08/31 - Striking Saturday!

Title: Psychology Study — Harvard University (20 minutes) | PANDA
Requester: Yi-Chia Chen [A1AUC55UAIUHBG]
TurkerView: [ $12.59 / hour ]
Description: Look at pictures and make simple judgments (20 minutes)
Duration: 360 Min
Available: 0
Reward: $2.00
Qualifications:

• Tex Img DoesNotExist
• HIT approval rate (%) GreaterThan 95,
• Total approved HITs GreaterThanOrEqualTo 50,
• Location EqualTo US,

[3KHO6UOMI8QH25GAQLR209JLN4OBPD]​

 

Title: Which Sentences Confirm Stereotypes?` [MANY HITS INCOMING]! | PANDA
Requester: MIT [A1WZ5548AAD4H5]
TurkerView: [ $3.00 / hour ]
Description: Can you determine which sentences confirm stereotypes? There will be many HITs uploaded soon!
Duration: 20 Min
Available: 7
Reward: $0.07
Qualifications:

• HIT approval rate (%) GreaterThan 95,
• Location EqualTo US,

[33D52F3J44HJEZO34F58J2PI233WVO]​

 

Title: Which Sentences Confirm Stereotypes?` [MANY HITS INCOMING]! | PANDA
Requester: MIT [A1WZ5548AAD4H5]
TurkerView: [ $3.00 / hour ]
Description: Can you determine which sentences confirm stereotypes? There will be many HITs uploaded soon!
Duration: 20 Min
Available: 7
Reward: $0.07
Qualifications:

• HIT approval rate (%) GreaterThan 95,
• Location EqualTo US,

[33D52F3J44HJEZO34F58J2PI233WVO]​

 

Title: Which Sentences Confirm Stereotypes?` [MANY HITS INCOMING]! | PANDA
Requester: MIT [A1WZ5548AAD4H5]
TurkerView: [ $3.00 / hour ]
Description: Can you determine which sentences confirm stereotypes? There will be many HITs uploaded soon!
Duration: 20 Min
Available: 7
Reward: $0.07
Qualifications:

• HIT approval rate (%) GreaterThan 95,
• Location EqualTo US,

[33D52F3J44HJEZO34F58J2PI233WVO]​

 

Cinnabon and red wine, not bad..

 

Title: Transcribe Approximately 12-Minute Audio File [D23 190831-7] | PANDA
Requester: PR Brigade [A3CFR5JHIC6U80]
TurkerView: [ $ / hour ]
Description: You will be transcribing a portion of a conversation that has been split into two different audio files.
Duration: 480 Min
Available: 1
Reward: $4.00
Qualifications:

• HIT approval rate (%) GreaterThanOrEqualTo 100,

[3CTCX9NXB5MRA7HZVJJGSEN79G4LJO]​

 

Title: Transcribe Approximately 12-Minute Audio File [CAM 190831-3] | PANDA
Requester: PR Brigade [A3CFR5JHIC6U80]
TurkerView: [ $ / hour ]
Description: You will be transcribing a portion of a conversation that has been split into two different audio files.
Duration: 480 Min
Available: 1
Reward: $4.00
Qualifications:

• HIT approval rate (%) GreaterThanOrEqualTo 100,

[3BW6RX62FE5JZRV7HI8PVYAF47BLE4]​

 

W. Warner Burke [AWWFA10ZKC3YG] Short survey about cross-racial interactions in the workplace (5-7 min or less) - $0.50 | PANDA

Low

Unrated

Unrated

$8.82 / hour
00:03:24 / completion time

​

Pros:

Cons:

Qualifications: Total approved HITs >= 10000; HIT approval rate (%) >= 99; Location EqualTo US;

To read W. Warner Burke's full profile check out TurkerView!​

 

Title: Answer a survey(~ 3 minutes) | PANDA
Requester: Ali Tezer [A1UG8S77ZBD1KJ]
TurkerView: [ $11.81 / hour ]
Description: Fill out a survey in less than 3 minutes for $.31
Duration: 60 Min
Available: 11
Reward: $0.31
Qualifications:

• Total approved HITs GreaterThanOrEqualTo 1000,
• HIT approval rate (%) GreaterThanOrEqualTo 95,
• Location In US,
• Exc: [298719400-168174] DoesNotExist

[349I5E7LVC9ZADH72T6YGG3RKCC9JK]​

 

In the hypothetical scheme proposed, would some "fairest algorithm imaginable" necessarily be any "less fair" than the current scheme with panda-polling? Could such a redesign be more-or-less almost equally "fair" but maybe more efficient in some ways?

From what little I understand of networked software design, clients constantly "polling" can be relatively inefficient at scale. Whether that's true here or not, a question arises as to whether it even matters in this case.

We have swarms of panda requests flying every 700ms or whatever from thousands of panda'ing clients. That works, and it's OK. Maybe it's the best way to operate, or maybe it isn't. Maybe it hits the right points of efficiency without the drawbacks of the alternatives. Or maybe it falls short of a hypothetical ideal design that could be more efficient in some ways. But even if the current scheme might be imperfect, nobody's really complaining about it (except maybe me, but I'm not really complaining, I'm just musing about hypotheticals).

Could a redesign be more "efficient" and thereby "better" (at least in some abstract sense of engineering design, if maybe not in actual benefit to us), if it could reduce that network traffic, by changing the request methodology from "constant polling" to something more "event driven" – by having a client register a request and "go idle", until the service on AWS determines something the client needs to know (and then notifies the client)?

Consider this sequence:

Client registers a request with a short list of HITs, and goes idle.
(Each item in the request can be an ID or a search query.)

An MTurk service on AWS monitors HIT availability, and (using some hypothetically "fair"-ish algorithm to determine the allocations of "who gets what and when"), matches the client's request to one or more available HITs when they become available.

Immediately after receiving that notification, the client script must automatically reply quickly (within a "few seconds", some short time-limit defined in API rules) to acknowledge and accept the HIT. — Then the server returns a final confirmation to make sure, and then the client has the HIT.

Otherwise (if the client script is disconnected, or just not running anymore, or for whatever reason it fails to reply within the agreed "few seconds" time-limit), then the service on AWS will reallocate and offer the HIT to some other worker's client script instead. ... (If a too-slow client's response arrives "too late", the server returns a refusal because the HIT has now been reallocated to someone else.)

No more than now, e.g. having PandaCrazy polling and accepting HITs in a separate window (in "background", maybe muted) while we're working. — Likewise the proposed "request and wait" client can be put in background, so it needn't interrupt a user who's busy working.

Interesting point that I hadn't considered.

 

It's for CT.

 

I'll distract....

 

I was looking for a TL;DR gif but this one is fine too.

 

CT & I could take that discussion off-thread and go to PM's ... but currently I'm not doing PM's on this forum, sorry.

Anyway, I've already said enough about it ... 'twas just an idea ... I probably don't want to get into a much longer discussion about it.