Wisdom of the crowds vs. electronic sensor

In 1907, Francis Galton, half-cousin of Charles Darwin, who was an anthropologist  and a statistician, observed a curious phenomenon that is called the “Wisdom of the crowd”. During this year, he assisted a game where the players were supposed to guess the weight of an ox just by looking at it. Individually, the responses were generally wrong and far from the right answer. But when he took all the 787 estimations, he observed that the median of those (it means that 50% of the participants answered above and 50% under) was the right answer with only 1% of error.

With this observation he published a paper in Nature in 1907 which was the first about the Wisdom of the crowd.

In the context of the second Biosensors week that focused on force sensing, our team, ScalingHuman, decided to compare the accuracy power of the crowd to an electronic scale to determine the weight of an object.

To do so, we designed a protocol that enabled us to collect the crowd measures as well as the scale ones. We first prepared three different weighed boxes that looked exactly the same, so that their appearance would not interfere in our measures. We then masked the eyes of the person we were testing in order to reduce even more biais. After that, we handed the person a 1kg box, and told her it weighed 1kg in order to “tare” her, just like an electronic scale. At that point the experiment could start: we handed her the other boxes, asked how much she estimated their weight and wrote her response down. We handed 4 times the same box in a random order in order to test people’s precision. In order to reach a so called crowd, we reproduced the protocol on 35 different persons.

Finally, we compared both results and observed that the median and the average of the estimations are relatively close to the scale measurement. The results are shown in the graph below :

In those two graphics , the curves represent the average/median of the estimation made by the scale (red) and the crowd (blue).

We can clearly observe an evolution of the difference between the 2 estimations, it increases as the objects get heavier. We can observe that the average and the median of the estimations are quite close to the measurement made by the scale  for light object. As for  the 3kg object, the estimation were in average and with the median overestimated (of about 2kg). Our results allow us to make the assumption that the crowd manages to estimate light weights (<1kg) easier than heavy weights (>1kg) .

However, our results could heavily be contested because of the following bias:

The size of our crowd was for sure too small, Galton tested almost 800 persons we tested only 33, and that was not a sufficient sample. Furthermore, we tested only 3  random people on the street, the others being scientific students. That makes a rather homogeneous crowd, while the crowd should be heterogeneous to be relevant. Then, we also observed that the large volume of boxes could have led to biased weight perceptions.

The most significant bias in our experiment seems to stem from the way we collected our data. In fact, for each tested person we reproduced the experiment 4 times for each box, which may be too much. Then, we gave them the different weights randomly, and alternating heavy and light boxes may have altered people’s perception. A study on the influence of the order we follow to give our boxes could also be really interesting.

http://webinet.cafe-sciences.org/articles/etrange-perspicacite-collective-1/

https://www.youtube.com/watch?v=xtuh5zTa7mQ the video that introduced us to the concept of wisdom of the crowd

: https://www.youtube.com/watch?v=iOucwX7Z1HU : Explanation to the concept and application of wisdom of the crowd

Resources: F. Galton, Vox Populi, Nature, no1949  75, March 7, 1907.

1. Weight perception and the haptic size–weight illusion are functions of the inertia tensor. Amazeen, Eric L.; Turvey, M. T. Journal of Experimental Psychology: Human Perception and Performance, Vol 22(1), Feb 1996, 213-232.

Follow us on twitter : @SyllogikiS

Written by : Noémie Paillon, Lucile Szpiro, Léon Grillet and Aurélien DIEHL.

Cockroach perception VS iPhone Application

On the 01/25/2016, we began the second week of Biosensor project. The subject for this week was “Forces”, and it includes mechanic vibrations, sound vibrations, electrostatic forces, magnetic forces etc...

What force did we choose ?

We decided to work on sound wave. Before explaining what we did, it is important to figure out what a sound wave is to better understand our project.

An acoustic wave is a mechanic disturbance (compression then expansion) which is propagate into a media without material transport. The period is the time that the waves take to complete one cycle (it means to go back in the same position). The frequency is the number of times, that a phenomenon (here, the sound wave) occurs per time unit. The Humans can hear sound with frequency between: 200 - 20,000 Hz.

Illustration of acoustic wave on a microphone

An interesting video to understand what a sound is :

What was the project ?

We choose to work on rusty red cockroaches as biological sensor and an iPhone with Dicibel Ultra app as electric sensor. So our question is “How do both cockroaches & an iPhone sensor react to sound?” The aim is to determine which is the best sound sensor to detect high-frequency sounds.

At the beginning of the project, we wanted to identify both the sound frequencies which scares cockroaches and the one that attracts them. But why did we choose cockroaches (it’s so dirty!) ? This is because they are sensitive to vibrations! They have many receptors on their legs and their antennas which are very responsive to vibrations. Indeed, they can sense small vibration in the air. That’s why it is so difficult to catch them. Moreover, it is original to study this hated organism.

To emit our high-frequency sounds, we used “Frequency Sounds generator”, an app that can produce sounds from 20 Hz to 20,000 Hz.  To receive these sounds, we use Decibel Ultra, another smartphone application. Why this paticular application? Because it is accessible to a large number of person (it is free on appStrore & Android) and it measures sound level and its frequency with high precision. The first idea was to test sound with a frequency higher than 20 000Hz named ultra sound but our phone application cannot produce there frequency.

How did we the experiment ?

Figure 1 : Visual protocol

To measure if cockroaches are sentitive to high-frequency sound, we place only one cockroach into a transparent plastic tube because they cannot climb on wall, it allow us to see the movement of the cockroach as well. Then we placed the sound transmitter in front of the tube and the camera facing the recipe to film all the tube because we want to analise their film to calculate cockroach’s speed. Concerning the electronic sensor, we just replace the tube by the iPhone on the cockroach position marks and we note the value given by the application.

What are the results ?

Figure 2. Sensor responses to varying frequency

The values given by the iPhone receptor ( “Decibel Ultra App”) were quite precise, except for the value given at a 20 000Hz frequency. Indeed, at 20 000Hz, the receptor doesn’t detect the sound wave accurately and gives a value clearly underrated to reality (accuraty represents if a measure correponds to the reality).

For our cockroaches, our results were difficult to interpret. At all given frequencies, our cockroaches didn’t have clear responses. Sometimes they didn’t move at all, others they were repulsed, then attracted, then repulsed again before stopping, etc… As such, we did an average of their speeds at each frequency to create this graph.

What is the conclusion ?

Thanks to these results, we can deduce that cockroaches are less efficient than our IPhone to detect high-frequency sounds. But, it was really hard to detect and understand their reaction in front of sound waves. Their behavior was really different from one individual to another. That said, we can see on the graph that our application was not really good to detect sound with frequency greater than 20 000 Hz.

Why our experience didn’t work as expected ?

Our results, difficult to interpret may they be, are certainly due to some of these causes. First, our cockroaches were placed near crickets in the shop we bought them from which may explain their lack of particular observable behavior when in contact with sound waves. During the experiment, we could hear noise coming from the street, so our biological sensor wasn’t solely in contact with our sound frequencies. Also, with time, they were getting used to receiving high-frequency sounds, so they reacted less and less, though we did try to minimize this bias by exposing them in a random order to our souds. Additionally, we stressed them by moving them from one environment to another. We also discovered that their behavior was dependant on light conditions by doing our protocol during night time and during day time which caused clearly different behavior in our cockroaches.

Want to know more about the subject ?

An scientific article by S. Shaw on the "Detection of airborne sound by a cockroach 'vibration detector': A possible missing link in insect auditory evolution" Journal of Experimental Biology. (en)

Video about "A saptial robot built on the basis of cockroach sensors !" by the NASA

By Fairouz Gzara, Juliette Delahaye, Nathan Touboul and Nicolas Silva - @The_SpiderSpeed

Ear Force One project - Biosensors

The Ear Force One project is a one week teamwork proposal based on interdisciplinary knowledges, involving biology, physics and computer skills.

“Is human ear better at detecting a sound than an electronic sound sensor?”

                                        

The global aim of Biosensors project is to compare a biological sensor with an electronic one. Therefore, through this project, we wanted to compare the capacity  of human ears and of FC-04 sound sensors to detect sounds at different distances.

To do this amazing project, we realized two different experiments. In the first one, we tested our biological sensor: the human ear. We considered that human ear is one of the most essential organ of the human body because it allows us to detect sounds and keep balance. That is the reason why we chose this spectacular organ. In the second experiment, we test our electronic sensor: the FC-04 sound sensor. This type of electronic sensor was created in order to be more accurate than ears. In others words, we thought that it would be interesting to compare these both sensors on their capacity to detect a sound coming from long distances.

Our experiment was divided into three main parts : collecting the data, interpreting them, and finally criticize them.

Collecting the data

We created the same experimental setup for our both systems (biological and electronic). It was first constituted by a computer spreading a sound. In front of this computer, we made marks on the floor every meters until 12 meters.

For the biological experiment, we asked to volunteers to take place at the 12 meters’ mark. We spread a sound and the participant had to tell us if he was hearing the sound or not . We were waiting for a simple answer : yes or no. After their answer, the participant had to step forward and to take place at 10 meters. We continued the same experiment until the participant reached 0 meters. We repeated this whole biological experiment with 11 persons. If you want to know more about how our sense of hearing works, you can look at this video.

A participant at 12 meters from the sound source

For the electronic experiment, the setup was exactly the same except of the participant that we replaced by the electronic sensor. The sensor was connected to a LED which was turning on each time the sensor was detecting a sound. If you want to reproduce our electronic system here is a simple video tutorial. Or if you just want to have fun with it don’t hesisate to check here.

So, we were able to collect data to determine the detection thresholds of our both sensors according to the distance. If you want to know more about threshold of hearing you can click here or read this article.

We collected our data and for a better visualization, we decided to represent them on an histogram.

Interpreting the data

We compared the sound detection of human ear and of FC-04 sound sensor according to the distance. We could observed on the histogram that from 7.20 m all participants heard the sound. However, at some points like 1.2m or 5.2m,  some participants had difficulties to hear the sound. It can be explained by the different ages of the participants. If you want to know more about age effect in hearing you can check this article. For our electronic system, we observed that our sound sensor was either detecting every sounds in the surrounding area or either detecting no sounds whatever the distance. It could be explained by the fact that the calibration of the sensibility threshold of the sound sensor wasn’t at all precise.

Critical analysis of the project

To conclude, we can say that the principal problem in our project was the sound sensor potentiometer . Indeed, the potentiometer was very sensitive which was problematic because it didn’t allows us to have precise sensibility thresholds.

During the experiment, we tried to change the potentiometer with an other. But we obtained results that were similar to the first one: the sensibility remained unchanged !

If we had to improve this project, we would like first to change the sensor and use another that sends analog values  so allows a more precise set up.

On an other hand, we would like to record the sound got by the sensor during the experiment because one of our mentors told us that sometimes, a LED could be on but too little for us to see it.

If you want to know more about ear sensibility and arduino sound sensor, here are some links:

• Letmeknow (2013). Letmeknow, [Tuto] Utiliser un capteur de son avec le Arduino Uno[online]. janvier 2016. http://letmeknow.fr/blog/2013/11/15/tuto-utiliser-un-capteur-de-son-avec-le-arduino-uno/

• "Arduino Frequency Detection"(2016). Instructables.com [Online]. janvier 2016. http://www.instructables.com/id/Arduino-Frequency-Detection/

• "Autocorrleation Code"(2016) . Instructables.com [Online] Janvier 2016. http://www.instructables.com/id/Reliable-Frequency-Detection-Using-DSP-Techniques/step3/Autocorrleation-Code/

• “/chapter: Son-Qui-Provoque-Lumiere / Arduino". Fr.flossmanuals.net [Online]. Janvier 2016. http://fr.flossmanuals.net/arduino/son-qui-provoque-lumiere/

Follow us on Twitter : @Ear_Force_One

Written by Alexandra PERRON, Maryam ARIF, Margaux BIEUVILLE & Isabelle JOUSSET

Biosensors: focusing on sensing light

Week 1 of Biosensors course focused on examining how biological and electronic systems perceive light in their environment.

Students designed their projects in a way to try to compare specific characteristics of both biological and electronic sensor, by carefully designing the experimental setup, choosing the appropriate biological model organism as well as the creation of the electronic device.

Credits: Wikipedia page on lenses and optics

Comparing angular precision in Daphnia's and LDR's response to light

By Gaspard Baudrin, Manon Curaudeau, Xander Hampel and Isabelle Jousset - @AngulightFDV

Hello to you lovely people ! This blogpost is about a one-week experiment our team conducted during the Biosensors course, which consists of a wonderful month giving birth to a myriad of very diverse scientific group-projects about biosensors. It was organised by our interdisciplinary bachelor program “Frontiers of Life Sciences” hosted by the CRI!!!

Where the light shines, the shrimp goes!

(credits to https://youtu.be/XmajoRnBW30)

Changes in light intensity are known to have a strong effect on Daphnia's behavior.

Like many species in the marine environment, Daphnia - a small planktonic crustacean - migrates to upper luminous water layers during the night and return to deeper layers during the day, before sunrise.

This behavior is due to their transparency: even though transparency protects Daphnia from being seen by their predators, it enhance the exposition of their molecular components (proteins, DNA) to solar irradiation. As a result, they react to UV light with negative phototaxis : they move away from the sunlight.

But Daphnia also perform positive phototaxis, which means they move towards sources of light. During the night, when the aggressive sunrays can't harm them, Daphnia return to surface seeking for their main source of food : photosynthetic phytoplankton. Phytoplankton floats near the surface for they need sunlight to accomplish photosynthesis. Daphnia use light from stars and moon as a trusted guide towards their food. All they need to do is swim upwards where the beaming, harmless moonlight spreads all over the ocean.

Fig 1 - Daphnia’s compound eye: It is made of thousands of photoreceptor units, and enable them to detect different light intensities. (credits to http://arthropoda.southernfriedscience.com/wp-content/uploads/2011/02/daphniaeye.jpg)

We were wondering how sharpened these receptors are, and therefore how precisely Daphnia can trend towards a located source of visible light. We decided to answer that question, or at least to get more insights about it, by comparing Daphnia and an electronic light sensor (LDR standing for Light Dependent Resistor). We focused on angular precision,

LDR is a light-controlled variable resistor, which means that its resistance varies with light intensity. Indeed, its resistance decreases when the light intensity increases. This features enables the device to detect light and measure the luminosity

So far, we supposed that the precision of both Daphnia and LDR will increase when the intensity rises. We assumed that the angular precision of the electronic sensor will overtake the precision of the Daphnia.

To investigate this, we ran two different experiments, a biological and an electronic one. We had six light intensity tested for both (including no light and maximal light we could produce with a white LED).

For the biological part, we first measured the average position of ten daphnia (disposed in a thin layer of water) at each of the six different values of intensity we exposed them to. Then we compared the resultant angular position with the value of the angle of the light source.

In the meantime, we tested the precision of the electronic sensor we’ve engineered in such a way that it takes measurements every 1° in 180° and gives us the angle of the highest perceived value of intensity.

Fig 2 - Our experimental setup:  We exposed both daphnia and a LDR (disposed on a movable arm) to six different light intensities. The light was produced by a LED that was disposed randomly but at a known angle (135°). Then we measured the angle between the source of light and the position of both Daphnia (their average position) and the LDR.

We expected the Daphnia to move towards the light, but they actually moved away from it. Anyway, we still could exploit our results : indeed, to avoid the light Daphnia have to go as far as possible from it. Their ability to do so still depends on the precision of their light-sensitive organs. Instead of moving towards the maximum of intensity they would move towards the minimum of intensity perceived. And where would the minimum of intensity be ? Well it would be at the farthest point of the position of the light source, which means at the exact opposite point !

Fig 3 - Data analysis for daphnia:  The average position of daphnia was calculated (cross) using ImageJ. The measured angle is between the light ray and the ray passing through the averaged position. It is the same wether the daphnia go to the light or if they escape from it.

Many studies have been conducted regarding the behavior of Daphnia when exposed to UV light, but little is known about the effect of visible light on marine zooplankton. Our results show that for Daphnia, an increase in light intensity goes along with a decrease in the measured angle. It means that the accuracy of daphnia photoreceptors increases. Our graph shows that a limit of accuracy of 10° is reached from 500 a.u.

Regarding the the electronic device, it seems that accuracy doesn’t depends on light intensity, but more measurements need to be done. Plus, the precision of the sensor depends mostly on the precision of the mobile arm.

Fig 4 - Our results: Angle in function of intensity for both tested devices.

Comparing the response of two different sensors to different light intensities appeared as the primordial aspect of our study. From our results, it appears that the LDR device is precise but not accurate, and that it’s accuracy does not depends on light intensity. As for daphnia’s accuracy, it depends on light intensity, and their photoreceptors are not super precise.

References:

The positively phototactic reaction of Daphnia magna Straus: a contribution to the understanding of diurnal vertical migration J Ringelberg - Netherlands Journal of Sea Research, 1964 - Elsevier

Interaction of polarized light and turbidity in the orientation of Daphnia and Mysidium TH Waterman - Zeitschrift für vergleichende Physiologie, 1960 - Springer

Individual swimming behavior of Daphnia: effects of food, light and container size in four clones Stanley I. Dodson, Shanna Ryan,Ralph Tollrian and Winfried Lampert

Phototaxis in water fleas (Daphnia magna) is differently influenced by visible and UV light

U. C. Storz, R. J. Paul

Some physical factors influencing the feeding behavior of daphnia magna straus  J. W. McMahon

OpenFlowers

OpenFlowers

By Nicolas Silva, Clara Haas, Margaux Bieuville and Amélie Bouissou

The main goal of the biosensor weeks is to compare biological and electronical sensors. To do so we decided to use a phototransistor as an electronic sensor for light. For the biological sensor, our first idea was to study the opening of flowers depending on the intensity of luminosity. The flower with the most amazing movement reacting to light is the nympheas as you can see here: https://www.youtube.com/watch?v=dem8ZDXycR4. The Oxalis triangularis has also impressive leaf movement with variation of light: https://www.youtube.com/watch?v=7mSBTkKqqOU.

Sadly, we are in January so finding plants with flowers sensitive to different luminosities is not really easy.  

We therefore decided to change the scale of our experiment and study the movement of chloroplasts in cells at different light intensities. Chloroplasts are little compartments in plant cells that allow photosynthesis to happen, therefore they are sensitive to different types and intensities of light. When there is little or no intense light, the chloroplasts migrate to the surface of the cell to capture the most light in order to do photosynthesis. On the contrary when there is too much light the chloroplasts migrate against the side cell walls to avoid photodamage. You can read more about this phenomenon in this Nature article: “Chloroplast avoidance movement reduces photodamage in plants” http://www.nature.com/nature/journal/v420/n6917/full/nature01213.html 

We compared the reactions of a biological sensor in the form of chloroplasts and an electronic sensor, a phototransistor, in different light intensities.  We put leaf cells on a microscope slide and put them for 20 minutes under either maximum light, medium intensity light or dark.  Simultaneously, we were measuring this light intensity with the phototransistor, an electronic component connected to an Arduino board then to a computer.  We then observed the leaf cells under the microscope.  Since chloroplasts contain chlorophyll, a fluorescent pigment, we were able to distinguish the chloroplasts thanks to fluorescence microscopy.  We took pictures and counted how many chloroplasts were touching the side cell walls.  You can see our experimental protocol on the following image:

Visual protocol of our experiment and data analysis

 

    Our results were very interesting since they allowed us to evaluate the efficiency and precision of both sensors.  In the dark, 4 chloroplasts on average touch the side membranes whereas in medium intensity light approximately 8 chloroplasts are touching and in maximum light, 11 chloroplasts were touching on average.  Similarly to what we expected, we observed that the more light the leaf cells were exposed to, the more chloroplasts touched the side cell membranes, as if to avoid being in the center of the cell.  This matches the previous research done on the topic as you can read in the Nature article.  The phototransistor, the electric component, returned non-linear values, meaning that the value that it measured did not correspond to the intensity sent.  For the intensity measured in the dark or in full light, the values were pretty accurate but for medium intensities, the value measured was not proportional to the current sent.

    Through these two experiments, we learned that though chloroplast layout is a good indicator of light intensity, it is not very precise.  From one cell to another, the number of chloroplasts can vary and their response to light will not be identical.  The response time, meaning the time from initial light exposition to the reading of the result, is also very long since we have to prepare the leaf cell sample, expose it for 20 minutes, observe the cells on the microscope then count the number of chloroplasts.  On the other hand, the phototransistor is connected to a computer, and the measured value can be seen almost instantly.  However, the phototransistor is not fully reliable since there were several problems.

Plant cell magnification x63 with mRFP filter after 20 minutes exposition with maximum intensity light

References:

Kasahara, Masahiro et al. "Chloroplast Avoidance Movement Reduces Photodamage in Plants." Nature.com. Letters to Nature, 19 Dec. 2002. Web. 19 Jan. 2016.

Islam, Sayeedul, and Shingo Takagi. "." NCBI. Plant Signaling and Behavior, 5 Feb. 2010. Web. 19 Jan. 2016. Co-localization of Mitochondria with Chloroplasts Is a Light-dependent Reversible Response

Chloroplast Photorelocation Movement: A Sophisticated Strategy for Chloroplasts to Perform Efficient Photosynthesis, Noriyuki Suetsugu et al. (2012)

The Noun Project: https://thenounproject.com/