March 28, 2024 • 29 mins
Imagine walking into a barn and instead of hearing random clucks and squawks, you're privy to a conversation about likes, dislikes, and even emotional states. This isn't a scene from a children's book; it's the reality that researchers at Dalhousie University are uncovering through the power of artificial intelligence. The idea for this conversation came from a news release titled: Fowl language: Dalhousie University researcher uses AI to crack the code of clucks and unravel chicken chatter secrets, opening the door to an improved quality of life. Suresh Neethirajan is an associate professor and the University Research Chair in Digital Livestock Farming in Dalhousie's Faculty of Computer Science. In this conversation, he explains how he is applying artificial intelligence to chicken sounds from more than 30 poultry farms across North America and Europe, including birds with Avian flu and a wide range of health conditions. Using sophisticated machine learning algorithms, he has translated chickens' nonsensical sounds to specific words, like "Danger!" and "Fox!" For instance, he can now understand expressions of distress or a need for help in diseased chickens, much like a human saying, "I have to go to the hospital." Humourist Bill Carr posted a piece highlighting the ethical questions this raises and Dr. Neethirajan addresses some of them in this conversation.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:01):
Okay, Now, I know whatI'm about to discuss is going to upset
some people. But I just heardyesterday on CBC Radio that at DOW they're
doing AI research into how chickens communicate, and the fact that they actually do
communicate and they can talk about howupset they are, and they talk to
(00:22):
each other about about things they likeand they don't like, and about things
that excite them. And I can'timagine they're any too crazy about how they're
you know, raised. And thenthe fact that their friends and they're eventually
them are getting the you know,they can't. That's got to be a
bad thing. And the more weknow about that, the more we understand
their sentient feelings, the more wegot to take that into consideration. That
(00:44):
voice you hear is humorist, actor, keynote speaker and all round goodfellow Bill
Carr. And Bill and I usedto talk about things like this quite frequently
back when we work together on aradio program. Culled shift for Brains,
say fast shift for brains. Now. That was Bill's response his post to
a recent piece he'd heard on theradio. Now I received the same news
(01:06):
release or pitch from Dalhousie University,and it was titled Foul Language. Dalhousie
Researcher uses AI or artificial intelligence tocrack the code of clocks and unraveled chicken
chatter secrets, opening the door toan improved quality of life. I don't
(01:26):
know if you're like me, butwhen I first heard that, I actually
thought it was some kind of AprilFool's prank given the timing it all.
And I even said as much tothe now communications person who sent out the
email in the first place. Sheassured me this was real and Bill's not
wrong. This does crack open thedoor to some really deep thinking about this
(01:47):
subject. Here was the rest ofwhat Bill had posted, and the implications
are huge. I mean, we'regoing to find out that cow's probably aren't
crazy about what's going on in theirlives, you know, and lambs and
I bring that up at Easter.They can't be crazy about what's going on
in the old cows. Let's noteven talk about that. My point is,
if we find out for sure thatsentient beings are not happy, we
(02:12):
at least owe them the opportunity todiscuss their fate. It's the least we
can do. And that's why I'mall the more committed to vegetarianism because I
know, I know that if youknow, eventually the chickens get together and
really, you know, talk aboutit, they're gonna tell us Colonel Sanders
(02:32):
is not a nice man, justfor example. I know it's going to
upset you, but thats your factsnow. Just to be clear, that
work was being done at Valhousie's Agriculturalcampus in Bible Hill, and it was
being done by the Research Chair inDigital Livestock Farming. And as the release
noted, it's like having a universaltranslator for chicken speech that could detect stress
(02:58):
early and lead to more empathetic treatmentof farm animals. But it's still about
farming and again, really deep thinkingon this one. I'm Sheldon McLeod.
This is my podcast, presented exclusivelyby salt Wire. This is Speaking Out
Loud. Sais NiTi Rajan. I'man Associate professor and University Research Chair in
(03:23):
Digital Livestock Farming at Delahousie University,Canada. So let's talk a little bit
about your research and use of artificialintelligence. First of all, tell me
a little bit about where this projectcame from, how this was a subject
that you wanted to take on.We've been working with farm animals such as
(03:46):
horses, the dairy, cows,swine, pigs, and chickens for almost
ten years. I recently moved fromBahanigan University and research in Atlans to here
in Halifax, Delhasa University. We'vebeen looking at how do we improve the
quality of life of farm animals,how do we develop new solutions for tackling
(04:11):
climate change, improve welfare for farmanimals via artificial intelligence and modern technologies.
So basically I'm from a farming communityas well, so there is a personal
affinity to develop solutions for farm animals. So these are the driving factors behind
(04:33):
this particular project. And chickens inparticular, why was this particular method used
to try and understand how chickens werecommunicating or what it was that they were
saying. So right now, Iknow when we walk into a powtry house
(04:53):
or a poy husbandry, we canhear lots of sounds, cacophony, of
nonsensical languages, different kinds sound.Sound is basically an energy. You know,
we as human beings often you know, tend to misinterpret what other people
say. But once we start payingattention, enough energy is being given to
(05:16):
the attention, then we can actuallyunderstood what is that that they are trying
to communicate and how we can respond. So the question here is how do
we go about using the technology tobetter understand the communication of these birds.
So artificial intelligence and natural language processinghelp us to look at things that are
(05:40):
easily visible, easily understandable. Theyhelp us to unravel dig deeper, find
meaning behind very subtle signs and signaturesfrom these sounds and vocalizations the birds make.
How do you carry out this research? What are you doing to
(06:00):
try and unravel that mysterious communication betweenanother species? Yeah, so the way
we did was we collected a lotof data over past year year and a
half. Here in Pluro Campus atNova Scotia, we have a Atlantic Powertrey
Research Institute where we did a seriesof experiments in collecting audio as well as
(06:25):
visual data. So some of theboats. We also collected data from Asia
as well as Europe and the UnitedStates via a collaborators and from industries.
So the sound, for example,there are many different context The sound is
coming from some boats when they aresick, when they are affected by avian
(06:46):
influence, or even a low pathogenicinfluenza. They peculiarly express certain eternality when
boats are happy. When you introduceas certain playful material, we call them
a enrichment material. It could bea bail up straw. It could be
just a stick like a pitch forthe birds to sit after they feed.
(07:12):
So they suddenly changes their emotional context, their energy, amplitude, tone,
pitch behind the vocalization changes. Ifwe can pay attention to changes in the
frequency and the interval of the feedingthe hygiene protocols, there is a change
in the vocalizations. So we dida variety of experiments. For example,
(07:35):
we were stressing. The birds weresuddenly opening an umbrella. As a usual
stress serve, we played a loud, dark barking sound and wolf sound like
a wild animal using speakers inside thepountry bond. Some birds were actually going
through certain unhealthy, stressful situations.So we recorded all of those songs.
(08:01):
Then we converted these variety of soundsinto some of them into images. So
and then once we get the images, we were looking at the buried signatures
behind each and every pixels inside thoseimages. For example, healthy versus and
healthy bird that is happy versus abit delayed feed frustrated vocalization call. So
(08:26):
some data were converted to images.Some data, what we did was we
converted them into semantics, which meansit's not directly the human language the words
we speak, but we use differentmodels hidden moreco model and other models using
artificial intelligence and deep learning via naturallanguage processing and converted into specific words.
(08:52):
These words are basically representation of thecombination of amplitude energy of the vocalization the
sound that comes from these chickens basedon different context. And then we did
a what we call a topic modelinghow each of those in the context of
the chicken bird vocalization. We werelooking at identifying patterns and clusters of sounds.
(09:18):
How each of these sounds, youknow, relate to specific behavior a
specific context. For example, uniqueword that basically represents either as a feeding
call or a mating call or analerting to predators. So those are the
(09:39):
type of calls we were looking atbased on the words expressed in the form
of vocalization by these chickens. Soagain this doesn't mean we are translating sounds
of these words directly into human language, but we are categorizing these words according
to their meanings and intentions behind theemotions based on the ethology for example,
(10:03):
whether it's alarm call or a foodcall, et cetera. The next step
what we did was we did asentimental analysis and emotion investigation. This is
very commonly then. For example,if we walk into a big, huge
supermarket in the United States, wehuman beings. We are being recorded all
the time, our video as wellas audio, so our the choice of
(10:26):
words we speak among our own maybefamily member, and then when we interact
with the sales agent or the cashierat the supermarket, they also look for
what the what are the type ofwords we choose, and they can quickly,
using natural language processing, map theemotions of our brain in that particular
(10:48):
moment and say, okay, thisperson is going to buy this product or
will close the sale of the purchaseof car, et cetera. They can
make the judgment right away. Sothere is a lot of research happened.
So the inspiration comes from that aswell. How do we use the human
based sentimental analysis and emotion investigation foranimals? Here we were relying on acoustic
(11:13):
futures of the chicken bird vocalizations andwe use artificial intelligence to infer the emotional
states and the well being of theseanimals. For example, stress calls,
they have different acoustic signatures compared tothe acoustic signatures collected from dataime when the
(11:33):
birds are in comfort or contentment.So that's the way we did analysis,
and then we come up with adifferent classes of emotions and different words they
actually are in terms of the birdsspeaking among themselves and also express their frustration,
disappointment, contentment through words in theform of differences in vocalization signatures.
(11:58):
Terribly fascinating, and I have toask, we're not trying to put a
human context to the communication between theseanimals, clearly, but what are the
implications for farming. Anyone who's tendedanimals or perhaps has a pet knows that
their animal is communicating to them.They don't often know exactly what it is
they're saying. They can guess arewe talking about an educated guest here?
(12:22):
What is the endgame? What isit that you're hoping to accomplish here?
So we now know that chickens haveemotions. Chickens also have their own special
way of talking, with certain soundslike clucks, squawks, purs and other
sounds over and over, kind oflike their own words and sentences that means
(12:45):
something really important to them. Forexample, when we analyze over data,
we noticed some words for popping upquite frequently. For example, irrespective of
whether the board is healthy or neutralor sick, words such as going us
coming on frequently all over the place. We also found a certain word called
(13:07):
kneehi nii. These may sound nonsensicalactually even from a human perspective, but
from the word perspective, based onthe context and the environmental conditions they go
through, they really provide very inthe meaning to the conditions of the words
they experience. So it's a bitof head scratcher. What does it?
(13:31):
What does it mean? Knee hi? There are also words like room,
I'm going to next, room,back, other side, So these are
some of the words that are poppingup based on the vocalizations when we use
the semantic modeling to convert the vocalizationinto specific words. So they really have
(13:52):
a pretty consistent way of talking toeach other no matter what's happening around them.
They also share their way of speakinghighlights. So this clearly shows how
clever these birds are in terms ofcommunicating and getting along. So it's just
that we are at the verge oftranslating these sounds into some really meaning.
(14:13):
For example, how do we developintervention strategies? How do we develop animal
welfare protocols? If there is apoint five degree selfious change in the animal
husbandry system, are they feeling happy? Can we be able to decode and
understand the real needs of these birds? Right now? In forming animal management,
(14:37):
we have diary cows, swine,up bigs, horses, and these
birds. So we classify them intotwo groups bigger animals smaller animals. So
it's a injural animal management versus flocklevel or group level management. Right now,
most of the time, especially forturkey broilers and chicken industry, it's
(14:58):
all a group level management interventions arebeing done. So we are proposing,
hey, you know, can webe able to shift from group level to
intodo a level or a subgroup levelmanagement of these form animals such as boats.
So an example is if an LEDlight is being placed in a powertry
bond, the same level of wavelength, intensity and energy of these like may
(15:24):
not be comfortable for all the boats. There may be a group of boats
they try to move around, migrateto a different zone where they feel less
intensity. So we propose instead ofgiving a universal level of intervention strategies,
give a zonal level group level ofintervention strategies, so we make the happy
(15:46):
much more in a positive emotional state. Another example I could give is if
there is a difference in the hygieneprotocol how often they clean the powetry bond
removal of litters and other materials,can we be able to increase the frequency
(16:06):
slightly because it directly it's proportional tothe composition of the gas levels and the
micro oder composition. So that alsoimproves the betterment and the quality of life
of these birds. So timely littlechanges in the life boats and directly proportional
to the better well being of thebirds, and is also related to resource
(16:30):
use utilization as well as climate changein the long run. So if we
are able to save even one percentageof water use or even one percentage of
the feeding and the resource uses inthe on form daily operations, that is
directly proportional to cutting down the emissionsin terms of methane and other gases we
(16:51):
produce during these farming So there area lot of benefits in terms of enhancing
profitability, sustainability as well as animalwelfare in the powdery industry. Is this
the first of this type of researchthat you're aware of in the poultry sector.
Yes, we may be the firstgroup to look at people have looked
(17:12):
at the audio before, but interms of decoding, specifically looking at the
words language and communication, we maybe the first to actually find out the
real meaning and finding decoding the languagebehind the chicken communications. To go back
(17:34):
to turning that audio into an image. Are we talking like a wave file
that people might be accustomed to seeingor is this some other type of image
that's being created. So we callthis a spectrogram. When we look at
it, it has an X andY access, It has a different distribution
and gradient level of RGB values.So when we look and it looks like
(18:00):
an abstract painting from you know,like what we see here behind my screen,
so van go something similar. Butwhen we look at each and every
pixel point it's come out, whythen it has specific meaning behind it.
It has atain stress level buried andso artificial neural network and deeply learning helps
(18:22):
to pick up those and then doa very detailed, insightful trend and patent
analysis to find out what exactly thesebirds are trying to tell us. And
I know this is somewhat speculative,but do you envision a time professor in
the future in which people will knowwhat their dog is saying to them or
(18:42):
their cat. Yes, there area few companies who are already interested in
looking at this in the United States, so we call them as an animal
bioacoustics research. How do we decode? There has been some research done in
whales, sperm whales, and dolphins, et cetera in terms of you know,
(19:06):
we actually can make those sounds inthe way the dolphins can understand and
listen and then they top back.So it has been demonstrated, so there
is a lot of potential to actuallydecode. Perhaps the near future, we
may have a gadget that will recordour voice and convert and try to communicate
(19:27):
with these birds. That sounds likethe stuff of science fiction, but as
you say, you're on that pathand it's due in part because of artificial
intelligence. That's fair. Yes,Well, is there anything else that we
didn't touch on that perhaps has germaineto this conversation? Like, I'm open
(19:48):
for any questions if you have interms of if there's nothing any clarifications or
additional details are needed, more thanhappy to share them. And this is
being carried out at the dal Housiecampus in Troure or Bible Hill. As
you say yes, and it's implicationright at the very heart of it.
It's about farming and raising or growingchickens and understanding what their needs may be.
(20:14):
That's fair to say yes. Andalso how do we provide sustainable profitability
for the powertry industry. It's likea two side of the same coin or
a double squad. Sometimes when wethink of welfare, it may mean that
we are bringing down the exerting additionalpressure bringing down the profitability, but it
is not necessarily the case. Weuse the technology so that we increase the
(20:38):
profitability and sustain the profitability for theproducers at the same time improve the welfare
of the boods. What are theethical implications of in some ways giving rise
to a notion that these are cogentbeings who communicate with one another and the
appropriateness of humans too have them incaptivity and to use them in that way.
(21:03):
Then may be concerns related to,for example, when the technology is
being developed if a company is tryingto sell the technology to the farmers the
farm First of all, the questionwe could ask is does the farm animals
have privacy? Do they have aprivacy? For example, if you put
a camera and constantly monitor them,sometimes maybe they like to shy away.
(21:26):
They like their own quiet zone.They don't like to be monitored. But
how do we know that we arecrossing the boundary in terms of violating their
actual real needs? So a bitof an ethological question we can ask.
The second question is data privacy.Who owns the actual data of these vocalizations?
(21:47):
Is it the farmers? Is itthe company that sells the technology to
the farmers? So how do wecreate a shared mutual benefit? Because in
terms of creating value for all thestakeholders here in terms of watchdog agencies Canadian
Food Inspection Agency and Agriculture and GreyfoodCanada, they are more about how do
(22:10):
we avoid spilling of diseases from thefarm animals to the humans? Getting rid
of zoonotic diseases possibility? So canwe be able to predict the diseases based
on these disturbances anomalies and the smallannuances from these vocalization sound calls. Is
there a possibility Is there a possibilityfor real time transmission of the vocalization of
(22:38):
data from the form to the vetternerianswho may not be able to access the
form in real time. So theseare some of the questions we need to
look at when we develop the technologyand bring multiple stakeholders and realization of this
potential. How do you know thatyour data is accurate? How do you
(23:00):
know that you're interpreting it properly?How do you know that this is really
what you're observing? So we doa variety of behavioral tests, variety of
biochemical tests. For example, wecollect the furthest of the boats, blend
of the chickens at specific time littersand then do biochemical analysis to look at
(23:21):
LAC date and other stress related indicators. So that's a validation study. We
also work with ethologists and behavioral scientistsprovide them video. It's not just sound
alone. We put thermal camera aswell as two dimensional high resolution camera.
For example, the blink of theeye of the chicken can provide a lot
(23:45):
of telltale signs. Unlike human beings, they have vertical eyelids. There are
three islets for chickens, so wecall them a snicitiating membrane. They close
and open like a shutter door.So we also look at how the heat
ymnates and the thermoregulation happens. Usingthermal camera we capture The beak is made
of keratin, which is a proteinaciousin them bring It has two small holes
(24:10):
through which the inhalation exhalation happens.There is a lot of heat transmission happens,
so surrounding the phase region. Thereis also kumba on the surface of
the head, so the variation inthe heat at a specific time point at
particular regions of the facial We alsohave that data in addition to the behavioral
(24:30):
data then we put Sometimes we dobehavioral tests such as a small little plastic
ball in which inside we put larvaewhich is alive. So the birds make
an extra effort to pick through theholes of the plastic ball to take the
larvae, which is a simulation ofthe mental activity, so they may they
(24:52):
feel the emotions shift directly to positive. If they are in negative or neutral,
they become excited and then they slowlyshift. So this combination of behavioral
tests, multimodal data, biochemical analysisand ethological information help us to say,
hey, yes, what we arelooking at is scientifically validated. We are
(25:17):
continuously working with neuroscientists and ethologists tocreate more consensus. For example, when
I utter the word emotion, itbasically starts with the letter E emotion emotion,
which means the energy in motion.But Paul Ekman and Darwin, they
were the earliest scientists who actually decodedand understood and created different classes of the
(25:44):
emotions for human beings. But forform animals, we are not yet there
in terms of having Okay, thisis exactly frustration, this is exactly disappointment
and sadness and happy. So weare still continuously working with our European powers
and some in the United States tocome up the consensus of what exactly constitutes
(26:06):
an emotion for form animals. Howdo we create a baseline and so on
and so forth. So it's awhat can progress in terms of creating a
scientifically valid principle for this. Andyou touched on emotions rather than intelligence,
and that is a distinction that youare very clear on. We could also
(26:29):
call the emotional intelligence EI. Forexample, yes, the boats also have
they have sympathy. If you drivein a rural area, especially the backyard
coops. You might have seen ifthere is an eagle flying down, sometimes
the mother hen makes a peculiar loudsound and the smaller chicks immediately gather around,
(26:51):
and then the mother hen spreads herwings and then try to protect.
So we can see the sympathy andthe protection and the otherly what we call
as a imprinting process, social imprintingprocess for the chicks very clearly, so
they do have emotions. For biggerform animals such as pigs, they demonstrate
(27:14):
certain empathy. Uh it has beenclearly scientifically you know, validated, and
evidence for diary cous also they havethey are much more highly emotionally evolved along
with intelligence. For example, theyhave the ability to when an animal care
taker who walks into the poultry bonor a specific diary bond, they can
(27:38):
dig okay, this person is morefriendly, this person is not more friendly.
So they have the ability to turnaway their face or welcome in their
own way the care taker that walksinto the bond. So we have there
are certain not only addact total butscientifically validated evidences. Some form animals they
(28:00):
have the ability to check out themselvesin the mirror. They just check out
their body. Oh okay, thisis me. So they have that ability
to recognize themselves in the mirror.We call them as mirror tests in their
behavioral science. For form animals,yes, but in the chickens it's not
much fully experimented or standard yet,so we are still looking at how do
(28:22):
we these are For larger animals andsmall animals, we do know that they
have emotions, but in terms ofintelligence, for example, as soon as
the egg hatches, because the industrialproduction animals it is a chickens broilers,
as soon as the egg hatch theycome out, they never have the opportunity
to see or impact with their mother, so there is no opportunity for social
(28:45):
imprinting at all. So they hadto learn everything on their own by pecking
or this is food, this isnot food. So the intelligence comes in
that itself. It's a very basicevolutionary intelligence. But when we look deeper
using these artificial intelligence, we nowhave the possibility to look at the different
(29:07):
classes and grades and colors and levelsof intelligence, and we can actually group
them and classify them in boards suchas chickens. Very fascinating research and Professor
I thank you very much for sharingand talking about that with me today.
Thank you, thank you, Sheldon
Okay, Now, I know whatI'm about to discuss is going to upset
some people. But I just heardyesterday on CBC Radio that at DOW they're
doing AI research into how chickens communicate, and the fact that they actually do
communicate and they can talk about howupset they are, and they talk to
(00:22):
each other about about things they likeand they don't like, and about things
that excite them. And I can'timagine they're any too crazy about how they're
you know, raised. And thenthe fact that their friends and they're eventually
them are getting the you know,they can't. That's got to be a
bad thing. And the more weknow about that, the more we understand
their sentient feelings, the more wegot to take that into consideration. That
(00:44):
voice you hear is humorist, actor, keynote speaker and all round goodfellow Bill
Carr. And Bill and I usedto talk about things like this quite frequently
back when we work together on aradio program. Culled shift for Brains,
say fast shift for brains. Now. That was Bill's response his post to
a recent piece he'd heard on theradio. Now I received the same news
(01:06):
release or pitch from Dalhousie University,and it was titled Foul Language. Dalhousie
Researcher uses AI or artificial intelligence tocrack the code of clocks and unraveled chicken
chatter secrets, opening the door toan improved quality of life. I don't
(01:26):
know if you're like me, butwhen I first heard that, I actually
thought it was some kind of AprilFool's prank given the timing it all.
And I even said as much tothe now communications person who sent out the
email in the first place. Sheassured me this was real and Bill's not
wrong. This does crack open thedoor to some really deep thinking about this
(01:47):
subject. Here was the rest ofwhat Bill had posted, and the implications
are huge. I mean, we'regoing to find out that cow's probably aren't
crazy about what's going on in theirlives, you know, and lambs and
I bring that up at Easter.They can't be crazy about what's going on
in the old cows. Let's noteven talk about that. My point is,
if we find out for sure thatsentient beings are not happy, we
(02:12):
at least owe them the opportunity todiscuss their fate. It's the least we
can do. And that's why I'mall the more committed to vegetarianism because I
know, I know that if youknow, eventually the chickens get together and
really, you know, talk aboutit, they're gonna tell us Colonel Sanders
(02:32):
is not a nice man, justfor example. I know it's going to
upset you, but thats your factsnow. Just to be clear, that
work was being done at Valhousie's Agriculturalcampus in Bible Hill, and it was
being done by the Research Chair inDigital Livestock Farming. And as the release
noted, it's like having a universaltranslator for chicken speech that could detect stress
(02:58):
early and lead to more empathetic treatmentof farm animals. But it's still about
farming and again, really deep thinkingon this one. I'm Sheldon McLeod.
This is my podcast, presented exclusivelyby salt Wire. This is Speaking Out
Loud. Sais NiTi Rajan. I'man Associate professor and University Research Chair in
(03:23):
Digital Livestock Farming at Delahousie University,Canada. So let's talk a little bit
about your research and use of artificialintelligence. First of all, tell me
a little bit about where this projectcame from, how this was a subject
that you wanted to take on.We've been working with farm animals such as
(03:46):
horses, the dairy, cows,swine, pigs, and chickens for almost
ten years. I recently moved fromBahanigan University and research in Atlans to here
in Halifax, Delhasa University. We'vebeen looking at how do we improve the
quality of life of farm animals,how do we develop new solutions for tackling
(04:11):
climate change, improve welfare for farmanimals via artificial intelligence and modern technologies.
So basically I'm from a farming communityas well, so there is a personal
affinity to develop solutions for farm animals. So these are the driving factors behind
(04:33):
this particular project. And chickens inparticular, why was this particular method used
to try and understand how chickens werecommunicating or what it was that they were
saying. So right now, Iknow when we walk into a powtry house
(04:53):
or a poy husbandry, we canhear lots of sounds, cacophony, of
nonsensical languages, different kinds sound.Sound is basically an energy. You know,
we as human beings often you know, tend to misinterpret what other people
say. But once we start payingattention, enough energy is being given to
(05:16):
the attention, then we can actuallyunderstood what is that that they are trying
to communicate and how we can respond. So the question here is how do
we go about using the technology tobetter understand the communication of these birds.
So artificial intelligence and natural language processinghelp us to look at things that are
(05:40):
easily visible, easily understandable. Theyhelp us to unravel dig deeper, find
meaning behind very subtle signs and signaturesfrom these sounds and vocalizations the birds make.
How do you carry out this research? What are you doing to
(06:00):
try and unravel that mysterious communication betweenanother species? Yeah, so the way
we did was we collected a lotof data over past year year and a
half. Here in Pluro Campus atNova Scotia, we have a Atlantic Powertrey
Research Institute where we did a seriesof experiments in collecting audio as well as
(06:25):
visual data. So some of theboats. We also collected data from Asia
as well as Europe and the UnitedStates via a collaborators and from industries.
So the sound, for example,there are many different context The sound is
coming from some boats when they aresick, when they are affected by avian
(06:46):
influence, or even a low pathogenicinfluenza. They peculiarly express certain eternality when
boats are happy. When you introduceas certain playful material, we call them
a enrichment material. It could bea bail up straw. It could be
just a stick like a pitch forthe birds to sit after they feed.
(07:12):
So they suddenly changes their emotional context, their energy, amplitude, tone,
pitch behind the vocalization changes. Ifwe can pay attention to changes in the
frequency and the interval of the feedingthe hygiene protocols, there is a change
in the vocalizations. So we dida variety of experiments. For example,
(07:35):
we were stressing. The birds weresuddenly opening an umbrella. As a usual
stress serve, we played a loud, dark barking sound and wolf sound like
a wild animal using speakers inside thepountry bond. Some birds were actually going
through certain unhealthy, stressful situations.So we recorded all of those songs.
(08:01):
Then we converted these variety of soundsinto some of them into images. So
and then once we get the images, we were looking at the buried signatures
behind each and every pixels inside thoseimages. For example, healthy versus and
healthy bird that is happy versus abit delayed feed frustrated vocalization call. So
(08:26):
some data were converted to images.Some data, what we did was we
converted them into semantics, which meansit's not directly the human language the words
we speak, but we use differentmodels hidden moreco model and other models using
artificial intelligence and deep learning via naturallanguage processing and converted into specific words.
(08:52):
These words are basically representation of thecombination of amplitude energy of the vocalization the
sound that comes from these chickens basedon different context. And then we did
a what we call a topic modelinghow each of those in the context of
the chicken bird vocalization. We werelooking at identifying patterns and clusters of sounds.
(09:18):
How each of these sounds, youknow, relate to specific behavior a
specific context. For example, uniqueword that basically represents either as a feeding
call or a mating call or analerting to predators. So those are the
(09:39):
type of calls we were looking atbased on the words expressed in the form
of vocalization by these chickens. Soagain this doesn't mean we are translating sounds
of these words directly into human language, but we are categorizing these words according
to their meanings and intentions behind theemotions based on the ethology for example,
(10:03):
whether it's alarm call or a foodcall, et cetera. The next step
what we did was we did asentimental analysis and emotion investigation. This is
very commonly then. For example,if we walk into a big, huge
supermarket in the United States, wehuman beings. We are being recorded all
the time, our video as wellas audio, so our the choice of
(10:26):
words we speak among our own maybefamily member, and then when we interact
with the sales agent or the cashierat the supermarket, they also look for
what the what are the type ofwords we choose, and they can quickly,
using natural language processing, map theemotions of our brain in that particular
(10:48):
moment and say, okay, thisperson is going to buy this product or
will close the sale of the purchaseof car, et cetera. They can
make the judgment right away. Sothere is a lot of research happened.
So the inspiration comes from that aswell. How do we use the human
based sentimental analysis and emotion investigation foranimals? Here we were relying on acoustic
(11:13):
futures of the chicken bird vocalizations andwe use artificial intelligence to infer the emotional
states and the well being of theseanimals. For example, stress calls,
they have different acoustic signatures compared tothe acoustic signatures collected from dataime when the
(11:33):
birds are in comfort or contentment.So that's the way we did analysis,
and then we come up with adifferent classes of emotions and different words they
actually are in terms of the birdsspeaking among themselves and also express their frustration,
disappointment, contentment through words in theform of differences in vocalization signatures.
(11:58):
Terribly fascinating, and I have toask, we're not trying to put a
human context to the communication between theseanimals, clearly, but what are the
implications for farming. Anyone who's tendedanimals or perhaps has a pet knows that
their animal is communicating to them.They don't often know exactly what it is
they're saying. They can guess arewe talking about an educated guest here?
(12:22):
What is the endgame? What isit that you're hoping to accomplish here?
So we now know that chickens haveemotions. Chickens also have their own special
way of talking, with certain soundslike clucks, squawks, purs and other
sounds over and over, kind oflike their own words and sentences that means
(12:45):
something really important to them. Forexample, when we analyze over data,
we noticed some words for popping upquite frequently. For example, irrespective of
whether the board is healthy or neutralor sick, words such as going us
coming on frequently all over the place. We also found a certain word called
(13:07):
kneehi nii. These may sound nonsensicalactually even from a human perspective, but
from the word perspective, based onthe context and the environmental conditions they go
through, they really provide very inthe meaning to the conditions of the words
they experience. So it's a bitof head scratcher. What does it?
(13:31):
What does it mean? Knee hi? There are also words like room,
I'm going to next, room,back, other side, So these are
some of the words that are poppingup based on the vocalizations when we use
the semantic modeling to convert the vocalizationinto specific words. So they really have
(13:52):
a pretty consistent way of talking toeach other no matter what's happening around them.
They also share their way of speakinghighlights. So this clearly shows how
clever these birds are in terms ofcommunicating and getting along. So it's just
that we are at the verge oftranslating these sounds into some really meaning.
(14:13):
For example, how do we developintervention strategies? How do we develop animal
welfare protocols? If there is apoint five degree selfious change in the animal
husbandry system, are they feeling happy? Can we be able to decode and
understand the real needs of these birds? Right now? In forming animal management,
(14:37):
we have diary cows, swine,up bigs, horses, and these
birds. So we classify them intotwo groups bigger animals smaller animals. So
it's a injural animal management versus flocklevel or group level management. Right now,
most of the time, especially forturkey broilers and chicken industry, it's
(14:58):
all a group level management interventions arebeing done. So we are proposing,
hey, you know, can webe able to shift from group level to
intodo a level or a subgroup levelmanagement of these form animals such as boats.
So an example is if an LEDlight is being placed in a powertry
bond, the same level of wavelength, intensity and energy of these like may
(15:24):
not be comfortable for all the boats. There may be a group of boats
they try to move around, migrateto a different zone where they feel less
intensity. So we propose instead ofgiving a universal level of intervention strategies,
give a zonal level group level ofintervention strategies, so we make the happy
(15:46):
much more in a positive emotional state. Another example I could give is if
there is a difference in the hygieneprotocol how often they clean the powetry bond
removal of litters and other materials,can we be able to increase the frequency
(16:06):
slightly because it directly it's proportional tothe composition of the gas levels and the
micro oder composition. So that alsoimproves the betterment and the quality of life
of these birds. So timely littlechanges in the life boats and directly proportional
to the better well being of thebirds, and is also related to resource
(16:30):
use utilization as well as climate changein the long run. So if we
are able to save even one percentageof water use or even one percentage of
the feeding and the resource uses inthe on form daily operations, that is
directly proportional to cutting down the emissionsin terms of methane and other gases we
(16:51):
produce during these farming So there area lot of benefits in terms of enhancing
profitability, sustainability as well as animalwelfare in the powdery industry. Is this
the first of this type of researchthat you're aware of in the poultry sector.
Yes, we may be the firstgroup to look at people have looked
(17:12):
at the audio before, but interms of decoding, specifically looking at the
words language and communication, we maybe the first to actually find out the
real meaning and finding decoding the languagebehind the chicken communications. To go back
(17:34):
to turning that audio into an image. Are we talking like a wave file
that people might be accustomed to seeingor is this some other type of image
that's being created. So we callthis a spectrogram. When we look at
it, it has an X andY access, It has a different distribution
and gradient level of RGB values.So when we look and it looks like
(18:00):
an abstract painting from you know,like what we see here behind my screen,
so van go something similar. Butwhen we look at each and every
pixel point it's come out, whythen it has specific meaning behind it.
It has atain stress level buried andso artificial neural network and deeply learning helps
(18:22):
to pick up those and then doa very detailed, insightful trend and patent
analysis to find out what exactly thesebirds are trying to tell us. And
I know this is somewhat speculative,but do you envision a time professor in
the future in which people will knowwhat their dog is saying to them or
(18:42):
their cat. Yes, there area few companies who are already interested in
looking at this in the United States, so we call them as an animal
bioacoustics research. How do we decode? There has been some research done in
whales, sperm whales, and dolphins, et cetera in terms of you know,
(19:06):
we actually can make those sounds inthe way the dolphins can understand and
listen and then they top back.So it has been demonstrated, so there
is a lot of potential to actuallydecode. Perhaps the near future, we
may have a gadget that will recordour voice and convert and try to communicate
(19:27):
with these birds. That sounds likethe stuff of science fiction, but as
you say, you're on that pathand it's due in part because of artificial
intelligence. That's fair. Yes,Well, is there anything else that we
didn't touch on that perhaps has germaineto this conversation? Like, I'm open
(19:48):
for any questions if you have interms of if there's nothing any clarifications or
additional details are needed, more thanhappy to share them. And this is
being carried out at the dal Housiecampus in Troure or Bible Hill. As
you say yes, and it's implicationright at the very heart of it.
It's about farming and raising or growingchickens and understanding what their needs may be.
(20:14):
That's fair to say yes. Andalso how do we provide sustainable profitability
for the powertry industry. It's likea two side of the same coin or
a double squad. Sometimes when wethink of welfare, it may mean that
we are bringing down the exerting additionalpressure bringing down the profitability, but it
is not necessarily the case. Weuse the technology so that we increase the
(20:38):
profitability and sustain the profitability for theproducers at the same time improve the welfare
of the boods. What are theethical implications of in some ways giving rise
to a notion that these are cogentbeings who communicate with one another and the
appropriateness of humans too have them incaptivity and to use them in that way.
(21:03):
Then may be concerns related to,for example, when the technology is
being developed if a company is tryingto sell the technology to the farmers the
farm First of all, the questionwe could ask is does the farm animals
have privacy? Do they have aprivacy? For example, if you put
a camera and constantly monitor them,sometimes maybe they like to shy away.
(21:26):
They like their own quiet zone.They don't like to be monitored. But
how do we know that we arecrossing the boundary in terms of violating their
actual real needs? So a bitof an ethological question we can ask.
The second question is data privacy.Who owns the actual data of these vocalizations?
(21:47):
Is it the farmers? Is itthe company that sells the technology to
the farmers? So how do wecreate a shared mutual benefit? Because in
terms of creating value for all thestakeholders here in terms of watchdog agencies Canadian
Food Inspection Agency and Agriculture and GreyfoodCanada, they are more about how do
(22:10):
we avoid spilling of diseases from thefarm animals to the humans? Getting rid
of zoonotic diseases possibility? So canwe be able to predict the diseases based
on these disturbances anomalies and the smallannuances from these vocalization sound calls. Is
there a possibility Is there a possibilityfor real time transmission of the vocalization of
(22:38):
data from the form to the vetternerianswho may not be able to access the
form in real time. So theseare some of the questions we need to
look at when we develop the technologyand bring multiple stakeholders and realization of this
potential. How do you know thatyour data is accurate? How do you
(23:00):
know that you're interpreting it properly?How do you know that this is really
what you're observing? So we doa variety of behavioral tests, variety of
biochemical tests. For example, wecollect the furthest of the boats, blend
of the chickens at specific time littersand then do biochemical analysis to look at
(23:21):
LAC date and other stress related indicators. So that's a validation study. We
also work with ethologists and behavioral scientistsprovide them video. It's not just sound
alone. We put thermal camera aswell as two dimensional high resolution camera.
For example, the blink of theeye of the chicken can provide a lot
(23:45):
of telltale signs. Unlike human beings, they have vertical eyelids. There are
three islets for chickens, so wecall them a snicitiating membrane. They close
and open like a shutter door.So we also look at how the heat
ymnates and the thermoregulation happens. Usingthermal camera we capture The beak is made
of keratin, which is a proteinaciousin them bring It has two small holes
(24:10):
through which the inhalation exhalation happens.There is a lot of heat transmission happens,
so surrounding the phase region. Thereis also kumba on the surface of
the head, so the variation inthe heat at a specific time point at
particular regions of the facial We alsohave that data in addition to the behavioral
(24:30):
data then we put Sometimes we dobehavioral tests such as a small little plastic
ball in which inside we put larvaewhich is alive. So the birds make
an extra effort to pick through theholes of the plastic ball to take the
larvae, which is a simulation ofthe mental activity, so they may they
(24:52):
feel the emotions shift directly to positive. If they are in negative or neutral,
they become excited and then they slowlyshift. So this combination of behavioral
tests, multimodal data, biochemical analysisand ethological information help us to say,
hey, yes, what we arelooking at is scientifically validated. We are
(25:17):
continuously working with neuroscientists and ethologists tocreate more consensus. For example, when
I utter the word emotion, itbasically starts with the letter E emotion emotion,
which means the energy in motion.But Paul Ekman and Darwin, they
were the earliest scientists who actually decodedand understood and created different classes of the
(25:44):
emotions for human beings. But forform animals, we are not yet there
in terms of having Okay, thisis exactly frustration, this is exactly disappointment
and sadness and happy. So weare still continuously working with our European powers
and some in the United States tocome up the consensus of what exactly constitutes
(26:06):
an emotion for form animals. Howdo we create a baseline and so on
and so forth. So it's awhat can progress in terms of creating a
scientifically valid principle for this. Andyou touched on emotions rather than intelligence,
and that is a distinction that youare very clear on. We could also
(26:29):
call the emotional intelligence EI. Forexample, yes, the boats also have
they have sympathy. If you drivein a rural area, especially the backyard
coops. You might have seen ifthere is an eagle flying down, sometimes
the mother hen makes a peculiar loudsound and the smaller chicks immediately gather around,
(26:51):
and then the mother hen spreads herwings and then try to protect.
So we can see the sympathy andthe protection and the otherly what we call
as a imprinting process, social imprintingprocess for the chicks very clearly, so
they do have emotions. For biggerform animals such as pigs, they demonstrate
(27:14):
certain empathy. Uh it has beenclearly scientifically you know, validated, and
evidence for diary cous also they havethey are much more highly emotionally evolved along
with intelligence. For example, theyhave the ability to when an animal care
taker who walks into the poultry bonor a specific diary bond, they can
(27:38):
dig okay, this person is morefriendly, this person is not more friendly.
So they have the ability to turnaway their face or welcome in their
own way the care taker that walksinto the bond. So we have there
are certain not only addact total butscientifically validated evidences. Some form animals they
(28:00):
have the ability to check out themselvesin the mirror. They just check out
their body. Oh okay, thisis me. So they have that ability
to recognize themselves in the mirror.We call them as mirror tests in their
behavioral science. For form animals,yes, but in the chickens it's not
much fully experimented or standard yet,so we are still looking at how do
(28:22):
we these are For larger animals andsmall animals, we do know that they
have emotions, but in terms ofintelligence, for example, as soon as
the egg hatches, because the industrialproduction animals it is a chickens broilers,
as soon as the egg hatch theycome out, they never have the opportunity
to see or impact with their mother, so there is no opportunity for social
(28:45):
imprinting at all. So they hadto learn everything on their own by pecking
or this is food, this isnot food. So the intelligence comes in
that itself. It's a very basicevolutionary intelligence. But when we look deeper
using these artificial intelligence, we nowhave the possibility to look at the different
(29:07):
classes and grades and colors and levelsof intelligence, and we can actually group
them and classify them in boards suchas chickens. Very fascinating research and Professor
I thank you very much for sharingand talking about that with me today.
Thank you, thank you, Sheldon
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
The Herd with Colin Cowherd
The Herd with Colin Cowherd is a thought-provoking, opinionated, and topic-driven journey through the top sports stories of the day.