fast and efficient as possible, etc.


"Given this rate of improvement it seems likely that robots will soon be able to outperform all Olympic athletes and all soldiers."


How about this :shock:
https://www.youtube.com/watch?v=mAjnNhT ... andalorian

More seriously though...

Cars and bikes and planes and trains, and missiles, etc, can already outperform athletes and soldiers in terms of physical power and speed, etc.

Tanks, guns, missiles, etc, can all outperform soldiers in physical terms in battle.

Human soldiers going into battle today, are already augmented by technology anyway, so is there any really great concern if the Boston Dynamics robots are getting close to physically outperforming athletes and soldiers?

In most cases of battle these days, the presence of a human 'soldier' is only really to act as a 'driver' - be that a tank driver, an aircraft pilot, etc. There only to decide what to fire at and when. In physical terms, the presence of the human solider usually hinders the primary aim. A pilot in an aircraft can't stand anywhere near the forces the aircraft itself can withstand. Putting people in tanks requires all sorts of air conditioning, space, etc. If you could remotely operate the tank, it would be smaller, lighter and more effective. And so on.

But the (humanoid) technology that looks so impressive in that video, isn't the kind of technology that you would use if you were to automate the 'driver' in military vehicles / aircraft, etc.

You wouldn't go to the trouble of creating a humanoid robot to sit in the cockpit of an aircraft, any more than Waymo would create a humanoid robot to sit in the driving seat of their autonomous car to be the driver.

So I'm not sure that outperforming athletes or soldiers in terms of physical capability is all that much of a milestone when (other) technology can already do that by a huge margin.

odysseus2000
Lemon Half
Posts: 6358
Joined: November 8th, 2016, 11:33 pm
Has thanked: 1527 times
Been thanked: 958 times

Re: AI endeavours

#371331

Postby odysseus2000 » December 30th, 2020, 4:44 pm

So I'm not sure that outperforming athletes or soldiers in terms of physical capability is all that much of a milestone when (other) technology can already do that by a huge margin.


Yes, but after every battle one needs infantry to go in and 'mop-up' small pockets of resistance and then one needs infantry to enforce and maintain your domination.

Having a biped soldier, able to go anywhere that a human can go, but being more resistant to small arms, able to enter standby mode to dominate an area 24-7, able to see in the dark, have images from all surrounding areas via drones and/or satellites, able to self destruct if this gives increased probability to kill enemy and before long dressed as humans is imho a step change in military capability.

Imagine if you fighting against such humanoid robots, how would you do it? They can go everywhere you can go, they speak your language, they are many times stronger, they can 'see' all around, can call in artillery or drone strikes as needed, etc and are neither troubled by emotions, needing the bathroom, food or sleep.

Regards,

onthemove
Lemon Slice
Posts: 540
Joined: June 24th, 2017, 4:03 pm
Has thanked: 722 times
Been thanked: 471 times

Re: AI endeavours

#371481

Postby onthemove » December 30th, 2020, 11:00 pm

odysseus2000 wrote:
So I'm not sure that outperforming athletes or soldiers in terms of physical capability is all that much of a milestone when (other) technology can already do that by a huge margin.


Yes, but after every battle one needs infantry to go in and 'mop-up' small pockets of resistance and then one needs infantry to enforce and maintain your domination.

Having a biped soldier, able to go anywhere that a human can go, but being more resistant to small arms, able to enter standby mode to dominate an area 24-7, able to see in the dark, have images from all surrounding areas via drones and/or satellites, able to self destruct if this gives increased probability to kill enemy and before long dressed as humans is imho a step change in military capability.

Imagine if you fighting against such humanoid robots, how would you do it? They can go everywhere you can go, they speak your language, they are many times stronger, they can 'see' all around, can call in artillery or drone strikes as needed, etc and are neither troubled by emotions, needing the bathroom, food or sleep.

Regards,


An alternative to food will certainly be required.

But more generally, what is there militarily that one of those robots could do that couldn't be achieve simpler, quicker, cheaper, easier with a drone or swarm of drones.

Swarms of drones could even sacrifice one or two of themselves to blow a path for the following drones to enter into buildings.

I'm not sure that the ISIS militants in Iraq would be any more worried about humanoid bipeds than the drones that already are taking them out...

https://www.ibtimes.sg/what-ninja-bomb- ... yria-47261

"Imagine if you fighting against such humanoid robots, how would you do it? "


If you're armed, quite easily. The cleverness of those dancing robots is their efficiency. Making components light and (relatively) low power. It could easily be destroyed by a single bullet. Bulk them up to be bullet proof and you'll lose a substantial part of what makes them so versatile and impressive.

Not to mention the amount of metal in them would make magnetic triggered mines fairly selective at destroying the robots without destroying people walking over them.

Electromagnetic pulse would likely be effective, though not easy to create.

Easier to create, you could likely use lasers (just simple laser pointers) to interfere with their vision. Enough people working together could easily disable such a robot.

Trip wires would probably be quite effective. Robots using vision are going to be subject to many of the limitations that people have with vision as well. Trip them up then ambush with well targeted projectiles thrown into the joints.

Water jets would also be a good way of defeating them. A simple garden hose would probably do the trick. Something that complex and nimble is unlikely to be fully waterproof.

Traps... the covered hole kind would probably also be fairly effective.

They'd also be susceptible to misdirection, etc, ... use someone to act as a lure to lead them into any of the other traps as per above.

Simple fire would probably also be fairly effective. Petrol and matches.

Cutting off power supplies to a town for a few days would likely render the majority of any such robots unusable when their batteries run out.

Wearing a disguise so that you appear like one of 'them' may also help to deter them from attacking you and fool them.

Wearing a disguise to appear like a different person simply to not be targeted if they're looking for a specific target would also likely be effective.

I mean, presumably if these robots are going in, the aggressor isn't looking for a scorched earth policy, otherwise they'd just drop a big bomb on the place. No need to send in these robots.

Also large coils embedded in the road, etc, might also be effective. If turned on at the appropriate time the electric field could potentially induce current in any coils in the robot's motors and damage them. Or at least disable them from functioning correctly while the mob moves in for the kill.

Driving at them with a regular car would probably put them out of action. Swerve into it at the last minute to catch it off guard.

Attaching a small amount of explosive to a consumer drone with a remote detonator and flying it at such a robot would probably be pretty effective. Particularly if it's attention / use of limbs, etc, is distracted by a deliberate distraction, etc.

Or dressing a remote control car (carrying explosives) as a mouse or dog would probably also enable you to destroy such robots.

I wouldn't be surprised if a first generation of these sorts of robots would be easily confused by their own reflection in a well placed mirror.

Hidden rooms / hidden passages would also easily fool such robots.

If they're many times stronger than people, it's not as though they're going to be able to hide themselves amongst the population the way special forces often do. So stealth isn't really going to be one of their features. Which would make it easy for a disgruntled population to then implement any / all the above suggestions.

All in all, I don't really see any situation where such robots could be used without the consent of the population in which they are placed.

If you tried to swarm the place with a large number of them, can you imagine what the charging points are going to be like! It'd be worse than queuing up for a Tesla car charger!

odysseus2000
Lemon Half
Posts: 6358
Joined: November 8th, 2016, 11:33 pm
Has thanked: 1527 times
Been thanked: 958 times

Re: AI endeavours

#371494

Postby odysseus2000 » December 30th, 2020, 11:17 pm

I believe you are missing a vital point.

It takes about 12 years to turn a human into a soldier, more like 16 to be effective.

Robots can be cranked out in very large numbers very quickly.

Destroy a robot and a new one is sent in, destroy a resistance fighter & it takes a long time to replace him/her.

Other points are that robots would not fight alone, sure you could take them out by some of the means you discuss, but in doing so you reveal yourself to other robots or drones etc & we return to replacement times.

Also in standby mode one of these machines could endure for weeks using small amounts of power & then reactivate in milli-seconds.

Each robot would be continually updated with images of known enemy & able to see through many disguises to execute these hostile.

I think the whole capability of resistance/terrorists fighter will soon be under mined by this kind of technology.

Regards,

odysseus2000
Lemon Half
Posts: 6358
Joined: November 8th, 2016, 11:33 pm
Has thanked: 1527 times
Been thanked: 958 times

Re: AI endeavours

#396149

Postby odysseus2000 » March 16th, 2021, 1:23 pm


odysseus2000
Lemon Half
Posts: 6358
Joined: November 8th, 2016, 11:33 pm
Has thanked: 1527 times
Been thanked: 958 times

Re: AI endeavours

#442077

Postby odysseus2000 » September 14th, 2021, 10:38 am

Interesting items from a Chinese fair on AI:

https://youtu.be/ez1KnhUx9m4

I was discussing with an acquaintance the shortage of GP's. He was of the opinion that we needed to train a lot more, but I thought that most of what a GP does could be replaced with an AI system. There are several business working on this and in the video we see a Chinese version along with a mobile CT unit that looks to me to be how health care will go. It takes about 26 years from birth to a junior Dr, another 10 before they are good. AI systems can be created in days. Results I have seen from various comparisons of AI to humans show similar & often better results from the AI.

Regards,

murraypaul
Lemon Slice
Posts: 785
Joined: April 9th, 2021, 5:54 pm
Has thanked: 225 times
Been thanked: 265 times

Re: AI endeavours

#442094

Postby murraypaul » September 14th, 2021, 11:23 am

odysseus2000 wrote:AI systems can be created in days.


Standing up another instance of an existing AI system can be done quickly.
Creating AI systems takes years.
Otherwise, why don't we have our level 5 autonomous driving cars yet?

odysseus2000
Lemon Half
Posts: 6358
Joined: November 8th, 2016, 11:33 pm
Has thanked: 1527 times
Been thanked: 958 times

Re: AI endeavours

#442104

Postby odysseus2000 » September 14th, 2021, 12:01 pm

murraypaul wrote:
odysseus2000 wrote:AI systems can be created in days.


Standing up another instance of an existing AI system can be done quickly.
Creating AI systems takes years.
Otherwise, why don't we have our level 5 autonomous driving cars yet?


Yes, creating the first narrow AI for some job takes time, but having got it working one can duplicate very quickly. Driving is one of the most difficult challenges as the AI has to be fast and the consequence of a mistake is very bad.

Most other AI does not have the very fast speed constraint and can naturally explore possibilities and then come to a conclusion. E.g. I can currently see no technical reason why AI systems can not replace GP's. There will be a number of union objections but as far as I can tell it is only a matter of time (a few years) before the GP as we know it is gone and is replaced by something far cheaper and better. This applies to very many jobs.

Regards,

murraypaul
Lemon Slice
Posts: 785
Joined: April 9th, 2021, 5:54 pm
Has thanked: 225 times
Been thanked: 265 times

Re: AI endeavours

#442109

Postby murraypaul » September 14th, 2021, 12:06 pm

odysseus2000 wrote:Yes, creating the first narrow AI for some job takes time, but having got it working one can duplicate very quickly. Driving is one of the most difficult challenges as the AI has to be fast and the consequence of a mistake is very bad.


I would suggest to you that the consequences of a mistake for a medical AI could also be very bad?

E.g. I can currently see no technical reason why AI systems can not replace GP's. There will be a number of union objections but as far as I can tell it is only a matter of time (a few years) before the GP as we know it is gone and is replaced by something far cheaper and better.


And what technical qualifications do you have to make that prediction?

Automated assistance for simple problems, yes. Replacing all human intervention for all areas covered by GPs? In the next few years? No.

odysseus2000
Lemon Half
Posts: 6358
Joined: November 8th, 2016, 11:33 pm
Has thanked: 1527 times
Been thanked: 958 times

Re: AI endeavours

#442119

Postby odysseus2000 » September 14th, 2021, 12:26 pm

murraypaul wrote:
odysseus2000 wrote:Yes, creating the first narrow AI for some job takes time, but having got it working one can duplicate very quickly. Driving is one of the most difficult challenges as the AI has to be fast and the consequence of a mistake is very bad.


I would suggest to you that the consequences of a mistake for a medical AI could also be very bad?

E.g. I can currently see no technical reason why AI systems can not replace GP's. There will be a number of union objections but as far as I can tell it is only a matter of time (a few years) before the GP as we know it is gone and is replaced by something far cheaper and better.


And what technical qualifications do you have to make that prediction?

Automated assistance for simple problems, yes. Replacing all human intervention for all areas covered by GPs? In the next few years? No.


GP's make mistakes all the time, I have known many people incorrectly diagnosed by a GP leading to some very serious outcomes. AI will be better.

I believe you have not interacted with AI neural net systems and consequently have no understanding of what they can do. These are nothing like the old system where one was asked a series of questions and the machine then tried to make a diagnosis based on your answers. Modern AI interacts with you, listens to what you say and builds a model of you and the area it is trained to work upon. In essence you are dealing with an intelligence that has an intense memory of all human ailments and human psychology, something impossible for a GP but approximated to after 10+ years of experience.

The last time I spoke to a GP she did not listen to much of what I said, picking up on something without the context of my previous statements and going off on a tangent that I had to correct.

Most people I deal with have similar prejudice and don't have any experience or understanding of what AI can do or of the limits of human systems in comparison, nor that AI systems are cheap, operate 24-7.

Everything that we know is changing and there are huge investment opportunities here and huge changes in what humans will do going forwards.

Regards,

doolally
Lemon Slice
Posts: 631
Joined: February 8th, 2021, 10:55 am
Has thanked: 107 times
Been thanked: 509 times

Re: AI endeavours

#442131

Postby doolally » September 14th, 2021, 12:37 pm

odysseus2000 wrote:
murraypaul wrote:
odysseus2000 wrote:AI systems can be created in days.


Standing up another instance of an existing AI system can be done quickly.
Creating AI systems takes years.
Otherwise, why don't we have our level 5 autonomous driving cars yet?


Yes, creating the first narrow AI for some job takes time, but having got it working one can duplicate very quickly. Driving is one of the most difficult challenges as the AI has to be fast and the consequence of a mistake is very bad.

Most other AI does not have the very fast speed constraint and can naturally explore possibilities and then come to a conclusion. E.g. I can currently see no technical reason why AI systems can not replace GP's. There will be a number of union objections but as far as I can tell it is only a matter of time (a few years) before the GP as we know it is gone and is replaced by something far cheaper and better. This applies to very many jobs.

Regards,

AI does have its place for some ailments, but some issues, especially mental health issues where logic flies out the window, absolutely need human interaction.
And mental health issues represent an ever-growing proportion of GP's time

doolally

odysseus2000
Lemon Half
Posts: 6358
Joined: November 8th, 2016, 11:33 pm
Has thanked: 1527 times
Been thanked: 958 times

Re: AI endeavours

#442159

Postby odysseus2000 » September 14th, 2021, 1:47 pm

AI does have its place for some ailments, but some issues, especially mental health issues where logic flies out the window, absolutely need human interaction.
And mental health issues represent an ever-growing proportion of GP's time

doolally


Actually I think AI will be far better for mental health issues.

As we stand now a patient has to wait a long time for a telephone appointment, often sees or audios with a different person every time and then gets prescribed drugs with addictive properties that treat the symptoms not the underlying problems.

With an AI the patient can get through in seconds, speak to the same AI which has all the patients details and before long will know more about them than their mother. The AI can do both the job of the GP and the psychologist which a human GP may refer the patient to.

In essence AI creates a knowledgeable Samartian, programmed with all known human psychology and able to interact with the patient for as long as the patient wants, simultaneously dealing with 1000's of other patients at the same time and while maintaining a memory of all the patient has told the AI in previous sessions and a knowledge of the patients physiological state as well.

This kind of service is now available to very wealthy and powerful folk who can afford private physicians etc, but most have to make to do with terrible service.

Regards,

murraypaul
Lemon Slice
Posts: 785
Joined: April 9th, 2021, 5:54 pm
Has thanked: 225 times
Been thanked: 265 times

Re: AI endeavours

#442167

Postby murraypaul » September 14th, 2021, 2:02 pm

odysseus2000 wrote:With an AI the patient can get through in seconds, speak to the same AI which has all the patients details and before long will know more about them than their mother. The AI can do both the job of the GP and the psychologist which a human GP may refer the patient to.

In essence AI creates a knowledgeable Samartian, programmed with all known human psychology and able to interact with the patient for as long as the patient wants, simultaneously dealing with 1000's of other patients at the same time and while maintaining a memory of all the patient has told the AI in previous sessions and a knowledge of the patients physiological state as well.


That is wonderful, but it doesn't exist yet, and nothing we have is even close.
You are describing science fiction, not science fact.


Return to “Macro and Global Topics”

Who is online

Users browsing this forum: No registered users and 28 guests