0:00

This podcast is supported by the Icahn

0:02

School of Medicine at Mount Sinai, one

0:04

of America's leading medical research schools. How

0:08

will advances in artificial intelligence transform

0:10

medical research and medical care? To

0:13

find out, we invite you to read

0:15

a special supplement to Science Magazine, prepared

0:17

by Icahn Mount Sinai, in partnership with

0:19

Science. Just visit our

0:22

website at science.org and search

0:24

for Frontiers of Medical Research-Artificial

0:26

Intelligence. On May 1st

0:28

and May 2nd, Icahn Mount Sinai

0:30

and the New York Academy of Sciences

0:33

will be convening a major symposium in

0:35

New York City on the

0:37

new wave of AI in healthcare. For

0:40

more information and

0:42

to register, please

0:44

visit events.nyas.org/AI Health.

0:47

That's events.nyas.org/AI

0:51

Health. The

0:53

Icahn School of Medicine at Mount Sinai. We

0:55

find a way. This

1:03

is a science podcast for May 3rd, 2024. I'm

1:06

Sarah Krustepe. First up on

1:08

the show, a roundup with newsletter editor,

1:10

Christy Wilcox. We talk about the oldest

1:12

ice ever found, how to

1:14

combat mosquitoes with our own microbes,

1:17

and about how recent research into

1:19

conservation efforts shows that broadly, they

1:22

seem to be working. Next, we're

1:24

evaluating seismic hazard maps. Leah

1:27

Saldich and colleagues used modern prediction

1:29

maps on past quakes and

1:31

found a mismatch. We talk about

1:33

where this bias comes from and how to fix it.

1:41

Now we have newsletter editor, Christy Wilcox.

1:43

We're going to talk about some recent

1:45

stories from the newsletter, Science Advisor. Hi,

1:47

Christy. Welcome back to the podcast. Hi, Sarah. Great

1:49

to be here. We're going to

1:52

first talk about your pick, the one that you said,

1:54

oh, I can talk your ear off about this one.

1:57

This is something that is really cool because you

1:59

had an ink. about it in an

2:01

earlier, earlier newsletter and now you get to follow

2:03

up and it's on mosquitoes which is maybe something

2:06

people are starting to think about this time of

2:08

year. I certainly am starting to think about it

2:10

as the weather gets a little nicer and you

2:12

start to want to go outside. Yeah and so

2:15

this is about how we can use our own

2:17

microbes to fight off these bitey guys. Yeah

2:20

so last year I wrote

2:22

about a study where they were looking at

2:24

the, they call them volatile chemicals, the stuff

2:26

that comes off of your skin. Especially

2:29

your smell but not

2:31

all of these have a distinct smell to

2:33

people. It's not all us right? We're not

2:36

creating all these odors. No, no. That was

2:38

interesting and so what they were looking at

2:40

is they were looking at the

2:42

different chemicals coming off of skin

2:44

that are made by microbes on our skin

2:47

and they wanted to know if some of these

2:49

were either repellent or attractant to

2:51

mosquitoes that are looking for a place to land

2:54

and feed and so they found

2:56

that there were several chemicals that you know

2:58

either repelled or attracted mosquitoes but there was

3:00

one in particular a version of lactic acid

3:02

that seemed to have a pretty potent effect

3:05

in terms of helping mosquitoes find their

3:07

spots that they can bite and feed

3:09

and drink from. This wasn't on everybody

3:11

this was just on some people? Well

3:14

the amount of it on different people

3:16

varied based on their skin microbiome and

3:18

so that the idea is that this

3:20

is produced by certain bacteria that naturally

3:22

live on our skin. Some

3:25

people have more of these bacteria than others

3:27

or versions of these bacteria that produce more

3:29

of it than others. Some people do smell

3:32

better to mosquitoes out there. They do, they

3:34

really do. It is absolutely a thing

3:36

that some people are mosquito magnets and some

3:38

people aren't. So what

3:40

they were sort of said in this

3:42

earlier paper is that hypothetically this means

3:45

that you could manipulate the microbes on

3:47

a person's skin and make them more

3:49

or less attractive to mosquitoes. So

3:51

bring us to today, which is the goal of

3:53

art. So then what they did is they took

3:55

a couple of bacteria that naturally

3:58

live on skin and they engineered

4:00

them, they deleted an enzyme involved

4:02

in producing this chemical. These

4:05

engineered microbes produce much, much less

4:07

of it. And then they tested

4:10

whether or not mice

4:12

with the engineered versions of these microbes

4:14

were more or less attractive to mosquitoes

4:16

than mice with the regular versions. And

4:19

lo and behold, getting rid of this

4:21

enzyme, getting rid of this compound essentially,

4:24

made it so the mice were less attractive

4:26

to mosquitoes. It worked! And the bacteria were

4:28

okay, even though they weren't producing that special

4:31

chemical anymore? Yeah, yeah, this version of lactic

4:33

acid or whatever, it didn't seem to harm

4:36

them in any noticeable way. And they

4:38

were able to colonize the skin and

4:40

do fine. And they actually made these

4:42

mice less attractive to mosquitoes

4:44

for two weeks. Wow. Yeah. I mean,

4:46

this was not just a temporary effect.

4:49

If you think about mosquito repellents, I

4:51

mean, I'm out there spraying

4:53

DEET on me every few hours, right?

4:55

So this is a potentially long lasting,

4:57

and they called it a living mosquito

4:59

repellent, which I thought was just the

5:01

coolest. Now I'm starting to fantasize about

5:03

microbes that we can smear on our

5:05

skin that protects us from the sun,

5:08

that protects us from mosquitoes. It's

5:10

so cool. Maybe that's a little far away

5:13

though. Yeah, yeah. I mean, possibilities

5:15

are endless, I would say. All

5:17

right. Okay, so we're only doing

5:19

that one animal story today, and it

5:21

wasn't even a cute animal story. Sorry.

5:24

But we have an extreme story. So

5:26

this is the oldest ice

5:28

ever dug up. And

5:31

what we can learn from something that is

5:33

just so, so old. How old, Christy? How

5:35

old? Six million years. I was

5:37

surprised by that number. I didn't know you

5:40

could have ice that was six million years

5:42

old. That blew my mind. Yeah. So where

5:44

was it then? So it was in Antarctica,

5:46

but it was not in the usual place

5:49

that they get Antarctic cores. So the idea

5:51

is they normally when they drill

5:53

for ice in Antarctica, they're sort

5:55

of going in the interior of the

5:58

island. That's where they have their campsites. That's

6:00

where they're doing their work, right? Yeah, and they've

6:02

got these big glaciers or whatever. They drill

6:04

down, get a big long core. But

6:07

in this case, they got what they called blue ice,

6:09

and it was actually from the coast. And

6:12

so this blue ice from the coast is

6:15

older, but it's less of a clear

6:17

record. So you're not

6:19

getting that nice straightforward core where you can say,

6:21

this one is this old, this one is this

6:23

old, this one is this old. They

6:25

had to sort of figure out how to date these

6:28

properly. And they're a little bit harder to date and

6:30

they're a little bit harder to do all

6:32

that with. But the payoff is

6:34

that they're more than twice as old as

6:36

the one that they had been getting. What

6:39

can we learn from looking so far back with ice? What's

6:42

captured in this ice record that we haven't been

6:44

able to see before? Are we looking at isotopes

6:46

or what's in there? So what they

6:48

have is they have actually trapped

6:50

bits of air, so little bubbles

6:53

of air that is trapped in this ice.

6:55

And they can measure things like the CO2

6:57

in that air, so the carbon dioxide that

6:59

is in that air. And

7:02

one of the things that they found, for instance,

7:04

is that when you had this

7:06

giant temperature drop, right, when you

7:09

had the ice ages coming on

7:11

and the world got really cold,

7:14

you didn't have a huge drop

7:16

in carbon dioxide. And

7:19

what they said is that that meant that carbon

7:21

dioxide is really powerful, right? The fact

7:23

that it didn't take much of a

7:25

drop for that temperature

7:27

to change and that cooling effect

7:29

to occur. So they're able to

7:32

extract all of this amazing information from

7:34

these bubbles of air and such that is trapped

7:36

in the ice. Is this probably the

7:38

oldest ice we're going to get or is

7:40

there older ice out there somewhere? So I

7:42

don't know if they can get much more

7:44

ancient than six million years in terms of

7:46

this ice, but I know that they only

7:48

got really small samples this time around, so

7:50

they're going back and they're hoping

7:52

to get like really big samples so that they

7:55

can do more work with it. So

7:57

one more story and then I'm going to have to let you go.

7:59

This is actually... from a science paper

8:01

that was published last week on

8:04

wildlife conservation and they basically

8:06

asked a really big question. Are

8:08

these small efforts that are kind of going

8:11

on all over the globe, people saving

8:13

different populations or different ecosystems

8:15

in a piecemeal fashion, are

8:17

they working to stop or even

8:20

reverse the decline in biodiversity?

8:22

Are they worth it? So Christy

8:25

wants to answer. Yes, yes, I'd

8:27

love it. I'm leaving it there,

8:29

turning my mic off. Okay. No,

8:33

what was really cool about this study is,

8:35

I mean, we've had lots of studies that

8:37

look at individual projects and try to figure

8:39

out if this project is working. And

8:42

what this study did was really take all

8:44

of those studies and say,

8:48

on the whole, if we look at these

8:50

projects, are we more often

8:52

than not succeeding? And

8:55

when they looked at these projects, they didn't

8:57

just say like, oh, did biodiversity increase or

8:59

did the conservation work? What

9:01

they had to show is that it

9:03

improved over what would have happened or

9:06

some sort of control. So general

9:08

increases in biodiversity or increases

9:10

in plant cover that happened

9:12

everywhere, not just where this

9:15

special effort was being made, didn't

9:17

count. So like you didn't get

9:20

credit for just a general increase. It had

9:22

to have that control in there. That's what

9:24

makes it really powerful and really accurate. I'm

9:27

super excited. Positive news about the world

9:29

is always good. Is the recommendation then

9:32

to keep doing it the way we're doing or

9:34

do it harder? Like what

9:36

does it mean? Like just keep doing what we're doing?

9:38

I think it means that these efforts,

9:40

even if they seem like they're expensive

9:43

or they seem like they're hard to

9:45

arrange or they're hard to negotiate, they're

9:47

worth it. It is

9:49

absolutely worth it, and we need to keep

9:52

trying basically. And don't give

9:54

up is the message that I

9:56

got. That's a good message. All right. Chrissy,

9:58

what else would people be – interested in reading

10:00

from the news. Just this week,

10:03

we've had some really interesting ones. I mean, there was

10:05

one about how an AI

10:08

transcription service actually

10:11

hallucinates essentially. So

10:13

they've shown the chat GPT

10:15

that it'll make things up.

10:17

It'll make up fast, right? Like it'll just

10:20

pull things out of the ether.

10:22

Well, apparently a transcription version, a

10:25

version that is supposed to be listening

10:27

to your audio and then turning it

10:29

into words is hearing

10:31

things. And not just things,

10:34

really often, a lot of the

10:36

time they are inappropriate or like

10:38

racist or terrible things. Oh, wow.

10:40

So you don't want that doing

10:42

live transcription for you at your event.

10:45

That sounds like. Yeah. We

10:47

are not ready for live transcription. That is for

10:49

sure. And then another one that I thought was

10:51

really interesting from this week is that we

10:53

often think of like bloodhounds or Sherman

10:56

shepherds as these like super sniffers. And

10:58

so that's why we train them to

11:00

do bomb sniffing and all of the,

11:02

you know, tracking down cadavers

11:05

or whatever. Turns out

11:07

all dogs basically have the same sense

11:09

of smell, at least physiologically. And so

11:12

the only differences appear

11:14

to be in the motivation or

11:17

ability to be trained. Yeah. Like,

11:19

so you could be taking our

11:21

poodles, our labradoodles out to the

11:23

airport to have them be little

11:26

curly bouncy balls of fun and also bomb

11:28

snippers. See, I'm voting for pugs. I want

11:30

to see a bunch of little bomb sniffing

11:32

pugs. They sound like they can smell really

11:35

good. Trades to little

11:37

snorters like running around. All

11:39

right, Christy. Thanks so much for coming on the show.

11:41

Always fun to have you. And I'm glad we got

11:43

to squeeze a few animals there

11:46

in the end. Christy Wilcox

11:48

is the newsletter editor for Science

11:50

Advisor. Thanks, Christy. Thank you, Sarah.

11:54

Stay tuned for a chat with researcher

11:56

Leah Saldich about using modern seismic

11:58

prediction methods on past. makes. Seismic

12:09

hazard assessments are used to set

12:11

up building codes and to even

12:13

plan earthquake damage mitigation

12:16

strategies. But how good

12:18

are these assessments that are used for

12:20

such practical purposes? How good

12:22

are they actually at predicting earthquake

12:24

intensity? This week in Science

12:26

Advances, Leah Saldich and colleagues looked at

12:29

this odd disconnect between what

12:31

the seismic hazard assessments say and

12:33

what happens in the real world. Hi

12:35

Leah, welcome to Science Podcast. Hi,

12:38

thank you so much for having me,

12:40

Sarah. Sure. So what brought this disconnect

12:42

or this idea that maybe these assessments

12:44

that people are using for structures

12:47

or for insurance, that they

12:49

might not actually match up with real

12:51

world shaking intensity? We

12:54

have a saying in seismology that

12:57

earthquakes don't kill people,

12:59

buildings kill people. And

13:02

a crucial input to building

13:04

design codes is seismic

13:06

hazard maps that try to forecast

13:08

how much shaking to expect with

13:11

a certain probability over many years,

13:13

given the lifetimes of buildings and

13:15

other structures so that engineers can

13:17

design them appropriately. Hazard

13:20

maps are also important for

13:22

insurance and reinsurance rates and

13:24

emergency management and mitigation, like you said.

13:26

Seismologists and

13:28

earthquake engineers have been making these maps

13:30

for a long time, but it turns

13:32

out they knew very little about

13:34

how well they actually forecast shaking

13:37

given the short record of

13:39

past earthquakes. Our team

13:41

was very interested in figuring out

13:43

how to evaluate the performance of these

13:45

maps to see if they were actually

13:47

accomplishing what they're supposed to do. Why

13:49

do you say that the records are

13:51

short? Yeah, so a

13:53

big obstacle to evaluating hazard

13:55

map performance is what we

13:57

call the short instrumental record of earthquakes.

14:00

earthquakes. Reporting of earthquakes

14:02

by seismometers, which can

14:04

really accurately measure ground

14:06

motions and those

14:09

seismometers were invented around 1900. But

14:14

the human record of earthquakes

14:16

and earthquakes shaking goes back

14:18

much further. Humans

14:21

have always been curious about earthquakes and

14:23

have been keeping records of damage

14:25

from earthquakes for a long time.

14:28

And so there's this measure of

14:30

shaking that seismologists use, which is

14:32

based directly on human perceptions of

14:35

shaking and the damage caused to

14:37

structures by shaking. So we

14:39

don't have a little needle that's like shaking on

14:41

a piece of paper that's recording this event. We

14:44

have a building fell down or

14:46

this bridge collapse. Exactly. Historians collected

14:48

that information from whatever kind of

14:50

public records have been taken

14:53

down throughout much longer history. Oh,

14:55

that's super interesting. Can you

14:57

give an example of a record that would

14:59

be in that data set? Do you have anything

15:01

that comes to mind? Me and

15:04

my team worked on creating an

15:06

intensity data set for the state

15:08

of California. It goes

15:11

back to 1857 when an

15:13

earthquake happened called the Fort Tahoe

15:15

earthquake. We looked through lots of

15:17

historical compilations of those shaking belt

15:19

reports that were collected by the

15:21

government at the time. The U.S.

15:24

Coast and Geodetic Survey, which eventually

15:26

became the U.S. Geological Survey. And

15:28

an example of some of the felt reports

15:31

would be hanging pictures

15:33

swung on walls. That's

15:35

a really low intensity. It might move

15:37

up to windows rattled and

15:40

then things fell off shelves, up

15:43

to heavy furniture being

15:45

moved, and then finally to

15:47

structural damage, cracks in the

15:49

walls, corners falling off

15:52

all the way to homes slipping

15:54

off their foundations. Eventually,

15:56

the top of the scale is

15:58

complete destruction. reminds me that when

16:01

there is an earthquake today, you can go

16:03

onto, I can't remember what the website is, and

16:05

just enter your shake report, right? These are

16:07

still collected even today. That's

16:09

right. Today we call it, Did

16:11

You Feel It? is the program that's

16:14

operated by the US Geological Survey. So

16:16

if you felt an earthquake, you can

16:18

go onto that website. And there's a

16:21

questionnaire which will ask you questions that

16:23

are related to that intensity scale. So

16:25

to those observations that I was just

16:28

listing before, if things swung on your

16:30

wall, if things fell over on your

16:32

tables, up to severe structural damage. How

16:35

can you translate a

16:37

building fell down, my hut fell

16:39

down, into a sensible record

16:42

that we can use today to kind of

16:44

estimate back in the past how strong an

16:46

earthquake was? We combine all

16:48

of the historical reports and photographs of

16:50

shaking and damage that we can find

16:52

in a region to map out

16:55

the distribution of ground motions. We

16:57

combine these shaking footprints to create

16:59

catalogs of maximum observed shaking

17:01

in a region over time.

17:04

How we

17:06

use that to compare directly

17:08

to the hazard models and

17:10

maps is through, it has

17:12

a funny name, but

17:15

basically it's called

17:17

ground motion intensity conversion

17:20

equations, GMICE. These

17:22

are conversion equations that

17:24

allow us to directly

17:26

compare those historical shaking intensity

17:29

data to the numerical

17:31

methods that we use in hazard

17:33

modeling. Yeah, I guess

17:35

I'm a little surprised that the modeling

17:38

for the seismic hazard maps, it doesn't

17:40

have any reference to this older stuff. It

17:42

is only based on seismometer readings. Is that kind of

17:44

what you're saying here? The seismic

17:47

hazard models are the

17:49

end result of many

17:51

other models. We've got

17:53

models of known fault

17:55

locations, models of unknown

17:58

fault locations, models of

18:00

of the frequency and magnitude of

18:02

earthquakes on those faults, and

18:04

then models of how ground shaking decays

18:06

away from the epicenter of an

18:08

earthquake. Those all combined

18:11

result in this hazard model,

18:13

which forecasts expected shaking with

18:15

a given probability over a

18:17

given time in one

18:19

of the instrumental measures from

18:21

a seismometer. All of that

18:24

to say that you can take the

18:26

hazard maps that you've constructed from these models

18:28

today, and then look at past

18:30

shaking incidents that were not measured with

18:33

seismometers and say, how

18:35

accurate your hazard map is? Is

18:37

that kind of what you did here? Exactly. How

18:39

did they line up with each other?

18:41

When you looked at the past, were

18:43

these seismic hazard models, were they predicting

18:45

what people would have felt in those

18:47

times? Wherever we looked

18:49

around the world, from

18:51

California to Italy to Nepal

18:54

to Japan to France, the

18:56

hazard map seemed to predict

18:58

much higher shaking than the

19:01

historic record shows. It's

19:03

important to note that even in

19:05

a perfect world, we would not

19:07

expect the forecasted maps to perfectly

19:09

predict shaking, as there is a

19:12

component of randomness to earthquake occurrence.

19:14

But around the world, seismic hazard

19:16

maps always over predicted the observed

19:18

shaking, which indicates that it

19:20

is a very general phenomenon. So there

19:22

has to be a very general explanation.

19:25

Japan and Nepal, they have these historic

19:27

data sets as well. That's right. And

19:30

they go back even longer than the

19:32

historical record in California. In regions like

19:34

that, we have records of the shaking

19:36

data that go back over 1,000 years

19:39

of human history. Wow. Since this

19:41

discrepancy is all in one

19:43

direction, do you have some ideas about

19:45

why there might be this

19:47

overestimation of intensity from the modern seismic

19:49

assessments? There are a lot of things

19:52

which affect the comparison in a subtle

19:54

way. But we found that the biggest

19:56

contributor to that result is

19:59

the convergence. of the

20:01

historical shaking intensity data to

20:03

the numerical methods that we

20:05

use in modeling. So it

20:07

was those conversion equations that allow

20:09

us to compare the different kinds of measures

20:12

of shaking. Does that mean that

20:14

those conversions are incorrect? Like

20:16

what can you say? Like can you say that they just, what

20:19

does it mean that those conversion equations

20:21

were incorrect? We would say

20:23

that the conversion equations give

20:26

a biased result. So once

20:28

we use those conversion equations,

20:30

the output is bias high.

20:33

From the standpoint of mitigating

20:35

earthquake risk, it's encouraging that

20:37

much of the apparent over

20:39

prediction of earthquake hazard results

20:41

from these conversion equations rather

20:44

than a systematic effect in

20:46

the earthquake hazard modeling approach. Wait,

20:49

can you say that again? I don't think I followed

20:51

that. We find it encouraging

20:53

that much of the apparent over

20:55

prediction of earthquake hazards with

20:58

respect to the observation results

21:00

from these conversion equations. So

21:03

rather than there being a

21:06

systematic effect or problem in

21:08

the way that we approach

21:10

earthquake hazard modeling, it's

21:13

just a small piece of

21:15

the evaluation that is biasing

21:17

the results in one direction

21:19

every time. So you're saying

21:21

that the modern seismic modeling

21:23

that they use for the

21:25

hazard maps, that doesn't have

21:28

this piece inside of it. So we're

21:30

not making mistakes today, but like when

21:32

we go to evaluate our models based

21:34

on past feeling reports, that is a

21:36

problem that hasn't been working correctly. That's

21:39

correct. So that is reassuring. And

21:41

it's also reassuring that it's in one direction. So

21:43

if we were depending on it, we

21:45

would be overbuilding, not underbuilding. Exactly.

21:49

What does this mean for the field? Does that suggest

21:51

that it needs to be a different way to make

21:53

the comparisons that you wanted to do here? Yes.

21:57

So improvements to these conversion equations

21:59

have been... proposed by my colleagues

22:01

and co-authors on this paper, Molly

22:03

Galihue and Norman Abrahamson. Using

22:06

these new unbiased

22:08

conversion equations will improve the

22:11

comparisons of hazard forecasts to

22:13

observed shaking. What would that do

22:15

in the long run? Does that just mean you're going to have more confidence

22:18

in predicting or understanding different

22:20

regions of the world, their

22:22

seismic activity? Yes, exactly. Thank

22:24

you so much, Leah. Thank

22:27

you, Sarah. Leah Saldich is

22:29

a Geoscience Peril Advisor at

22:32

Guy Carpenter. During

22:34

the time of the research for this paper,

22:36

she was a geoscientist in the USGS. You

22:39

can find a link to the Science

22:41

Advances paper we discussed at science.org/podcast. And

22:45

that concludes this edition of the Science

22:47

Podcast. If you have any comments or

22:49

suggestions, write to us at sciencepodcasts at

22:52

aaas.org. To

22:54

find us on a podcasting app, search

22:56

for Science Magazine. Or you can

22:59

listen on our website, science.org/podcast.

23:03

This show was edited by me, Sarah Crespy,

23:05

and Teva McLean. We also had

23:07

production help from Megan Tuck at Podigy.

23:10

Jeffrey Cook composed the music on

23:12

behalf of Science and its publisher, AAAS. Thanks

23:15

for joining us.