Suggestions for using AI

[mistermack]

I have one or two ideas for what AI can be used for. 

If you have any, and would like it to be chewed over by others, post it here. Might be interesting. 

 

To start it off, I would like to see it used to "re-create" old images, making a modern very hi-def photo, from an old blurry bit of rubbish. I'm not talking about playing with contrast, brightness and colour and focus in the usual sense. 

What I would do is amass a database of modern, very high definition pictures, of say people, animals, cars, landscapes. Whatever subject is commonly captured. At every angle that you could. You would need a huge database, the bigger it got, the better it would work. Then divide each picture into many millions of individual squares, and digitise it's properties. 

Then you take your old fuzzy photo, and divide that into millions of squares, and digitise it's properties. Now use AI to find a match for each fuzzy square, in the database of hi-def squares, and do that for the whole image. What you would end up with ideally would be a very high definition image, that had nothing of the original in it, but was to the human eye a vastly improved high-definition version of it. Use the learning capability to compare the result with the original, note the bad bits, and run it again and again, until the best possible result was gained. 

[npts2020]

It could be used to automate the entire transit system to cut the ~40,000 annual fatalities to nearly zero and allow everyone to use the system.

[TheVat]

Though it might not be real fun to be in a fast-moving vehicle when a strong solar flare causes a massive EMP on Earth.  Maybe it could work okay with robust Faraday caging and a good onboard mechanical backup that reliably stops the vehicle if there is electronic failure.

3 hours ago, mistermack said:

Then you take your old fuzzy photo, and divide that into millions of squares, and digitise it's properties. Now use AI to find a match for each fuzzy square, in the database of hi-def squares, and do that for the whole image. What you would end up with ideally would be a very high definition image, that had nothing of the original in it, but was to the human eye a vastly improved high-definition version

They seem to be working that direction.

https://www.engadget.com/free-ai-old-photo-restoration-tool-190502361.html

While I enjoy the fresh air and fitness derived from physical tasks around my property, I would welcome a safe AI lawnmower.  One smart enough to recognize patches we leave in a natural state (blossoming clover and alfalfa, sage, Virginia Creeper, wild grape, burdock, pineapple weed, spear thistle, echinacea, et al) and bend around them.  

[mistermack]

38 minutes ago, TheVat said:

They seem to be working that direction.

https://www.engadget.com/free-ai-old-photo-restoration-tool-190502361.html

That's a good find. I had a feeling someone would be trying it already, and it looks like they're doing well with it. 

In a short while, you should be able to construct totally new scenes, by splicing a few images together, and then letting this kind of app loose on it. So you could create a hi-def image of the Pope getting jiggy with Queen Elizabeth II . So images will no longer have any value as evidence, unless the evidence trail is unimpeachably recorded. 

1 hour ago, npts2020 said:

It could be used to automate the entire transit system to cut the ~40,000 annual fatalities to nearly zero and allow everyone to use the system.

I reckon the way to do that, would be to put everything on rails, which you could drop down off for the last few yards of your trip. So have steel wheel/tyre hybrids. On rails, you can computer-control position very precisely, and pick up energy along the way. Running on rails would also use less energy, and less resources for wear and tear of tyres. And less road maintenance and maybe less noise. 

[wtf]

7 hours ago, npts2020 said:

It could be used to automate the entire transit system to cut the ~40,000 annual fatalities to nearly zero

18 confirmed autonomous vehicle deaths so far and counting. The evidence does not support your hope.

https://www.slashgear.com/1202594/how-many-people-have-actually-been-killed-by-self-driving-cars/

[Endy0816]

I've been using them for original recipe ideas. Would be cool if could be done as part of an experimental restraunt menu.

Likewise something similar for products. They come up with interesting and unconventional design concepts. A simple system to bring them the rest of the way into reality would be incredible.

Edited August 6, 2023 by Endy0816

[geordief]

Could AI maybe just live my life for me?

 I dunno ,just divide my present circumstance into a million pixels and forecast a subsequent array of pixels  that were run against a database of  almost identical arrays  and stitched together  to form a new ensemble.

This would be presented to me as a possible reaction to my present circumstance  which I could accept (a serotonin hit)  or reject (an electric impulse in the anal area)

 

Or maybe even I might just get a life.(I hear living is easy and the cotton is high)

[Peterkin]

2 hours ago, wtf said:

18 confirmed autonomous vehicle deaths so far and counting. The evidence does not support your hope.

https://www.slashgear.com/1202594/how-many-people-have-actually-been-killed-by-self-driving-cars/

That just means the autonomous cars and the existing traffic grid are incompatible. The properly co-ordinated mass transit system of cities could very well be run by a central computer. The vehicles would be on the central maglev lane and the rest of the road would be bicycle paths and pedestrian walkway, which would liberate the sidewalk for shop displays, outdoor seating, public art and potted plants. Private vehicles would be allowed only outside city limits.

Emergency services, too. It could answer and track calls at the same time and dispatch the appropriate response unit faster than any human operator can. I understand many urban systems are already more or less computerized, but I don't know how well they're integrated. An AI central controller would know exactly which and how many units were available at any given time and their exact locations, and could both direct them to to the right facility (vacant beds, staff on duty, equipment, preparedness) by the most efficient route but also clear their way by adjusting traffic lights.

Most of all, I'd like to see one in charge of tracing, collecting and recycling guns. All of them! I want great hulking cybermen to stomp up to gun-owners' houses, crash through the door, find and scoop up the wretched guns (unless, of course, the owners chose to bring them out and surrender them) and carry them away to oblivion - maybe have them come back eventually as baby buggies and gardening tools. It's a modest little dream....    

 

Edited August 7, 2023 by Peterkin

[wtf]

28 minutes ago, Peterkin said:

I want great hulking cybermen to stomp up to gun-owners' houses, crash through the door, find and scoop up the wretched guns (unless, of course, the owners chose to bring them out and surrender them) and carry them away to oblivion

Authoritarian much? I push back gently on your dream to have a computer running our entire transportation system and you reply by fantasizing about breaking down citizens' doors because they're exercising Constitutional rights you don't happen to agree with?

What else have you got planned for us?

Edited August 7, 2023 by wtf

[Peterkin]

58 minutes ago, wtf said:

What else have you got planned for us?

Peace, prosperity and tolerance too long a reach?

The original transportation system suggestion wasn't mine, btw; I had no personal brief with you. It's the killing machines I dislike.

Edited August 7, 2023 by Peterkin

[wtf]

9 minutes ago, Peterkin said:

It's the killing machines I dislike.

Now that's funny. But enough about the autonomous cars

[Peterkin]

5 hours ago, wtf said:

18 confirmed autonomous vehicle deaths so far and counting. The evidence does not support your hope.

That happened over 9 years?  In that period, at least 100,000 were killed by drunk drivers.

Evidence is as evidence does.

[swansont]

5 hours ago, Peterkin said:

That happened over 9 years?  In that period, at least 100,000 were killed by drunk drivers.

Evidence is as evidence does.

What’s the rate, though? You can’t just compare the raw numbers.

[Sensei]

14 hours ago, wtf said:

18 confirmed autonomous vehicle deaths so far and counting. The evidence does not support your hope.

How many of these were caused by autonomous vehicles? How many of them were deaths of the owner of the autonomous vehicle?

If a motorcycle is traveling at 120 km/h on the road while an autonomous car is moving at 120 km/h in the opposite direction, and the motorcycle is heading straight for your (i.e., wrong) path (because it is overtaking another vehicle), a collision is unavoidable..

Radars, lidars, cameras, have limited (as everything) capabilities (resolution, refresh frequency, range)..

 

To avoid collisions with other vehicles, ALL vehicles must be equipped with automatic assistance. They will exchange data with sufficient frequency and delay to tell everyone else where they are, so they will e.g. negotiate whether it is safe to overtake at a given place at a given time.

 

You should not compare absolute numbers, but the percentage of deaths per number of autonomous cars (and it was turned on during the crash) with the percentage of deaths per number of non-autonomous cars.

Without a investigation, it is not possible to distinguish between autonomous cars with autonomous driving systems enabled and autonomous cars with autonomous driving systems disabled. They could have simply collected them for statistics by looking at the car model.

 

 

Edited August 7, 2023 by Sensei

[TheVat]

13 hours ago, geordief said:

Could AI maybe just live my life for me?

 I dunno ,just divide my present circumstance into a million pixels and forecast a subsequent array of pixels  that were run against a database of  almost identical arrays  and stitched together  to form a new ensemble.

This would be presented to me as a possible reaction to my present circumstance  which I could accept (a serotonin hit)  or reject (an electric impulse in the anal area)

 

Or maybe even I might just get a life.(I hear living is easy and the cotton is high)

Only if yo daddy's rich and yo mama's good-lookin.  

[Peterkin]

5 hours ago, swansont said:

What’s the rate, though? You can’t just compare the raw numbers.

OK. I was too lazy to do look up all the stats and do the arithmetic, so I estimated low.

Quote

In 2021, 13,384 people died in alcohol-impaired driving traffic deaths https://www.nhtsa.gov/risky-driving/drunk-driving

Assuming the average is lower than that, a more accurate estimate would be 118,000 for the period I mentioned. However, I was wrong to count the nine years since the first road-test; self-driving cars have only been legal on public roads since 2017, which means those 18 death must have occurred in only 6 years and the corresponding drunk driver deaths would be closer to 82,000. I didn't factor in speeding, distracted and tired driver driver error, road rage or vehicular homicide. My uninformed guess is that the autonomous car deaths would be most nearly comparable to the last named category. I do have reason to believe if just the drunks were decanted into ai driven vehicles, the roads would already be safer.  

A better way to compare would perhaps be to find out the number of fatal accidents caused by mechanical malfunction and include autonomous control malfunction among those, then compare vehicle error with human error.  But I don't have those numbers.

Edited August 7, 2023 by Peterkin

[swansont]

Haven’t found fatalities (yet), but according to these sources driverless vehicles have ~9 crashes per million miles driven, while the overall US rate is 0.26 per million miles

https://www.knrlegal.com/car-accident-lawyer/self-driving-car-accident-statistics/#:~:text=Self-Driving Car Quick Facts,autonomous car to drive them.

https://safetydawg.com/measure-collision-rate-safety-consultant/#:~:text=Recently the “American Transportation Research,average or better than average%3F

Fatalities depend on several factors, including how close you are to a hospital, so driverless vehicles not being legal everywhere might not give a true safety comparison, but having a ~36x higher collision rate surely suggests the technology isn’t there yet, and we’d not be safer 

edit: 

2nd source has a somewhat restrictive definition of collision; it doesn’t include “fender benders”

https://jalopnik.com/self-driving-car-vs-human-99-percent-safe-crash-data-1850170268

Cites a higher rate - 1 crash per .55 million miles. Still lower than driverless

[Sensei]

19 minutes ago, swansont said:

https://www.knrlegal.com/car-accident-lawyer/self-driving-car-accident-statistics/#:~:text=Self-Driving Car Quick Facts,autonomous car to drive them.

"For instance, there is no way to definitively confirm if the self-driving feature was in use at the time of a collision, nor does the data consider if the other vehicle was the cause of the crash."

19 minutes ago, swansont said:

Haven’t found fatalities (yet), but according to these sources driverless vehicles have ~9 crashes per million miles driven, while the overall US rate is 0.26 per million miles

People with electric cars, such as Tesla, drive mostly in cities. Meanwhile, statistics on miles traveled by non-electric cars include highways and interstate travels..

Edited August 7, 2023 by Sensei

[Peterkin]

A whole lot of data is as yet unavailable. Comparisons are made even more difficult by the fact that it cannot be determined with any certainty how great a part each factor - road conditions, vehicle capability, other vehicles, pedestrians, alcohol, speed, available options and unavoidable obstacles - played in the outcome.

I agree that the technology is not quite there yet, and part of what needs to change from now to when the world is ready for a fully automated transportation system is the configuration of roads and traffic markers, which were originally made for human drivers. Changes had to me made quite rapidly when the automobile took over more and more of the road from horse-powered vehicles (That, too, was a dangerous time on busy thoroughfares.) but the adaptations were made as the gas-powered car took over. So, too, will this transformation take place, growing pains and all.

[mistermack]

18 hours ago, wtf said:

18 confirmed autonomous vehicle deaths so far and counting. The evidence does not support your hope.

https://www.slashgear.com/1202594/how-many-people-have-actually-been-killed-by-self-driving-cars/   

 

Well, I clicked on that link, and it says that all of the fatal incidents occurred in cars that had a driver present, and there were no fatalities notified in a car that was fully driverless. I don't really know what to make of that, except that supervision by a human appears to be detrimental in some way. It's very hard to picture why that should be so, but maybe the fully driverless cars go slower, or are being used in safer environments. 

With a totally new technology, you would expect dramatic improvements in the early days, and given that governments are not pulling the plug on driverless vehicles, it looks like it's already made it as an acceptable risk on the roads. The true judgement will come from the insurance companies.

[Peterkin]

8 minutes ago, mistermack said:

With a totally new technology, you would expect dramatic improvements in the early days

When tested in a purposed-made environment, that is the case. When incorporating new technology into an existing (and frequently dysfunctional) system, problems arise. (Oh, like you never had Microsoft force some new upgrade on your computer that crashed three of your programs!)

[swansont]

2 hours ago, Sensei said:

"For instance, there is no way to definitively confirm if the self-driving feature was in use at the time of a collision, nor does the data consider if the other vehicle was the cause of the crash."

We’d expect the crash rate to be the baseline if the self-driving feature was not employed.

 

[mistermack]

32 minutes ago, Peterkin said:

When incorporating new technology into an existing (and frequently dysfunctional) system, problems arise.

Yes but problems are usually temporary and are only relevant long-term if they are totally insurmountable, or turn out to make the whole operation financially non-viable. I haven't heard anything regarding driverless vehicles that is threatening either of those things. It looks right now like driverless is the inevitable future.

[Genady]

To convert a YouTube video to a well-organized written article.

[mistermack]

I think AI is ideal for crime detection, and offender profiling.

I believe that there is already some sort of software to identify patterns in offending, but running AI over the details of offences will be much more efficient, I would have thought. 

There are loads of examples where police had details of serious offenders, interviews, police notes and suggestions etc, that got filed, ignored or dumped by some cop who had his own theory and ignored anything else. The Yorkshire Ripper case is a notorious example. The top cop got fooled by a few hoax messages, and the edict went out that only suspects with a north-eastern accent should be considered. The real ripper was interviewed and the reports buried, even though the interviewing cops were very suspicious of him.

If the AI was pointing away from the hoax at every turn, they could have solved it much quicker. In fact, they never solved it, he fell into their hands in the end, caught in the act.