Their "powerful AI engine" is almost certainly just humans. It might have a few off-the-shelf components like face detection but most of what they claim to do is just so easy to outsource that almost all companies do it. If there is any delay between the system observing a suspect behaviour and the student being told to correct it then they are definitely using humans.
An institution using a service like this is a huge red flag. You should take it as an indicator of a low quality administration if not a low quality institution.
As an engineering problem this task is hard. Ryan Calo (Prof of Law, UW) once presented a fascinating bit of research on trying to automate something as simple as fining someone for speeding. Given perfect information how do you build the system? If someone exceeds the speed limit for 1 second, do you fine them? If everyone around the person is exceeding the speed limit do you use the same rules? If someone oscillates between just above and just below the speed limit, how many times do you fine them? If someone exceeds the speed limit and stays there does this result in fewer fines? How do you square the code written with the law as written? The problems are so extensive it may be that application of rules like this require human level judgement. Proctoring an exam may turn out to be an AI-complete problem.
> Their "powerful AI engine" is almost certainly just humans.
I wonder how parents feel that an anonymous foreign, most probably male, can watch their daughter in her room working on her exams and she can't opt-out else she'll be treated like a criminal.
Wonder if at some point we'll find a zip file shared among employees with screenshots of students from one of these outsourced proctoring vendors.
> I wonder how parents feel that an anonymous foreign, most probably male, can watch their daughter in her room working on her exams and she can't opt-out else she'll be treated like a criminal.
No legal recourse, and cultural differences. An adult male in Japan propositioning a 13 year old for sex is legal in their current system, whereas that doesn't really fly in America.
It's definitely a face detection thing for some of them. My spouse had to deal with all of her black students being hectored nonstop by the software because it had a hard time reading dark skin faces and was constantly interrupting them to accuse them of not looking in the correct direction.
I've seen something related actually play out 20 years back. My dad needed to drive through a particular toll booth multiple times one weekend. He figured out that same weekend you could go pretty fast and the ezpass would still register. So he went fast every time.
A week later he got a half dozen letters in the mail all at once:
* warning do not speed
* warning (+fine)
* final warning (+fine)
* ezpass revocation (+fine)
* etc.
I think he ended up arguing that he didn't get the first warning before the others, and they agreed to waive the fines and roll back to the first warning. I.E. human judgement applied after the fines "fixed" the issue.
In any case: toll booths are a workplace, there's people there, go slow.
This is getting really far off topic, but the solution is clearly to make speeding fines a "dollars per mile per hour, per hour" system. It should scale continuously both with speed over the limit and with time spent speeding. Programmers who focus on discrete systems are too prone to forget about real analysis. ;)
So, on a 65mph highway, driving at 66mph for an hour has the same fine as driving at 125mph for one minute? (Minus speedup and slowdown time.) Don't think a linear scale would make sense.
I'm not sure if any scale would make sense, though. The true thing to be fined for should be "driving unsafely", of course, whether that means driving too fast, braking suddenly, swerving, changing lanes without warning, etc. The thing is, the unsafety of all those things depends in very large part on the cars around you and sometimes on the road. (Driving at 100mph on a straight, flat freeway with no cars nearby is safer than weaving between a bunch of 65-70mph cars to maintain a speed of 80mph.)
But it would be really hard to come up with an objective algorithm to calculate how unsafe a rule violation is, and even harder to implement it without a vast array of cameras. So ... they implement an enforceable set of rules even if it's not a good one. (On a related note, it bothers me that driving with more than some arbitrary blood alcohol level is illegal, but driving after e.g. staying awake all night is not, even though the latter is worse[1]. Likely this is partly because checking BAC can be done fairly directly with cheap equipment, while checking how recently someone slept is... I dunno, there might be ways to mostly do that with good equipment, but I assume it's currently impractical.)
This is clearly not the solution, unless it is accompanied by changes to speed limits in North America- which itself introduces a bunch of complexity and negative side-effects.
It won't solve speed limits as a civic policy issue, but it does solve the specific thing the question was posed to ask: the morass of trouble created by trying to write discrete rules for continuous phenomena.
Or you could just set the speed limit appropriately, at a level that most drivers will naturally conform to.
This optimizes for both expediency and safety, but, sadly, not for revenue, or for readily-available probable cause to pull over essentially anyone the police want to pull over.
> Or you could just set the speed limit appropriately, at a level that most drivers will naturally conform to.
The problem is that such a limit would not be a safe one, you would basically be replacing the limit with a sign that posts the rate that people like to drive at.
>The problem is that such a limit would not be a safe one,
According to who? The minority who wants the limit set elsewhere. Things like "what speed is safe and reasonable" are matters of social consensus. The "experts" can pontificate all they want and the vocal minority can gripe all they want but the median or average person and society at large is always going to be right on matters of social consensus. It's a tautology.
The problem is that posting a lower limit doesn't really stop most people from driving faster. Then, some people will obey the speed limit and they become hazards. Other people might believe the posted speed limit tells you about the speed you can expect cars to go on the road. It doesn't.
The main safety increase I can see is with respect to commercial trucking. Not all vehicles have the same acceleration and braking ability. But the fact that people just ignore the speed limit is a hard problem.
So, yes, as far as is practical, that's the idea. We just post the speed people are driving at.
There is a real difference, in both actual safety and perceived safety, between divided roads and undivided roads with the same number of lanes. Wider lanes seem (and are) safer than narrower lanes. See https://en.wikipedia.org/wiki/Traffic_calming for more.
Reasonable drivers conform their driving habits to the environment.
Unreasonable drivers should not be allowed to drive.
> There is a real difference, in both actual safety and perceived safety, between divided roads and undivided roads ... Wider lanes seem (and are) safer than narrower lanes.
> [Wiki] Traffic calming can include the following engineering measures ... Narrowing traffic lanes ... Converting one-way streets into two-way streets forces opposing traffic into close proximity, which requires more careful driving
What is happening here? Did the engineers forget that they were supposed to optimize for safety, and instead optimized for getting people to slow down? Does the net effect of this improve safety, or worsen it?
If everyone drives as fast as they can while meeting some perceived safety level, then does that mean these efforts are unlikely to affect safety and merely to slow everyone down? Is the best set of measures the maximally deceptive set, which looks as dangerous as possible while being as safe as possible?
> If everyone drives as fast as they can while meeting some perceived safety level, then does that mean these efforts are unlikely to affect safety and merely to slow everyone down?
No. It means that people are usually wrong when they judge the safety level of their driving speed on a nice clear road that happens to have pedestrians next to it or trying to cross it, but are more accurate at judging safety when they are primed to expect obstacles, steering challenges, other actors moving not in parallel with them on the road itself.
My understanding is that speed limits have been largely unchanged for the past 50+ years. Car safety, on the other hand, has increased by a great deal.
It seems likely to me that speed limits are conservative when it comes to safety.
I suspect it has more to do with the incentives faced by those who set the speed limits.
I don't know exactly how it goes. If it's chosen (or heavily influenced) by politicians, I suspect politicians can win some votes by saying they'll improve safety by lowering speed limits (particularly around schools or other places where children might be); while they'd be less likely to win as many votes by saying they'll raise the limits (exposing them to the risk of their opponents calling them reckless/irresponsible). If it's chosen by non-elected officials, that's less of an issue, but something like it may still be there; or it may be that raising the limit and then there being a fatal accident will damage your career, while lowering the limit and irritating everyone will not damage your career.
The speed limit is (in the USA) supposed to adhere to the 85th percentile rule, that being the speed at which people drive on a free-flowing road with no traffic and no enforcement. Most freeways have speed limits that are probably closer to the 0th percentile than the 85th percentile. I know driving the speed limit on some roads in my neck of the woods would be utterly hazardous.
People like to drive at the rate that is safe. Humans are pretty good at measuring risk.
I've driven on many roads in countries that don't have any speed limits. The overwhelming majority are driving a speed that makes sense given the road, driving conditions, etc.
>Ryan Calo (Prof of Law, UW) once presented a fascinating bit of research on trying to automate something as simple as fining someone for speeding. [...]
Most of the problems you've listed only exist because of expectations caused by inconsistent/lenient enforcement by human police officers. People don't seem to have a problem with strictly enforced rules in finance. eg. "your bank account can't below zero without triggering a overdraft".
> People don't seem to have a problem with strictly enforced rules in finance. eg. "your bank account can't below zero without triggering a overdraft".
I think a lot of people have problems with strictly enforced rules in finance. Especially when banks re-order daily transactions in order to maximize overdraft fees.
>I think a lot of people have problems with strictly enforced rules in finance
They have problems with it because the rules cost them money, not because it's hard to understand or rigidly enforced. If the rules just resulted in a "transaction declined" they wouldn't really care either way. For instance, I don't think anyone thinks that it's unreasonable for your debit card to get declined if you don't have enough money.
In the case referenced, there were 3 transactions on one day, 2 of $10 & 1 of $20. User has $20 in account. Do you take the true chronological order of transactions (10/10/20) & charge 1 fee, or do you go highest to lowest (20/10/10) & charge 2 fees?
A bank executive would likely argue why it's 'fair' to charge 2 fees because a business day is the relevant period for a bank (closing at end of day, etc.) & that method of accounting is mentioned on page 85 of the checking account TOS that a user signed.
Many consumers would argue that the bank's relevant period is meaningless, especially in the age of computers. They would also argue that it is unfair to lay out complex rules like this in an opaque way because of the asymmetrical information advantage that a bank has. They wouldn't complain that the rule was enforced per se, they would complain that the rule (which is plainly anti-consumer) exists.
Same is true of this cheating stuff - I think in general, students want a fair platform for grading. They just want that platform to be actually fair.
> They would also argue that it is unfair to lay out complex rules like this in an opaque way because of the asymmetrical information advantage that a bank has. They wouldn't complain that the rule was enforced per se, they would complain that the rule (which is plainly anti-consumer) exists.
Doesn't that justify my point? If we go back to the speeding example, if the rule was that you can't go over the speed limit (within the capabilities of the measuring device), then everyone would drive a little more slowly. The only reason people 5-10 miles above the speed limit is that 5-10 miles over is generally accepted to be "fine". If anything strict enforcement of speed limits reduce the amount of room for abuse by law enforcement (eg. pulling over someone for going 1 mph over).
Also, at the risk of victim blaming, maybe it isn't such a good idea to have your deposits/withdraws lined up on the same day? Deposits can get delayed/withheld, and withdraws can be moved up unexpectedly. Leaving zero days between a deposit and a withdraw is just asking for trouble. I agree that reordering the transactions from largest to smallest is probably greed motivated, but at the same time expecting it to behave differently is optimistic at best and foolish at worst. It's the equivalent of relying on undefined behavior in programming (eg. assuming that reading one byte after the end of an array wouldn't cause a fault).
> Also, at the risk of victim blaming, maybe it isn't such a good idea to have your deposits/withdraws lined up on the same day? Deposits can get delayed/withheld, and withdraws can be moved up unexpectedly. Leaving zero days between a deposit and a withdraw is just asking for trouble.
I'm not sure you understood the example that GP gave; there was no mention of withdrawals. The idea was that you have $20 in your account, you buy something for $10, then later buy something else for $10, and then later buy something for $20, and the bank reverses the order of applying the transaction and says that you made two purchases after your account was empty, so you get charged two overdraft fees.
Their "powerful AI engine" is almost certainly just humans. It might have a few off-the-shelf components like face detection but most of what they claim to do is just so easy to outsource that almost all companies do it. If there is any delay between the system observing a suspect behaviour and the student being told to correct it then they are definitely using humans.
An institution using a service like this is a huge red flag. You should take it as an indicator of a low quality administration if not a low quality institution.
As an engineering problem this task is hard. Ryan Calo (Prof of Law, UW) once presented a fascinating bit of research on trying to automate something as simple as fining someone for speeding. Given perfect information how do you build the system? If someone exceeds the speed limit for 1 second, do you fine them? If everyone around the person is exceeding the speed limit do you use the same rules? If someone oscillates between just above and just below the speed limit, how many times do you fine them? If someone exceeds the speed limit and stays there does this result in fewer fines? How do you square the code written with the law as written? The problems are so extensive it may be that application of rules like this require human level judgement. Proctoring an exam may turn out to be an AI-complete problem.