I can see your point. To my mind, there should be consequences to any company getting a subsidy if they don't deliver. Maybe like "if you don't deliver, we expect repayment of the subsidy."
And/or idiotically incompetent. I can see a scenario where someone said "hey, we have code that does what we want it to do already, let's just reuse that code", without realizing that code did other things too. As a professional programmer, it's incompetent bordering on negligent to reuse code without serious analysis, and that's an entirely believable explanation for what may have happened here.
Having said that, they're still liable for whatever the consequences of messing up that badly are, even if it wasn't intended.
1. According to the original article, "AI has more cases of mistaken identity with women and minorities, multiple research projects have found." So yes, they do have some evidence for it.
2. The Officers in the field would be likely to arrest and/or detain and/or question anyone that the system flagged, yes.
3. It is highly probable that the users (read: Police Officers) would be incapable of evaluating the effectiveness/correctness of the system. The agency involved was asking for installation in the real world, not in evaluation considerations. They'd already made up their mind that they wanted it and wanted to use it.
4. Fixing the system after abuses caused by blindly taking actions based on the flawed output of the system is too late. Those abuses would have already happened.
MS is, for a rare time in their history, taking the appropriate course here. They're saying, "we choose not to take this contract because we are aware the system, in its current form, has flaws, and flaws that can lead to abuse. We don't want a part of that."
At the moment, all of this is projections of performance and cost. Until we see it working, it's all just projections. If they can pull it off, my hat's off to them! But I don't think we should say things like "about to come on-line" - that makes it sound like it's gonna happen next week.
Nobody's been lied to - yet. To quote you, "Starlink is projected to operate around 25 to 50ms". Projected. All I'm saying is let's not tout this as the be-all and end-all solution until we see it actually working at the projected performance and price points.
Proof-of-concept is one thing, but some of the posters here are talking about it as if it were commercially available now, or will be next month. That's what I'm taking exception to. And as far as quoting pricing loooong before commercial availability, that's nothing more than wishful thinking.
I'm saying an absolute theoretical minimum of 2ms added, and that's only if you're directly underneath one of the satellites, otherwise the travel path is longer, and therefore so is the added delay from the propagation of the electromagnetic waves.
That's not taking into account any delay introduced by the electronics in the satellite itself, or the ground stations.
I'm not saying that usable satellite internet is impossible, but I am saying that it's probably not as easy or as close to rollout as some might think.
I'd love a free phone! Unfortunately, I have to pay for my phone, my service, and my broadband, with the money I earn from my job, you know, like it's supposed to work.
What I'm saying is that by going to a satellite, you're adding those delays - an absolute minimum of adding 2ms, and that 2ms is only if you're directly underneath one of the satellites - onto whatever time it will take for the ping to travel through the terrestrial infrastructure (note where I said "Ping goes from ground station through terrestrial infrastructure (no faster than with wired broadband)" and "Response comes through terrestrial infrastructure to ground station (no faster than with wired broadband)"). Unless you've found a way for those electromagnetic waves to travel faster than our current understanding of physics allows them to.
According to Wikipedia, "Low Earth orbit (LEO): geocentric orbits with altitudes below 2,000 km (100–1,240 miles)"
Let's look at the minimum of that range: 100 miles.
Electromagnetic waves travel at approximately 1 foot / nanosecond.
Meaning 5.28 microseconds per mile.
So, 528 microseconds for 100 miles.
1056 microseconds for the round-trip to the satellite.
That's 1 ms right there, just for the electromagnetic waves to travel up to the satellite, and the response to travel back down.
Double that, because here's what has to happen:
Ping goes from user to satellite (~0.5ms).
Ping goes from satellite to ground station (~0.5ms).
Ping goes from ground station through terrestrial infrastructure (no faster than with wired broadband).
Response comes through terrestrial infrastructure to ground station (no faster than with wired broadband).
Response goes from ground station to satellite (~0.5ms)
Response goes from satellite to original ping initiator (~0.5ms).
As an electrical engineer with an extensive physics background, I'd be pretty impressed if they could get that all done in 20ms or less.
I'm not against some sort of oversight or at least accountability here, but at least this is a step in the right direction - even if it's a small step.
The fact is, that the USF already exists, and unless the laws change, it's not going away. So why not use it for its intended purpose?
As far as market forces doing it better and cheaper, they haven't. The very fact that even Republicans at this point are advocating use of the fund for helping with some of the costs bears that out.
While I'm no supporter of a lot of Republican policies, I'm actually with them on this one. According to the original article, they're planning on reallocating those funds from the Universal Service Fund - and this sort of thing (getting telecom services to underserved / rural areas) is exactly what the USF was intended to help with.
As I said in a previous response, it's more about perception and acceptance than reality.
If an autonomous vehicle has a 1 incident in 10k miles driven rate, and the average human driver has that exact same rate, the autonomous system is going to get way more press, and knee-jerk reactions will occur. That's just the nature of the press and human nature.
The general public doesn't seem to care about comparing incident rates between autonomous vehicles and human-controlled vehicles.
The complexity of autonomous vehicles is immense, especially since the general public and regulators are expecting them to be better at making decisions and safer than human drivers.
While the complexity is immense, it wouldn't surprise me at all if autonomous vehicles are already safer than human drivers. There are two problems, though: (1) there are still many bad human drivers on the road, and (2) the general public and regulators aren't just expecting the autonomous vehicles to be safer than humans, they're unrealistically expecting them to be perfect drivers while being forced to share the road with human drivers.
That's kinda what I meant. It's more about perception than reality. It wouldn't surprise me either if they were already safer. But the perception is that unless they're perfect, they're not safer.
The complexity of autonomous vehicles is immense, especially since the general public and regulators are expecting them to be better at making decisions and safer than human drivers. I'd be willing to say that it's orders of magnitude bigger than the difference between reusable and non-reusable rockets.
Is anyone besides the CEOs themselves surprised that the corporate-types vastly underestimated the complexity of a problem they didn't truly understand in the first place?
Agreed 100%. It's important for people in general to understand that any statistical analysis shouldn't be used to decide how to treat/deal with a given individual.
There's a difference, though, between algorithms which use non-statistical mathematics, and those that use statistics.
Any time statistical analyses are involved, there are going to be times when it leads to a non-optimal answer. For instance, if an algorithm is based on data like "Steve likes classic rock songs 80% of the time", then that algorithm, when asked "will Steve like this particular classic rock song?" will get it right about 80% of the time. 20% of the time the algorithm will get the answer wrong.
That's not necessarily a biased algorithm, but it is a flawed one, in the respect that it isn't comprehensive enough, or doesn't have enough data to be better.
Slashdot Mirror
I can see your point. To my mind, there should be consequences to any company getting a subsidy if they don't deliver. Maybe like "if you don't deliver, we expect repayment of the subsidy."
... the more evil Facebook looks.
And/or idiotically incompetent. I can see a scenario where someone said "hey, we have code that does what we want it to do already, let's just reuse that code", without realizing that code did other things too. As a professional programmer, it's incompetent bordering on negligent to reuse code without serious analysis, and that's an entirely believable explanation for what may have happened here.
Having said that, they're still liable for whatever the consequences of messing up that badly are, even if it wasn't intended.
1. According to the original article, "AI has more cases of mistaken identity with women and minorities, multiple research projects have found." So yes, they do have some evidence for it.
2. The Officers in the field would be likely to arrest and/or detain and/or question anyone that the system flagged, yes.
3. It is highly probable that the users (read: Police Officers) would be incapable of evaluating the effectiveness/correctness of the system. The agency involved was asking for installation in the real world, not in evaluation considerations. They'd already made up their mind that they wanted it and wanted to use it.
4. Fixing the system after abuses caused by blindly taking actions based on the flawed output of the system is too late. Those abuses would have already happened.
MS is, for a rare time in their history, taking the appropriate course here. They're saying, "we choose not to take this contract because we are aware the system, in its current form, has flaws, and flaws that can lead to abuse. We don't want a part of that."
At the moment, all of this is projections of performance and cost. Until we see it working, it's all just projections. If they can pull it off, my hat's off to them! But I don't think we should say things like "about to come on-line" - that makes it sound like it's gonna happen next week.
Nobody's been lied to - yet. To quote you, "Starlink is projected to operate around 25 to 50ms". Projected. All I'm saying is let's not tout this as the be-all and end-all solution until we see it actually working at the projected performance and price points.
Proof-of-concept is one thing, but some of the posters here are talking about it as if it were commercially available now, or will be next month. That's what I'm taking exception to. And as far as quoting pricing loooong before commercial availability, that's nothing more than wishful thinking.
Well, if we're being fair here, we'd have to say that it's either OK in both cases, or in neither case.
And the outlawing of community-based providers? To me, outlawing them is an abomination. Let community-based players try, just like anyone else.
I'm saying an absolute theoretical minimum of 2ms added, and that's only if you're directly underneath one of the satellites, otherwise the travel path is longer, and therefore so is the added delay from the propagation of the electromagnetic waves.
That's not taking into account any delay introduced by the electronics in the satellite itself, or the ground stations.
I'm not saying that usable satellite internet is impossible, but I am saying that it's probably not as easy or as close to rollout as some might think.
I'd love a free phone! Unfortunately, I have to pay for my phone, my service, and my broadband, with the money I earn from my job, you know, like it's supposed to work.
What I'm saying is that by going to a satellite, you're adding those delays - an absolute minimum of adding 2ms, and that 2ms is only if you're directly underneath one of the satellites - onto whatever time it will take for the ping to travel through the terrestrial infrastructure (note where I said "Ping goes from ground station through terrestrial infrastructure (no faster than with wired broadband)" and "Response comes through terrestrial infrastructure to ground station (no faster than with wired broadband)"). Unless you've found a way for those electromagnetic waves to travel faster than our current understanding of physics allows them to.
I'd like to see that link, too.
Let's look at some basic electronics and physics:
According to Wikipedia, "Low Earth orbit (LEO): geocentric orbits with altitudes below 2,000 km (100–1,240 miles)"
Let's look at the minimum of that range: 100 miles.
Electromagnetic waves travel at approximately 1 foot / nanosecond.
Meaning 5.28 microseconds per mile.
So, 528 microseconds for 100 miles.
1056 microseconds for the round-trip to the satellite.
That's 1 ms right there, just for the electromagnetic waves to travel up to the satellite, and the response to travel back down.
Double that, because here's what has to happen:
Ping goes from user to satellite (~0.5ms).
Ping goes from satellite to ground station (~0.5ms).
Ping goes from ground station through terrestrial infrastructure (no faster than with wired broadband).
Response comes through terrestrial infrastructure to ground station (no faster than with wired broadband).
Response goes from ground station to satellite (~0.5ms)
Response goes from satellite to original ping initiator (~0.5ms).
As an electrical engineer with an extensive physics background, I'd be pretty impressed if they could get that all done in 20ms or less.
I'm not against some sort of oversight or at least accountability here, but at least this is a step in the right direction - even if it's a small step.
False all around.
I expected more from commenters on /.
You must be new here to /.
It's been some time since there wasn't a ton of crap comments on any given thread.
Downvoted? I don't think so. I doubt you had the mod points to do it.
The fact is, that the USF already exists, and unless the laws change, it's not going away. So why not use it for its intended purpose?
As far as market forces doing it better and cheaper, they haven't. The very fact that even Republicans at this point are advocating use of the fund for helping with some of the costs bears that out.
The cost doesn't change, but the USF was established so that it could help defray those costs.
While I'm no supporter of a lot of Republican policies, I'm actually with them on this one. According to the original article, they're planning on reallocating those funds from the Universal Service Fund - and this sort of thing (getting telecom services to underserved / rural areas) is exactly what the USF was intended to help with.
It's much more expensive to do that in underserved / rural areas, and that's exactly what the Universal Service Fund was designed to help with.
As I said in a previous response, it's more about perception and acceptance than reality.
If an autonomous vehicle has a 1 incident in 10k miles driven rate, and the average human driver has that exact same rate, the autonomous system is going to get way more press, and knee-jerk reactions will occur. That's just the nature of the press and human nature.
The general public doesn't seem to care about comparing incident rates between autonomous vehicles and human-controlled vehicles.
While the complexity is immense, it wouldn't surprise me at all if autonomous vehicles are already safer than human drivers. There are two problems, though: (1) there are still many bad human drivers on the road, and (2) the general public and regulators aren't just expecting the autonomous vehicles to be safer than humans, they're unrealistically expecting them to be perfect drivers while being forced to share the road with human drivers.
That's kinda what I meant. It's more about perception than reality. It wouldn't surprise me either if they were already safer. But the perception is that unless they're perfect, they're not safer.
Yep!
The complexity of autonomous vehicles is immense, especially since the general public and regulators are expecting them to be better at making decisions and safer than human drivers. I'd be willing to say that it's orders of magnitude bigger than the difference between reusable and non-reusable rockets.
Is anyone besides the CEOs themselves surprised that the corporate-types vastly underestimated the complexity of a problem they didn't truly understand in the first place?
Agreed 100%. It's important for people in general to understand that any statistical analysis shouldn't be used to decide how to treat/deal with a given individual.
It's a very interesting watch.
There's a difference, though, between algorithms which use non-statistical mathematics, and those that use statistics.
Any time statistical analyses are involved, there are going to be times when it leads to a non-optimal answer. For instance, if an algorithm is based on data like "Steve likes classic rock songs 80% of the time", then that algorithm, when asked "will Steve like this particular classic rock song?" will get it right about 80% of the time. 20% of the time the algorithm will get the answer wrong.
That's not necessarily a biased algorithm, but it is a flawed one, in the respect that it isn't comprehensive enough, or doesn't have enough data to be better.