Peter>The point is, the theory is predictive. In order for an acoustic based system to be useful it must accurately predict what we will find if we open the hive up.<
Peter, you ask about the dance language.  As per the symbolic dance language versus olfactory aspects of the dance,  new information suggests that the symbolic dance language may not be particularly accurate in its predictive part - one way or another, it gets some of the bees into the approximate location of the reward target, then other cues take over.

Since we condition bees to find things like landmines, we have had to evaluate the accuracy of odor-based conditioning.  The independently generated (blind testing by military) ROC curves for probability of detection (probability of a correct detection versus probability of a false detection) for a common explosive at ppb and pptr vapor levels are amazing (https://www.rand.org/pubs/monograph_reports/MR1608.html), nearly perfect. 
To date, I've never seen a statistical analysis or a ROC that rigorously exams the accuracy of the symbolic dance language.  There are lots of studies and reports about the dance language, but these mainly consist of someone decoding the dance, then seeing if any bees show up in the approximate target area - but how many of the bees attending any given dance actually find the intended target versus those that don't find it - that seems to be missing.  If you can find references to studies that take a hard look at dance language's accuracy, especially ROCs, I'd love to see them.  I'm not saying that they don't exist, but there are a lot of assumptions inherent in the decoding.
As per your comment>The point is, the theory is predictive<  

We are in full agreement here.   That's the whole purpose of our Kickstarter project -  Test and TUNE the app for accuracy.
Over nearly 15 years of research, we've shown the acoustic accuracy for each of eight colony condition variables, over and over.  It tends to range from 86-98%. We have done the studies and conducted  statistical analyses evaluating accuracy.  We've achieved these reasonably accurate 'predictions' using higher end equipment  - high definition digital recorders, workstation and benchtop computers.  Also most of our recordings were from N. American colonies of bees.

Our preliminary trials indicate that we should be able to achieve the same on at least the higher end, modern smartphones.  For example, last summer, we got detection of AFB in colonies showing visible signs of AFB on both Android and iPhone.  The AFB discrimination was consistent across both types of phones, various models levels, when employed by several different beekeepers.  So, I'm optimistic.

Since we're not independently wealthy, we can't afford to buy out the phone store to test every brand and model of phone.  And we also know bees have regional accents.  Small changes across much of N. America, much bigger differences in places like U.K., New Zealand, Australia - which should not be surprising.  I've not got a budget to globe hop to test and tune our app.

That's the entire reason that we need a large and widely distributed group of testers at this stage.  Since our systems use AI to analyze the recordings and provide probability predictions, we can tune for these variables, but only if we get lots of recordings, from lots of areas, and lots of different phones.  That's the purpose of the Kickstarter.  

Today we have slightly over 500 signed up for our Kickstarter citizen science project.  That's a lot of people, but still averages less than `10/state.  Preferably, we'd have at least 100 per state, 5000 for US, and others in other countries.   Just as human voice recognition systems build on the user base, so should our app.  500 users is better than 5, 5000 would be even better, 50,000 would help us dial this in better and faster.

 I've had success with citizen science projects before.  In the 1980s, I organized about 250 beekeepers, produced a landscape level study, met GLP requirements, and got it published in SCIENCE!

We need people like yourself who know what they are looking at inside beehives.  Any recording from you, with whatever the app analyses says, and your follow up hive inspection to either accept or reject the analyses results is EXACTLY what we need so that we can TUNE the app on smartphone platforms. 

The point is that we can tune the app, if we get the information.  We'll take all of the correct analyses and recordings and use them to TUNE the AIs.  Similarly we will use  the incorrect analyses to TUNE the apps.   AIs can be trained  to recognize the patterns underlying the correct results and to discriminate them from the patterns in the false readings. 

But this can ONLY be accomplished if we have people taking recordings, running the ~ 15 second recordings, AND conducting a visual inspection, followed by hitting  the upload button, so we get all of the information and use it to refine the app.

We don't expect the app to be spot on this summer, but with a large enough citizen science group, we should have  a better app by this winter (which can be tested and further refined in the southern hemisphere), then roll into next summer with an even better app.

We have to start somewhere.  We're not independently wealthy.  And at my age, I don't have decades left to pull this off.  

Your objections have been heard.  You don't think this will work!  I suggest you read some of our peer-reviewed papers and look to the  data in our patents.  In the meantime, I ask that you give us the benefit of the doubt and stop trying to actively discourage people from participating in our studies.  

We're both old dogs.  Sometimes we can learn new tricks.

If we fail, you'll have the last laugh.  Obviously, I'm risking failure.  But over the last 15 years, I've become convinced that this can and will work - but we need help to achieve success.   Success will be measured by whether this whole approach benefits bee health.

As of tonight, we have 508 backers/participants.  We've only 5 days left to go in the Kickstarter bee health guru project. I'm delighted to have 508, it's a reasonable start for a pilot study.  Keeping my fingers crossed that over the next 5 days, we may get significantly closer to a team of 1000 citizen scientists.






             ***********************************************
The BEE-L mailing list is powered by L-Soft's renowned
LISTSERV(R) list management software.  For more information, go to:
http://www.lsoft.com/LISTSERV-powered.html