0:00

You're listening to the Stephen Wolfram podcast,

0:02

an exploration of thoughts and

0:04

ideas from the founder and CEO

0:06

of Wolfram Research, creator of

0:08

Wolfram of the Wolfram and the In

0:10

this ongoing Q &A series, Stephen

0:13

answers questions from his live stream

0:15

audience about the future of

0:17

science and technology. the future session was

0:19

originally broadcast on November was originally broadcast

0:21

on November 22nd, 2024. Let's have a

0:23

listen. Hello,

0:26

everyone. Welcome to another episode

0:28

of Q episode of Q&A and technology. of

0:30

science to say I've been kind of

0:32

busy recently say we are getting ready

0:34

to Launch a product ready

0:37

hope will be which

0:39

I step towards some

0:41

future some science and technology.

0:43

and technology. Let's see.

0:45

see, asks, since you talked about

0:47

the history of Since you talked about

0:49

the history of quantum mechanics, which I did

0:51

just a couple of days ago, how how about

0:53

the future of quantum mechanics? mechanics? Okay, let's

0:55

try and talk about the

0:57

future of quantum mechanics. mechanics. So a couple

1:00

of are a couple of different things. One is One

1:02

is. how quantum how quantum

1:04

mechanics really works, that's kind of

1:06

a foundational science question. question. The The

1:08

other is, we how can we kind of

1:10

use quantum mechanics in daily life? life? there

1:13

are already all kinds

1:16

of devices that make

1:18

quite central use of

1:20

quantum mechanics. quantum mechanics. Examples,

1:22

O LED, televisions,

1:25

and quantum dot

1:27

television displays displays

1:29

and so on. a certain kind

1:31

of make a certain kind of use of quantum

1:33

mechanics. There are a

1:35

variety of quantum effects that get

1:37

used in different places. Another one

1:39

is quantum places. Another phenomenon

1:42

quantum where you can have something. where

1:45

though have something where even

1:47

though kind of like you're rolling a

1:49

ball up a hill up a a barrier

1:51

and then down the other side the other

1:54

the ball doesn't really have

1:56

enough momentum to make it

1:58

over to make it over barrier. But in

2:00

quantum mechanics, there's a small probability

2:02

that the ball will kind of

2:05

tunnel through the barrier to make

2:07

it to the other side. And

2:09

that's a thing that's been used

2:11

for years in things called scanning

2:14

tunneling microscopes. It's been there a

2:16

variety of techniques for kind of

2:18

visualizing things like individual atoms and

2:20

molecules and so on that make

2:23

use of this kind of phenomenon

2:25

of quantum tunneling. That's a couple

2:27

of places where quantum mechanics has

2:29

been used. I think that the

2:32

There are various frontiers of the

2:34

use of quantum mechanics. One is

2:36

quantum metrology, another which I think

2:38

is less hopeful is quantum computing,

2:41

but let's let's talk about quantum

2:43

metrology. See, the fundamental thing, one

2:45

of the fundamental things about quantum

2:47

mechanics is Things in quantum mechanics

2:50

are very precisely, they're very precise

2:52

in the sense that that you

2:54

can be in this energy level

2:56

or that energy level, or you

2:59

can have your spin up or

3:01

down, but it can't be kind

3:03

of, you can't be in between

3:06

those energy levels. You can't have

3:08

you sort of in at least

3:10

in some way of setting it

3:12

up. You can't sort of have

3:15

your spin halfway between these things.

3:17

It's kind of like it's, it's,

3:19

it's, you know, on or off,

3:21

it's, it's definite. And that's very

3:24

helpful when it comes to making

3:26

very precise measurements of things, because

3:28

you can end up with a

3:30

situation where you say, okay, I'm

3:33

going to have one quantized flux

3:35

unit or two quantum flux units,

3:37

but you can never have one

3:39

and a half. And it's much

3:42

easier to have something which can

3:44

figure out, oh, it's going to

3:46

be exactly, it's exactly one, exactly

3:48

two, exactly three, and so on,

3:51

then something where you have to

3:53

kind of decide, well, is it

3:55

really 2.3, is it 2.4, whatever

3:57

else? So one of the places

4:00

where that's some magnetic fields very

4:02

precisely, acceleration very precisely. Those are

4:04

two places where one can imagine

4:06

using sort of these quantum effects

4:09

and this kind of all or

4:11

nothing feature of quantum mechanics to

4:13

make very precise measurements. What consequences

4:15

does that have? Well, it gives

4:18

one. all kinds

4:20

of possibilities for more precise

4:22

positioning than GPS, for example,

4:25

using kind of using features

4:27

of, well, either the Earth's

4:30

gravitational field or just working

4:32

out the acceleration that's been

4:35

imposed on something, or in

4:37

the case of magnetic field,

4:39

using the Earth's magnetic field

4:42

and so on, and knowing

4:44

very precisely things about that.

4:47

That's, those are a few.

4:49

So I mean, I think

4:52

the ability to have a

4:54

very precise kind of quantum

4:57

mechanics-based measurement of things potentially

4:59

allows things one can imagine

5:01

like a pocket MRI machine,

5:04

for instance, is something one

5:06

might imagine. One might imagine

5:09

a, well, as I say,

5:11

more precise positioning information, anything

5:14

that involves sort of precise

5:16

measurement of magnetic fields is

5:19

sort of, I think, on

5:21

the table. Let's see, what

5:23

else? Well, there are... a

5:26

lot of things that become

5:28

possible when you can make

5:31

very precision measurements. Another class

5:33

of things that one can

5:36

imagine are things where you're

5:38

kind of measuring features of

5:41

molecules in things. So one

5:43

of the big ones, which

5:45

I don't specifically know of

5:48

a kind of quantum metrology-based

5:50

proposal for this, but I

5:53

can imagine that one could

5:55

make one, is being able

5:58

to have non-invasive measurements. of

6:01

kind of levels of different

6:03

kinds of chemicals in one's

6:05

blood. So for example, right

6:08

now, if you have your

6:10

average fitness watch, it has

6:12

those flashing green things on

6:14

the back, those are a

6:17

PPG sensor, they are measuring

6:19

the essentially change of color

6:21

of blood. as it is

6:24

oxygenated and not. So measuring,

6:26

measuring the, your, your pulse

6:28

as your blood is pumped

6:30

in oxygenated form, where it's

6:33

darker in color, an unoxygenated

6:35

form, where it's lighter, and

6:37

by looking at the way

6:39

that that light is reflected

6:42

back, when, from, from blood

6:44

underneath your skin, that sensor

6:46

can determine what, when your

6:48

pulse occurred and can measure

6:51

things like your heart rate.

6:53

You can also potentially measure

6:55

the amount of oxygen, the

6:57

oxygenation of your blood, pulse

7:00

oxometry. Well, that's something that's

7:02

just measuring, okay, how much

7:04

oxygen is there in the

7:06

blood, is there, how many

7:09

red blood cells have got

7:11

the hemoglobin's full of attached

7:13

to oxygen and so on.

7:16

But the thing that's sort

7:18

of a big challenge is

7:20

how about all the hundreds

7:22

thousands of other kinds of

7:25

possibly more other kinds of

7:27

chemicals that are among blood

7:29

from glucose to all kinds

7:31

of all kinds of hormones,

7:34

all kinds of nutrients, all

7:36

kinds of, well ultimately fragments

7:38

of proteins, bacteria, virus, all

7:40

kinds of things. Everything ends

7:43

up sort of going through

7:45

the blood. Now the question

7:47

is, is there a way

7:49

of making such precise measurements

7:52

that you can detect, for

7:54

example, the little vibrational of

7:56

different molecules and so on.

7:58

One could imagine that with

8:01

a new generation of kind

8:03

of quantum metrology that things

8:05

like that might become possible

8:08

and that you can then

8:10

have sort of a real-time

8:12

display not just of your

8:14

heart rate but also of

8:17

hundreds of other attributes of

8:19

you know I just ate

8:21

something what did that really

8:23

do to me and so

8:26

on. Well another kind of

8:28

area of quantum kind of

8:30

applications is in things

8:33

to do with computers. And there's

8:35

sort of been a hope that

8:37

the sort of feature of quantum

8:39

mechanics in which kind of there

8:41

are many threads of history sort

8:43

of operating in parallel inside the

8:45

quantum process, but somehow one can

8:48

take advantage of that to be

8:50

able to run many threads of

8:52

computation in parallel. That's been kind

8:54

of the idea for 40 years

8:56

or more of quantum computers. And

8:58

it's not yet really panned out.

9:00

There are a large number of

9:02

efforts using about little four or

9:04

five different technologies to try and

9:07

sort of take advantage of that

9:09

feature of quantum mechanics. The main

9:11

challenge is, yes, you can have

9:13

all those quantum threads run in

9:15

parallel, but in the end, we

9:17

humans want to know what the

9:19

answer is, and we want to

9:21

know a definite answer. And knitting

9:24

together all those sort of threads

9:26

of history is something that is

9:28

not well accounted for in the

9:30

standard kind of formalism of quantum

9:32

mechanics, and is something which, as

9:34

a matter of actually building quantum

9:36

computers, is something that least implicitly

9:38

causes trouble. It hasn't been as

9:41

explicitly realized as I think it

9:43

should be that there are these

9:45

kind of potential limitations on sort

9:47

of the quantumness of what you

9:49

can do based on kind of

9:51

what's involved in measuring quantum the

9:53

little quantum process. But that's kind

9:55

of been a hope. there's sort

9:57

of a way to do computations,

10:00

which for an ordinary computer would

10:02

have to be done each piece

10:04

one after another, that you could

10:06

do them sort of in parallel

10:08

on a quantum computer. That hasn't

10:10

really panned out yet, as I

10:12

say. I mean, there are small

10:14

experiments along those lines, but not

10:17

more than that. There are also,

10:19

and I happen to think that

10:21

there are sort of fundamental theoretical

10:23

limitations there, which are probably some

10:25

of the most interesting science that

10:27

can be discovered from this. Well,

10:30

that whole direction has led to

10:33

all sorts of investigation of kind

10:35

of physics alternatives to the way

10:37

that we build computers now with

10:39

semiconductors and electronics and so on.

10:42

And there are many interesting possibilities

10:44

along those lines to do with

10:46

using light to do with using.

10:49

kind of individual atoms trapped

10:51

in electrical magnetic fields and

10:53

so on to represent the

10:55

bits and computers and there

10:58

are lots of interesting directions

11:00

there. mention a few other

11:02

quantum applications. Another one people

11:04

have talked about for a

11:06

while is various kinds of

11:08

quantum cryptography and quantum being

11:10

able to kind of transmit

11:12

information while not really, but

11:14

while having it be the

11:16

case that you kind of

11:18

can't intercept that information being

11:20

transmitted because if you did

11:22

the sort of quantum mechanics

11:24

would reveal the fact that

11:26

you'd tampered with it and

11:28

you'd be able to kind

11:30

of, you'd be able to

11:32

sort of know that it

11:34

happened. There have been various

11:36

experiments along these lines. I

11:38

don't know how critically important.

11:40

I don't think that that's,

11:42

it's not convincing to me,

11:44

that that's the weak link

11:46

in the security of lots

11:48

of kinds of systems. I

11:50

mean, I remember seeing, oh,

11:52

must have been 25 years

11:54

ago now, kind of the

11:56

idea of using sort of

11:58

having, did the photon go

12:00

down that piece of optical

12:02

fiber or not? piece of

12:04

fibre, whether it's just one

12:07

photon or there isn't a

12:09

photon, and trying to figure

12:11

out whether, if there's just

12:13

one photon, then as soon

12:15

as anything kind of interrupts

12:17

it, you kind of know

12:19

that something went wrong. But

12:21

I think when it comes

12:23

to building practical, secure computer

12:25

systems, I don't think that's

12:27

the weakest link typically. I

12:29

think with kind of the

12:31

rise of sort of algorithmically

12:33

based cryptography and the strong

12:35

belief that that can't be

12:37

broken for just purely algorithmic

12:39

reasons having nothing to do

12:41

the physics of anything that

12:43

that's a sort of a

12:45

goal alternative to those things.

12:47

Well, let's see, there are

12:49

a variety of other applications.

12:51

People have certainly talked about

12:53

for quantum mechanics. Again, quantum

12:55

sensing is, I think, probably

12:57

the most important, most obviously

12:59

important direction, making use of

13:01

kind of getting down to

13:03

the level of individual sort

13:05

of quantum bits in terms

13:07

of sensing things and just

13:09

lots of stuff that you

13:11

just wouldn't normally be able

13:13

to tell, but if you

13:16

can measure things sufficiently precisely,

13:18

you can determine what's going

13:20

on. And all sorts of

13:22

things with individual molecules, all

13:24

sorts of things with very

13:26

small effects where you're trying

13:28

to see whether did any

13:30

light really get through this

13:32

thing that absorbed most of

13:34

it? Did any radiation get

13:36

through this thing that absorbed

13:38

most of it? Those kinds

13:40

of things. That's on the

13:42

practical side. On the theoretical

13:44

side, I think the real

13:46

question is whether there's what

13:48

we can do in terms

13:50

of making a better theoretical

13:52

framework for thinking about quantum

13:54

mechanics. What sort of happened

13:56

historically is there was sort

13:58

of the early days of

14:00

quantum mechanics nearly a hundred

14:02

years ago now. invention of

14:04

things like children's equation, Heisenberg's,

14:06

matrix mechanics, things like this,

14:08

kind of classical quantum theory

14:10

in which one was using

14:12

differential equations, matrices, techniques like

14:14

that to try to sort

14:16

of understand how quantum, very

14:18

small phenomena, things, electrons, and

14:20

atoms, things like this work.

14:23

the what has happened in recent

14:25

times is the rise of kind

14:27

of quantum information and the idea

14:30

that one can use much more

14:32

sort of obviously computational concepts to

14:34

think about quantum mechanics and that

14:36

one can really start with the

14:38

idea of qubits and then sort

14:40

of derive other features of quantum

14:42

mechanics from starting to think about

14:45

sort of the informational aspects of

14:47

quantum mechanics. That's been a new

14:49

thing of the last 25 years

14:51

or so. And I think that's,

14:53

it gives a different kind of

14:55

way of approaching quantum mechanics, that

14:57

in some ways is it's still

15:00

just a formal way of approaching

15:02

it. It's just kind of doing

15:04

mathematics to figure out what happens.

15:06

It's not sort of saying why

15:08

does it happen this way, but

15:10

it's provided sort of a different

15:12

approach. and that's led to the

15:15

sort of entrainment of some fancy

15:17

mathematics, particularly from areas like category

15:19

theory, as a way to sort

15:21

of characterize this sort of the

15:23

abstract representation of what's happening in

15:25

a quantum system. Well,

15:27

I tend to think that what

15:30

we've been able to do with

15:32

our physics project is sort of

15:34

a big step forward in this

15:36

direction, because instead of just saying

15:38

what we build up quantum information

15:40

as this kind of formal mathematical

15:43

thing, we're really saying there is

15:45

an underlying structure to do with

15:47

these things we call multiway graphs,

15:49

branch shield graphs and so on.

15:51

There's an underlying structure that is

15:53

the way the universe is actually

15:56

built. and that gives you this

15:58

kind of object that that leads

16:00

to quantum mechanics. There's vastly more

16:02

to figure out. how that works,

16:04

but I think that's the, that's

16:06

kind of, it gives one a

16:09

kind of a paradigm for thinking

16:11

about these things that's that's extremely

16:13

computational from the beginning. Now there

16:15

are all sorts of effects and

16:17

things that that suggests. One of

16:19

them is we know in physical

16:22

space that the speed of light

16:24

governs the maximum rate at which

16:26

information can travel. Well, we think

16:28

that in branchial space, this kind

16:30

of space of possible quantum histories,

16:32

possible quantum branches, that there's an

16:35

analogous thing that we call the

16:37

maximum entanglement speed, and that that's

16:39

something that, again, limits the rate

16:41

of information propagation in branchial space.

16:43

And I suspect that that quantity

16:45

is related to limitations on quantum

16:48

measurement processes and therefore on how

16:50

one can use things like quantum

16:52

computers. So that's a thing which

16:54

is sort of a direction to

16:56

go in. What's happening in quantum

16:58

mechanics in general is that the

17:01

understanding of what happens when you

17:03

have just one electron, two electrons,

17:05

three electrons, a small number of

17:07

atoms. it's become quite good in

17:09

many situations, at least when those

17:11

things are not coupled together too

17:14

strongly, it's been possible to get

17:16

pretty good calculations in quantum chemistry

17:18

and things like this, but one

17:20

can really work out what happens

17:22

to those small numbers of particles.

17:24

by the time you're dealing with

17:27

a large number of particles, and

17:29

that's what's important when you're going

17:31

to amplify some small quantum effect

17:33

to a size that we can

17:35

actually directly sense. When you're dealing

17:37

with larger numbers of particles, it's

17:40

been extremely difficult to come up

17:42

with a formalism that can describe

17:44

what's going on, except in very

17:46

special cases. There are plenty of

17:48

special cases to do with over

17:50

the propagation of electrons and solids,

17:53

the all sorts of other particular

17:55

situations where, for example, in crystals,

17:57

there's a kind of regular lattice

17:59

of kind of background lattice. which

18:01

things like electrons move and that

18:03

regularity makes it sort of tractable

18:06

to study the quantum mechanics of

18:08

what's going on. But the general

18:10

problem of quantum many body problem

18:12

has been very difficult to crack

18:14

and I think that's sort of

18:16

a frontier problem that affects, well

18:19

lots of things again, there are

18:21

special cases where it's been lots

18:23

of progress has been made. An

18:25

example of a very quantum many-body

18:27

kind of thing is something like

18:29

super fluid liquid helium, which is

18:32

kind of a very that the

18:34

phenomenon of super fluidity is a

18:36

sort of fundamentally quantum phenomenon, and

18:38

certain aspects of that superfluids have

18:40

the feature that sort of everything

18:42

moves together. And that means that

18:45

one can somewhat reduce the kind

18:47

of, oh, you have to account

18:49

for all of these many different

18:51

particles to something a bit simpler

18:53

in that case. And so that's

18:55

been intractable. But the general problem

18:58

of sort of dealing with large

19:00

numbers of quantum particles that operate

19:02

according to quantum mechanics has been

19:04

really hard. And I don't know

19:06

for sure, but I have a

19:08

suspicion that this kind of approach

19:11

that we've taken will actually give

19:13

on a good way to approach

19:15

that. Now, of course, you have

19:17

to rely on the fact that

19:19

we can only do a certain

19:21

amount of computation. And the universe,

19:24

when it figures out what happens

19:26

in some quantum system, is doing

19:28

lots of computation. And the question

19:30

is, can we sort of reduce

19:32

that computation? Can we kind of

19:34

figure out particular aspects of what

19:37

the universe does, where we can

19:39

do a sort of cheap computation

19:41

and jump ahead to see what

19:43

happens? And exactly what kinds of

19:45

cheap computations are possible is not

19:47

yet clear. Let's see. Any other

19:50

questions about quantum mechanics? Let's

19:52

continue on here with a few,

19:54

some other questions about completely different

19:57

topics. Brady asks, will AI 2...

19:59

over all education. Well first one

20:01

has to have one AI tutor

20:03

that really works. We're working on

20:05

that problem. It's AI tutors are

20:07

frustrating because it's very easy to

20:09

make a five-minute demo. It's very

20:12

unclear whether you can make something

20:14

that will sort of keep a

20:16

student engaged through a whole class.

20:18

There are great things you can

20:20

do with AI tutors. For example,

20:22

you can immediately personalize kind of

20:25

whatever you're talking about. something in

20:27

algebra or something like that, you

20:29

can immediately personalize that to the

20:31

interests of a student. You can

20:33

expect the AI to essentially get

20:35

a model for what's going on

20:38

in the mind of the student,

20:40

so that, for example, it can

20:42

figure out, oh, this is what

20:44

you're confused about, the thing I

20:46

should tell you to unconfuse you

20:48

is this. It can also sort

20:50

of listen to what you're saying

20:53

about something and kind of get

20:55

an idea of where you're at

20:57

in terms of understanding it from

20:59

the vague things that you say

21:01

without you just doing sort of

21:03

the multiple choice quiz as an

21:06

assessment mechanism. So there are many

21:08

things where you can do sort

21:10

of continual much richer assessment. You

21:12

can provide sort of just the

21:14

information that you need. You can

21:16

kind of personalize things to make

21:18

them more engaging for a particular

21:21

kind of student. Can you really

21:23

motivate a person to learn something

21:25

when they're interacting with a computer?

21:27

Depends on whether the person is

21:29

internally motivated or not, I think.

21:31

Obviously, games, for example, manage to

21:34

kind of pull people along, or

21:36

at least some kinds of people

21:38

along. I mean, I, for example,

21:40

I'm perhaps, you know, it's an

21:42

admission of something. I never play

21:44

games, computer games or others. It's

21:46

just not something I find engaging.

21:49

So that wouldn't be successful in

21:51

pulling me along, so to speak.

21:53

But many people who find that

21:55

very engaging and for that kind

21:57

of thing does pull them along,

21:59

although the general gamification of education

22:02

has not been terribly successful. So

22:04

it has to be something I

22:06

think richer and more engaging, more

22:08

sort of human-like. to form of

22:10

engagement to kind of pull students

22:12

along with an AI tutor. But

22:14

it's unclear whether people kind of,

22:17

when all they have around them

22:19

are machines, how motivated do they

22:21

get? Some people are internally motivated

22:23

and get very motivated. You know,

22:25

I really want to learn this.

22:27

I'm really interested in it. You

22:30

know, machine just tell me what

22:32

I want to know. other people

22:34

are like, why am I doing

22:36

this? You know, nobody is encouraging

22:38

me. No, no human is encouraging

22:40

me to do this. And it

22:43

would them be, you know, it's

22:45

a question of how much sort

22:47

of human encouragement is really needed,

22:49

how much is education really kind

22:51

of almost a selling of knowledge

22:53

type type situation where it has

22:55

to be a human doing it

22:58

and how much of it is

23:00

just delivery a sort of convenient

23:02

ergonomic delivery of things. I don't

23:04

think we know that yet. And

23:06

there are many many years of

23:08

probably 70 years now of kind

23:11

of failures in attempts to kind

23:13

of machinify the education process. I

23:15

remember when I was a kid,

23:17

so this is probably, I don't

23:19

know, I had these books about,

23:21

you know, the future. from when

23:23

I was, I don't know, seven,

23:26

six, seven years old in the

23:28

mid-1960s. And I remember one in

23:30

particular had these illustrations of the

23:32

teaching machines of the future and

23:34

it didn't happen. And of course,

23:36

different things happened. I mean, this

23:39

is one of the things about

23:41

sort of the arc of technology

23:43

is you kind of imagine, oh,

23:45

there's going to be this automated

23:47

teaching machine, but actually what happens

23:49

is instead you get the web,

23:51

for example, and you get Wolfram

23:54

Alpha and things like that, and

23:56

you get something which is sort

23:58

of delivering something which is sort

24:00

of kind of like what you

24:02

might have imagined, except what you

24:04

actually imagined was based on what

24:07

already existed, and what actually gets

24:09

delivered is something quite different. But

24:11

so I don't know what I

24:13

think AI tutors will be really

24:15

important in a lot of kind

24:17

of leveling of the playing field

24:19

for education. You know, there are

24:22

right now, you know, you can,

24:24

if you're, if you're, if you're

24:26

lucky, you can get a really

24:28

great human tutor to help you

24:30

understand stuff, but there just aren't

24:32

enough human tutors to go around.

24:35

And this is a way to

24:37

kind of make that be a

24:39

much more level playing field, and

24:41

I think that's quite important. But

24:43

I don't think it's immediately solved

24:45

the problem of how do you

24:48

kind of motivate people to do

24:50

things. I think, by the way,

24:52

another thing that I think will

24:54

happen, should happen, is that the

24:56

very rigid, almost industrial form of

24:58

sort of assessing, you know, did

25:00

you learn this? Do the multiple

25:03

choice quiz? And, you know, how

25:05

many questions do you get right

25:07

and so on? I think that

25:09

mechanism, that very industrial kind of

25:11

mechanism for assessing what you've learned,

25:13

I think that becomes something that

25:16

is kind of a pre-AI story.

25:18

And in the post-AI story, it's

25:20

much more like the AI is

25:22

kind of interviewing you, it's kind

25:24

of doing an oral exam type

25:26

thing, and it's kind of asking

25:28

you questions and you're chatting with

25:31

it and so on. And I

25:33

think it's a much more flexible

25:35

and in some ways friendly, much

25:37

more personalized way of kind of

25:39

assessing where you're at. Did you

25:41

actually understand this or not? And

25:44

so I think that's a thing

25:46

where, as I mentioned, I think

25:48

continual assessment becomes something quite realistic,

25:50

where it's really you're kind of

25:52

assessing where you've got to by

25:54

the actual form of the interaction

25:56

that you're doing in the teacher.

25:59

rather than, oh, you learnt this,

26:01

now do a quiz, you know,

26:03

as that separate step. And I

26:05

have to say, I mean, I

26:07

always thought that the sort of

26:09

now take a quiz is for

26:12

almost all jobs that get done

26:14

in the world, that's not a

26:16

good model of them. It's not

26:18

a good way. It may be

26:20

the best we've got in sort

26:22

of industrial scale education to figure

26:24

out where did people get to,

26:27

but it's not a good model

26:29

of what people will encounter sort

26:31

of in the world at large.

26:33

Let's see. It says, notice a

26:35

question. Well, Cam is asking, will

26:37

traditional classrooms still exist 20 years

26:40

from now? Will everything be online?

26:43

You know, I mean, I've seen

26:45

lots of online education and my

26:48

kids did a bunch of it

26:50

and it's, it's, there are lots

26:52

of features of it. I think

26:55

it can be with, with self

26:57

motivated people, it's probably can be

27:00

quite successful. I think that still

27:02

humans are get a lot out

27:04

of being around humans and that's,

27:07

I mean I think, well online

27:09

education breaks into two major categories.

27:11

One is sort of the, it's

27:14

the asynchronous, you're with an AI

27:16

tutor, you're on your own, and

27:19

the other is synchronous, you're in

27:21

a virtual classroom with a bunch

27:23

of other students. And I think

27:26

those are two rather different directions.

27:28

And then there's the are you

27:31

sort of more more directly around

27:33

actual humans in real life type

27:35

thing. And then there's the question

27:38

of what's the role of teachers?

27:40

You know, is the role of

27:43

teachers to sort of recite knowledge?

27:45

Or is the role of teachers

27:47

to kind of help individual students

27:50

with over issues that they have?

27:52

Or is the role of teachers

27:54

more in the role of coaching

27:57

and encouragement and not so much

27:59

directly? And

28:02

I think those things can get

28:04

sort of factored out at this

28:06

point. The role of teachers as

28:08

a way to just sort of

28:10

recite knowledge, even as textbooks became

28:12

common, starting well, I think big

28:14

time sort of in the 1950s,

28:16

that role started to decrease. I

28:18

mean, before that time, if you

28:20

wanted to hear kind of the

28:22

lectures of Professor so-and-so, you couldn't

28:24

get the knowledge from them in

28:26

some disembodied way. And then some

28:28

bright publishing entrepreneur had the idea

28:31

of taking lecture notes, I think

28:33

it was from MIT, and publishing

28:35

them as books, and then they

28:37

could be sort of distributed more

28:39

widely. And of course when online

28:41

videos, online courses and so on,

28:43

came into existence again that the

28:45

delivery of knowledge in that way

28:47

became much more distributed and much

28:49

less necessary to have you know

28:51

all those separate teachers providing that

28:53

delivery except insofar as the delivery

28:55

by a person standing in front

28:58

of you so to speak is

29:00

more compelling than delivery that you

29:02

have to sort of read motivated

29:04

by yourself in a book or

29:06

something. I think that the sort

29:08

of the flipped classroom idea of

29:10

you kind of learn the facts

29:12

online on your own, and then

29:14

the classroom and the teacher is

29:16

about kind of helping you work

29:18

through actual problems that you're trying

29:20

to solve or whatever else. That's

29:22

been a popular thing. I gather

29:24

it works quite well. I guess

29:27

that the, you know, at a

29:29

time when when AI tutors get

29:31

to the point where they can

29:33

do quite a lot of that

29:35

helping you through things, that again

29:37

becomes something that doesn't need to

29:39

be in person, except insofar as

29:41

it's kind of the human motivational

29:43

aspect of it. And that kind

29:45

of leaves the, you know, well

29:47

encourage somebody to actually, yeah, you

29:49

can learn this, it's worth learning.

29:51

And, you know, maybe that comes,

29:53

maybe it's really important for many

29:56

people at least to see that.

29:58

a human rather than just the

30:00

AI saying, hey, yes, you can

30:02

do it. And the person is

30:04

like, what do you mean? This

30:06

AI is telling me that, what

30:08

is it? No, it's not, you

30:10

know, it doesn't, it's not compelling.

30:12

So that's a thought there. Meredith

30:14

asks, are there enough guardrails in

30:16

place for K-12 application of AI

30:18

tutors? A good question in the

30:20

AI tutor that we have, it's

30:23

pretty complicated. That's one of the

30:25

issues. You have to have a

30:27

primary AI that's interacting with a

30:29

student. You have to have other

30:31

AIs that are watching that AI.

30:33

And you're also generating real time.

30:35

It's another interesting thing is that

30:37

you're generating kind of a real

30:39

time digested report of what the

30:41

student did and what the AI

30:43

did with the students, so to

30:45

speak. And so if the, you

30:47

know, you see it's kind of

30:49

interesting to watch the behind the

30:52

scenes of our AI tutor, you

30:54

know, you see the thing it's

30:56

actually talking to the student about,

30:58

and then you see it's kind

31:00

of internal discussion. with other AIs,

31:02

so to speak, about what's going

31:04

on. Like, the student is losing

31:06

concentration, is losing focus, or, you

31:08

know, the student just asked me

31:10

something totally outrageous, or, you know,

31:12

the thing I was about to

31:14

tell the student is something I

31:16

probably shouldn't tell this particular student,

31:18

or whatever else. So, yes, it

31:21

is a complicated thing. And it...

31:23

I mean, I don't know. We

31:25

don't know for sure yet. We're

31:27

just starting to test our AI

31:29

tutoring system on actual students and

31:31

so on. My impression is that

31:33

that part is working fairly well

31:35

at this point. It probably depends

31:37

on the subject matter. I mean,

31:39

we are the first major target

31:41

for us is teaching algebra one,

31:43

which is something where there isn't

31:45

a lot of highly controversial content.

31:47

and that is on topic, so

31:50

to speak. So if it gets

31:52

sort of highly controversial, it's probably

31:54

very off topic, and that's a

31:56

separate issue. How do you get

31:58

the thing back on topic? you're

32:01

you're teaching other kinds of things, it

32:03

might be more complicated be because the

32:05

actual material might be much more

32:07

controversial and it might be the might

32:09

between what's sort of on topic and

32:11

what's controversial might much, much thinner. I'm

32:13

not sure yet. what's sort of

32:16

on topic and what's Let's

32:18

see. might be much much

32:20

thinner. I'm not sure yet. Wiesel

32:23

is commenting Weasel is Um

32:25

that... kids need

32:27

human encouragement to motivate them. motivate them.

32:29

think it varies a bunch. There's

32:31

a spectrum as there is with

32:33

many things, of different people have

32:35

different kinds of motivation. And

32:37

sometimes that motivation can come from the

32:40

very virtual kind of online. can come There

32:42

is a person at a great distance

32:44

and sometimes it actually has to be

32:46

a the person is right there type

32:48

thing and so on. actually has to

32:50

be the person is right see. type thing

32:52

and so on. Let's see. The

32:56

question here from... here

32:58

is, do I think that

33:00

medical ethics

33:03

will change Do I

33:05

think that medical ethics will change

33:07

with the rapid advance of gene therapies?

33:10

So the real so the

33:12

real question there is is, I I

33:14

mean, let's be realistic. therapy

33:16

is still in its early

33:19

stages. in its I mean, being

33:21

able to edit genes kind

33:23

of in situ, genes, being able

33:25

to go situ, being able to go

33:27

tissues in a in a running, you

33:29

you know, existing human so

33:31

to speak speak. being able to

33:33

modify. the genetics of the so

33:35

that you of something so

33:37

that you can avoid sickle

33:39

cell anemia or deal

33:42

with some retinal something like degeneration

33:44

or something like this. to

33:46

be are There are

33:48

things that seem to be quite successful in

33:50

very specific cases, where

33:52

there is a very

33:54

specific genetic modification caused

33:56

caused the disease. Most

33:59

diseases are much more... complicated

34:01

than that. And it's much less clear

34:03

how to kind of just go in

34:05

and zap the gene, so to speak,

34:08

zap the single base pair change, for

34:10

example, isn't what you're doing. you

34:13

know, things like CRISPR

34:15

and it's, there are

34:17

many sort of follow-ons

34:20

from CRISPR, CRISPR, what

34:22

it's essentially doing. I

34:25

mean, it's, it's, you

34:27

know, it's leveraging something

34:29

that biological evolution already

34:32

discovered was already existed

34:34

in bacteria, as

34:37

a, actually in bacteria, it existed

34:39

as a sort of very primitive

34:41

immune system for the bacteria, but

34:43

we kind of, you know, harnessed

34:45

that to turn it into this

34:48

gene editing mechanism. It's a little

34:50

embarrassing how many of the advances

34:52

in biology and biotechnology, particularly were

34:54

actually things that were invented by

34:56

nature, so to speak, and we

34:59

just found it and harnessed it

35:01

for our technological purposes, rather than

35:03

things where we just sort of

35:05

started from scratch and said, how

35:07

can we build this molecular structure

35:10

that does this or that thing?

35:12

But be that as it may,

35:14

there's sort of a bunch of

35:16

issues about how much can you

35:18

edit, how well can you target

35:21

the edits, all these kinds of

35:23

things, lots there that isn't really

35:25

known yet, but advancing fairly rapidly.

35:27

So there's the question of modifying

35:29

an existing human. And that's one

35:32

question. Another question is modifying the

35:34

modifying genetics in a, you know,

35:36

in an egg cell or something

35:38

like this to modify the future

35:40

human, so to speak, to make

35:43

sort of design their genes for

35:45

a future human. Let's deal with

35:47

the human, you know, the human

35:49

who's sort of alive and running

35:51

first. You know, I've always thought

35:53

that kind of the sign of

35:56

something gene editing wise having happened.

35:58

when we get kind of glow

36:00

in the dark humans and when

36:02

it becomes sort of a fashion

36:04

statement to have, you know, your

36:07

skin glow in the dark. You

36:09

can insert jellyfish genes into other

36:11

organisms. There are fish that glow

36:13

in the dark, which have been,

36:15

which have had the jellyfish gene

36:18

inserted into them. There is supposedly

36:20

a plant. that glows in the

36:22

dark that you can get and

36:24

I ordered one and hasn't come

36:26

yet. So I'm a bit suspicious

36:29

about what's happened to this plant.

36:31

But there is supposedly as of

36:33

earlier this year a plant that

36:35

glows in the dark because it

36:37

has jellyfish genes in it. And

36:40

you know I've always thought it

36:42

would be I don't know whether

36:44

it's terribly useful. Well there are

36:46

aspects of glowing in the dark

36:48

that would be useful certainly in

36:51

in things like optogenetics where one's

36:53

trying to insert genes

36:55

that will show if a nerve,

36:58

if a neuron is active,

37:00

it'll produce a light signal which

37:02

you can then detect and insofar

37:04

as light can get through

37:06

brain tissue and so on, you

37:09

can potentially sort of light up,

37:11

you can tell tell what

37:13

the critter or ultimately person is

37:16

thinking just by detecting those pulses

37:18

of light from the modified neurons

37:20

modified by editing their genes to

37:23

sort of glow in the dark

37:25

when they produce light when they

37:28

do things. I think that the

37:30

There are a bunch of diseases.

37:32

I mean, one can start sort

37:35

of ranking of the 100,000 or

37:37

so classified human diseases, so to

37:40

speak. It will be interesting to

37:42

rank them, and I'm sure people

37:44

have done this, by sort of

37:47

how exposed they are, how much

37:49

they are really accessible by making

37:52

genetic changes. I mean, there are

37:54

lots of pieces of genetics where,

37:56

you know, it kind of its...

37:59

oh, you should be lactose intolerant

38:01

based on your genetics, but you're

38:04

not, because you happen to use

38:06

some secondary pathway and the genetics

38:09

didn't play out exactly that way

38:11

in you. It's not, genetics is

38:13

not as... as

38:15

immediately determinative as one might have thought,

38:18

and also the vast majority of kind

38:20

of biological code is what programmers would

38:22

call spaghetti code. It's a big mess

38:25

with lots of different things contributing, and

38:27

it isn't the case that you can

38:29

just say, let me change that one

38:32

piece of code, and then I'll get

38:34

this thing to happen. something I've been

38:36

studying this actually a bunch recently because

38:39

I've been studying kind of a model

38:41

for the foundations of biological evolution which

38:43

seems to also allow one to have

38:46

sort of a model for the foundations

38:48

of medicine, a kind of a formal

38:50

model for what it means to make

38:53

changes to an organism and see what

38:55

happens so to speak. and I think

38:57

that may help inform the extent to

39:00

which you can really say let's zap

39:02

the genome a bit and get what

39:04

changes does that make? Most changes to

39:07

genomes have all kinds of complicated effects

39:09

and knowing that this particular one is

39:11

going to have this one, oh I'll

39:14

just change, I'll just make this one

39:16

change and then this one thing will

39:18

happen, there are diseases where that's the

39:21

case, but there are a lot where

39:23

it isn't the case. Now,

39:26

when you have kind of

39:28

the sort of, I think

39:30

there are fewer ethical challenges

39:32

when it comes to sort

39:34

of humans who already the

39:36

way they are and you

39:38

are trying to fix something

39:40

that isn't going right, so

39:42

to speak. more complicated as

39:44

the challenge of should you

39:46

make humans, where you have

39:48

kind of, you know, not

39:51

rolled the dice of nature,

39:53

so to speak, but instead

39:55

tried to engineer a human

39:57

with certain characteristics. There's obviously

39:59

a bad to this.

40:01

I mean, people after, well,

40:03

by the, people have been

40:05

doing plant breeding for millennia.

40:07

By the 1800s, people knew

40:09

about animal breeding and were

40:11

sort of routinely doing that,

40:13

you know, breed the better,

40:16

the better horse, the better

40:18

cow, the better whatever. And

40:20

so it was kind of

40:22

a natural thought, particularly after

40:24

Darwin and origin of species

40:26

and so on, 1859, that

40:28

one would think about kind

40:30

of instead of using natural

40:32

selection to determine kind of

40:34

how you got humans, use

40:36

artificial selection, breed the humans,

40:38

so to speak. Lots of

40:40

bad things happened as a

40:42

result. It turned out to

40:44

be an ethically pretty awful

40:47

idea. It's, now the question

40:49

is, if it's not kind

40:51

of something that you're doing

40:53

by sort of, if you're

40:55

breeding the humans by constraining,

40:57

who gets to mate with

40:59

who, who survives this, that,

41:01

and the other, I think

41:03

that the world has sort

41:05

of decided that that's kind

41:07

of ethically off limits. when

41:10

it comes to if you have,

41:12

for example, if you have a

41:14

choice of 20, you know, fertilized

41:16

egg cells that you're going to

41:18

implant, and you make measurements on

41:20

those cells, on those egg cells,

41:22

and you can see, oh, this

41:24

one has a high risk of

41:26

developing some nasty disease, that one

41:29

doesn't. What are the ethics of

41:31

picking the one that doesn't? Well,

41:33

I think one is a little

41:35

bit saved by the fact that

41:37

it's all very complicated. And when

41:39

you have that thing that reduces

41:41

the chance of cancer, that will

41:43

also, you know, make some other

41:45

thing, you know, make the person

41:47

be able to run a marathon

41:49

slower or something. There are lots

41:51

of complicated tradeoffs. It is a

41:54

spaghetti code kind of mess of

41:56

how genetic works. So I think,

41:58

at least for a while, there's

42:00

a sort of the phenomenon of

42:02

computational irreducibility that I talk about

42:04

a lot. Just because you know

42:06

the rules for a system doesn't

42:08

mean you know how it will

42:10

actually behave. That sort of saves

42:12

one a little bit on some

42:14

of the ethics. Because you make

42:16

a change you say I really

42:18

want my kid to have green

42:21

eyes Okay, great. That's a thing

42:23

where one pretty much knows what

42:25

the genetics of that have to

42:27

be Then you can you know

42:29

you can pick that one but

42:31

but just picking that if you've

42:33

got a lineup of a bunch

42:35

of a bunch of egg cells

42:37

or whatever the then you

42:40

know you're going to get a bundle

42:42

of six billion other base pairs in

42:44

there and having picked that particular aspect

42:46

is not really determinative of everything else.

42:48

Now if you say let's get a

42:51

little bit more extreme and let's say

42:53

we can make more careful edits and

42:55

so on and we have this giant

42:57

dashboard board and it's kind of you

42:59

go into the the the child factory

43:01

so to speak and you say I

43:04

want you know I'm gonna click all

43:06

these all these things like like you

43:08

know more even more detailed than you

43:10

know buying a car or something like

43:12

that. There are all these options you

43:15

know do you want the kid to

43:17

be able to do this that and

43:19

the other. First thing

43:21

to say again, I think it's

43:24

going to be a little bit

43:26

more complicated than one imagines because

43:28

I think many things come with

43:31

tradeoffs. You know, you want the

43:33

kid to be, oh I don't

43:36

know, you want the kid to

43:38

be really smart in this particular

43:40

kind of way if that's something

43:43

that can be determined by a

43:45

small amount of genetics or you

43:48

want, well, that will mean that

43:50

they're kind of a geek. And

43:52

so this will be bad for

43:55

the kid. I think it becomes

43:57

difficult. You know, it's not something

44:00

where it's kind of a, there's

44:02

an obvious right answer, so to

44:04

speak. There probably are some things

44:07

where certain nasty diseases. can exclude

44:09

those nasty diseases, that's something one's

44:12

already doing with lots of fetal

44:14

genetics kind of testing. That's already

44:16

kind of a thing. I mean,

44:19

you know, the earlier one can

44:21

do that, probably the ethically happier

44:24

one is about it. It's, you

44:26

know, it's a complicated thing because,

44:28

you know, in a sense ethics.

44:31

is like often we don't want

44:33

to do something that if done

44:36

to us would make us feel

44:38

bad. So you know if if

44:40

we know what would make us

44:43

feel pain or something then we

44:45

feel like we don't want to

44:48

impose that on other things that

44:50

are like us enough. that we

44:52

can kind of infer that they

44:55

would feel pain as we feel

44:57

pain, so to speak. And that's

45:00

obviously an issue with knowing how

45:02

that works for animals, fetuses, all

45:04

these kinds of things. It's very

45:07

unclear. It becomes very definitionally complicated.

45:09

I think that It, you know,

45:12

one of the things that that's

45:14

often an issue and, you know,

45:16

it's certainly, I certainly notice when

45:19

people are thinking about, you know,

45:21

animals and animal rights and so

45:24

on, you know, at the point

45:26

where we can sort of decode

45:28

animal language and we can have

45:31

a serious conversation with our cat.

45:33

we might have a different view

45:36

of kind of the how like

45:38

us the cat really is and

45:40

how much the ethics that we

45:43

would extend to other humans should

45:45

also extend to the cat could

45:48

go either way I think. But

45:50

in any case I would say

45:52

that the this whole question of

45:55

kind of should you design a

45:57

human to be a particular way

46:00

I think the bad news is

46:02

it's really hard to do it,

46:04

but that's also good news because

46:07

it means you're really stuck kind

46:09

of rolling the dice of computational

46:12

irreducibility and not being able to

46:14

say, I will design my Frankenstein

46:16

monster type thing with very precise

46:19

kind of robotic way. So I

46:21

think that's a few thoughts about

46:24

that at least. Ah,

46:27

mad. Monk says

46:29

that people have

46:31

created going on

46:33

rabbits, okay. Yes,

46:36

yes, yes. Indeed,

46:38

there are, you

46:40

can, you can

46:42

absolutely get, um,

46:44

okay, the rabbits

46:47

grow, glow green

46:49

and ultraviolet light,

46:51

apparently. There's

46:56

a question from Kathy. Could bacterial

46:58

viruses evolve to outsmart all our

47:00

medical advances? Biology has been pretty

47:03

clever about this. The immune system,

47:05

the human and mammalian adaptive immune

47:07

system, is pretty universal. It's these

47:09

molecules that have a certain shape,

47:12

and there are, you know, 10

47:14

billion or a trillion, depending on

47:16

which subsystem you're looking at, of

47:19

differently shaped molecules. And it turns

47:21

out that that inventory of possible

47:23

shapes of molecules of antibodies or

47:25

T-cells, whatever, is kind of enough.

47:28

that it can kind of recognize,

47:30

here's a protein with a certain

47:32

shape, and it's a protein that

47:35

isn't like a protein that's in

47:37

me, but is some kind of

47:39

alien foreign bad protein, so to

47:41

speak, that I should attack with

47:44

the immune system. There is a

47:46

certain universality to that. We don't

47:48

really completely understand how it works.

47:51

I'm actually curious to try to

47:53

figure it out. This kind of

47:55

space of possible shapes that proteins

47:58

take on and the space of

48:00

possible shapes that the immune system

48:02

can discriminate. But it seems that

48:04

sort of any shape that is

48:07

thrown at us by biology, as

48:09

a protein, so to speak, and

48:11

biological things are biology on as

48:14

life as we know it is

48:16

always made from proteins. that it

48:18

seems like our immune system is

48:20

sort of universally able to do

48:23

that recognition of protein shapes. That's

48:25

a piece of good news. Now

48:27

there are certainly things where one

48:30

can sort of hack biology and

48:32

try and find these kind of

48:34

weak links. It becomes a problem

48:37

very much like computer security. It's

48:39

very much like insofar as we

48:41

manage to decode a lot of

48:43

biology and we're still not quite

48:46

there yet. it becomes like computer

48:48

security where it's like, here's an

48:50

operating system. Even if you see

48:53

the source code for the operating

48:55

system, how do you find a

48:57

hack that allows you to to

48:59

worm your way into the operating

49:02

system and take over, for example?

49:04

Well, the only way that, I

49:06

mean, we, we, that problem is

49:09

sort of similar to the problem

49:11

of designing drugs and designing, you

49:13

know, drugs to treat diseases and

49:15

so on, because typically a drug

49:18

to treat a disease that the

49:20

most common thing it will do

49:22

is be of the right shape

49:25

to bind to some particular site

49:27

and some particular molecule and make

49:29

it more active, less active, whatever

49:32

else. And so we haven't been

49:34

terribly successful at sort of hacking

49:36

biology to the point of being

49:38

able to make designer drugs that,

49:41

you know, that will target some

49:43

particular kind of feature of some

49:45

particular kind of receptor on some

49:48

cell or whatever else. There's obviously

49:50

some success. There are a couple

49:52

of thousand drugs that have been,

49:54

you know, approved for use in

49:57

humans, but it's a lot of

49:59

work to find another one. And

50:01

there's a lot of problems and

50:04

a lot of candidates that don't

50:06

don't succeed. So the idea that

50:08

one can just sort of come

50:11

up with a a drug, a

50:13

molecule that will be kind of

50:15

a, oh yes, it will do

50:17

exactly this thing. That's proved really

50:20

pretty difficult. You know, people hope

50:22

with a series of different approaches

50:24

rational drug design, combinatorial chemistry, now

50:27

AI-based drug design, sort of designer

50:29

proteins based on sort of, you

50:31

know, giving a text prompt. that

50:33

instead of generating sort of a

50:36

big essay will generate a protein

50:38

structure, these kinds of things, these

50:40

are all efforts that people have

50:43

tried to sort of solve the

50:45

problem of make a protein that

50:47

does what you want it to

50:49

do. So far, there hasn't been

50:52

a kind of a slam dunk

50:54

solution to that. So when it

50:56

comes to can you, for example,

50:59

is there a way of creating,

51:01

you know, the virus, the bacteria

51:03

that's going to sort of hack

51:06

biology, that's one thing, then there's

51:08

the question of will nature throw

51:10

that up of its own accord?

51:12

in terms of creating it, well

51:15

that's, you know, the story of

51:17

bio weapons and so on, and

51:19

the, there's the sort of like

51:22

drug design, it's something where you

51:24

could do it ab initio, or

51:26

you could try and find a

51:28

thing that is some nasty toxin

51:31

that some particular kind of snail

51:33

uses, and then you can bottle

51:35

that up, so to speak, and

51:38

weaponize that. I think that the,

51:40

the thing... Well, let's see, in

51:42

biological evolution is a way of

51:45

sort of finding solutions to things,

51:47

finding that way of managing to

51:49

have the the bacterium defending. against

51:51

the bacteriophage viruses that are attacking

51:54

it, these kinds of things. Biological

51:56

evolution goes through all these different

51:58

iterations to try to find something

52:01

that works. I've recently happened to

52:03

have studied kind of the foundations

52:05

of how that works and why

52:07

that works. I think I now

52:10

sort of understand that. It's quite

52:12

an interesting story. But as a

52:14

practical matter, biological evolution is able

52:17

to do certain kinds of things,

52:19

not able to do other kinds

52:21

of things, in you get a

52:23

certain number of iterations that you

52:26

can try, like in the history

52:28

of life on earth, about 10

52:30

to the 40th organisms have lived.

52:33

So there have been sort of

52:35

10 to the 40th different tries

52:37

at different kinds of things which

52:40

succeed or fail, and that gives

52:42

you some kind of sense of

52:44

scale. I think about a trillion

52:46

generations of organisms have lived. So

52:49

that's sort of a sense of

52:51

scale of what you get to

52:53

try there. Now, one of the

52:56

things that certainly is a story

52:58

of in the sort of this,

53:00

there's what happens naturally in biological

53:02

evolution. there's what we can do

53:05

by explicitly designing things, and there's

53:07

also what we can do by

53:09

artificial selection. So for example, if

53:12

we kind of do, if we're

53:14

breeding viruses, for example, we can

53:16

do, you know, the equivalent of

53:19

a thousand years of what will

53:21

be natural selection, we can potentially

53:23

do in a matter of months

53:25

by just sort of applying, by

53:28

saying, you know, infect

53:30

something with these viruses, sort

53:32

out the cells where the

53:35

virus did this or that,

53:37

you know, modify the modified

53:39

version of the virus, you

53:41

know, this modification did better,

53:43

this natural modification did better,

53:46

pick out that one, run

53:48

that one again, and sort

53:50

of enhance to breeding effectively

53:52

of viruses. So, you know,

53:55

now of those three different

53:57

different directions, I think the

53:59

question of natural nature

54:01

nature. a thing that will

54:03

kind a thing that will

54:06

kind of defeat our immune

54:08

system. there have been many think,

54:10

you know, there've been many

54:12

threats to our immune system

54:14

some some some awkward ones like

54:16

Some awkward ones like AIDS,

54:18

for example, which specifically

54:21

attacks the immune system. system and

54:23

but but I guess my I think

54:25

that the sort of coexistence

54:28

of viruses with higher coexistence

54:30

of viruses with higher

54:33

organisms for the last,

54:35

you know, billion years does tend to

54:37

suggest that sort of left to

54:39

their own devices the you know

54:41

that it's you know, that unlikely

54:43

at least that sort of the

54:45

virus that can defeat the

54:47

immune system will arise. arise. Now,

54:50

you know, you know, an interesting question

54:52

is, could there theoretically be a virus

54:54

that defeats the immune system? Can

54:56

there theoretically be a way there to

54:58

sort of avoid this supposedly universal kind

55:00

of shape kind of system? I don't

55:02

think we know. immune to know

55:04

that requires better knowledge of how the

55:06

requires how the immune system actually works.

55:08

details of how the see. actually

55:10

works. Let's see. Just asks, will

55:12

AI-driven -driven biological evolution make

55:15

Darwinian evolution obsolete? How

55:17

do we? do we... prevent

55:19

the of biological viruses by

55:21

AI. of biological viruses

55:24

by AI. Ah yes. Well, it is a

55:26

problem. I mean, if you have the

55:28

practical situation. of being. of

55:32

You know, you know a company

55:35

that makes. makes, that synthesizes

55:37

DNA, for DNA, for example, you

55:39

check that do you check that you're

55:41

not some really nasty

55:43

virus? Really virus? You can Really

55:45

hard to do. you

55:47

can check that it doesn't match

55:49

check, you know, it's You can check,

55:52

if it matches rejected but

55:54

matches smallpox, but this

55:56

is again, a story of

55:58

computational If you're If you're presented this and

56:00

somebody says just make this genome

56:03

and you say well is that

56:05

a bad genome or a good

56:07

genome you really can't tell other

56:09

than particular sort of blacklisted cases

56:11

where it's a no and bad

56:14

genome you really can't tell. Now

56:16

will is it as I was

56:18

saying I mean I think that

56:20

the the story so it's a

56:23

question of whether kind of simulated

56:25

evolution will successfully outrun actual evolution

56:27

done with proteins and labs and

56:29

things like this. Not yet, we're

56:31

pretty far away from being able

56:34

to simulate the reality of biology.

56:36

We can at best simulate, I

56:38

think I've gotten somewhat, a good

56:40

leg up on on simulating the

56:42

essence of what's going on in

56:45

biology, but it's not the actuality

56:47

of what's going on in biology.

56:49

It's not something where you know

56:51

it it it it produces a

56:53

thing that actually would you know

56:56

would eat things and so on

56:58

and it's some it's it's it's

57:00

very much more idealized than that.

57:02

Elsie When, if at all, do

57:04

you anticipate we'll have mostly softwareized

57:07

humans, meaning that we can reprogram

57:09

ourselves just as easily as we

57:11

can reprogram computer systems? It's not

57:13

so easy to reprogram computer systems.

57:15

If you have a running computer

57:18

and you say, I want to

57:20

change this computer and what it's

57:22

doing, it's not so easy. I

57:24

mean, if you wanted to go

57:26

in and hatch a running operating

57:29

system, good luck. It's unlikely to

57:31

work. It's more like you to

57:33

crash, which in the human case

57:35

will probably mean die. So I

57:38

think that in, you know, there

57:40

is a very complicated issue. If

57:42

you start to be able to

57:44

modify what's happening in the brain

57:46

as an external thing and you

57:49

say, well, you know, I decided

57:51

I eat too much chocolate, I

57:53

want to just press a button.

57:55

have my chocolate eating, you know,

57:57

craving, so to speak, reduced. It

58:00

becomes a very complicated thing to

58:02

know kind of what's you, what's

58:04

free will, what's, how does this

58:06

all work? If you get to

58:08

just select with a slider, how

58:11

you feel about something. I mean,

58:13

it's what does that mean? You

58:15

felt this way about how you

58:17

should feel about it at that

58:19

time, but then the actual feelings

58:22

you had have changed, and how

58:24

does that loop really work? I

58:26

think this is a bridge we

58:28

haven't crossed, and I think it's

58:30

a complicated ethical story. And I

58:33

think then we have to ask,

58:35

what is ethics? is ethics the

58:37

what we think we have a

58:39

view of ethics based on the

58:41

way we are as humans today,

58:44

if we were different than we

58:46

are, we probably would have a

58:48

very different view of ethics. If

58:50

it was the case that, for

58:53

example, oh, I don't know, we

58:55

knew, let's say, for instance, we

58:57

knew precisely how long we would

58:59

live. and nothing could change it.

59:01

That would, I think, present a

59:04

different kind of a set of

59:06

issues in ethics that are different

59:08

from the ones we have. If

59:10

we knew, if certain things we

59:12

knew in advance, they would have

59:15

a different consequence for ethics. And

59:17

I think as sort of a

59:19

question of ethics from what we

59:21

are today, we project a certain

59:23

way of how we think things

59:26

should be now and in the

59:28

future, but if we were different

59:30

than we are, then we don't

59:32

have that same sort of anchor

59:34

for knowing what ethics should be

59:37

like. And ethics for the different

59:39

human will not be the same

59:41

as ethics for the human as

59:43

the human as the human is

59:45

today. I

59:50

have to unfortunately go soon but

59:52

let me maybe try and take

59:54

one more question here from from

59:57

might talking about biological evolution can

59:59

this humans break the longevity limit

1:00:01

for humans and say, get humans

1:00:03

who can live a couple of

1:00:05

centuries. I don't know what it's

1:00:08

going to take to do that.

1:00:10

I'm sure it's possible. It's, you

1:00:12

know, it is surprising the extent

1:00:14

to which we have kind of

1:00:16

a biological clock that changes the

1:00:19

way we age. We know we

1:00:21

have short-term biological clocks, we know

1:00:23

we have a circadian rhythm that

1:00:25

operates on a roughly 24-hour basis,

1:00:27

we know we have monthly rhythms,

1:00:30

we know those kinds of things,

1:00:32

and we know that, you know,

1:00:34

as we're kids and mature and

1:00:36

so on, we have things that

1:00:38

are like, you know, take a

1:00:41

decade or something. We know that

1:00:43

even something like, I don't know,

1:00:45

the plates of the skull, I

1:00:47

think, finally kind of join up

1:00:49

around age 40 or something. There

1:00:52

are all these clocks and timers

1:00:54

that exist in us as humans.

1:00:56

And it is a little bit

1:00:58

kind of shocking to realize that

1:01:00

those timers just keep operating. you

1:01:03

know when you have a face

1:01:05

recognizing program the facial features system

1:01:07

it's pretty accurate at saying how

1:01:09

old a person is it's picking

1:01:11

up a lot of details of

1:01:14

how bone structure tends to change

1:01:16

and so on these are things

1:01:18

that sort of inexorably seem to

1:01:20

happen and biology, it's unclear exactly

1:01:23

why we age. There are different

1:01:25

phenomena going on. There's genetic damage,

1:01:27

there's oxidative damage, and then there

1:01:29

are specific forms of genetic damage,

1:01:31

which aren't probably straightforward, as was

1:01:34

first imagined, of just these end

1:01:36

caps for chromosomes kind of falling

1:01:38

off. and progressively falling off as

1:01:40

more applications happen. It isn't quite

1:01:42

as simple as that it seems.

1:01:45

But, you know, kind of there

1:01:47

are these various clocks operating that

1:01:49

seem to the question then is,

1:01:51

well, do those clocks, can we

1:01:53

just stop those clocks operating? Most

1:01:56

likely, knowing how biology works, it's

1:01:58

always a very complicated tradeoff. You

1:02:00

know, you have those clocks operating

1:02:02

and they operate this way because

1:02:04

if you stop those clocks, you

1:02:07

would get lots of tumors, let's

1:02:09

say. or if you stop those

1:02:11

clocks, some other thing would happen.

1:02:13

Or even, now it could be

1:02:15

that some of the things that

1:02:18

those clocks are there for, so

1:02:20

to speak, have been successfully selected

1:02:22

for are things we don't care

1:02:24

about or don't want anymore. For

1:02:26

example, it could be that in

1:02:29

the past, sort of that you

1:02:31

wanted to get rid of the

1:02:33

old organisms, because otherwise nothing new

1:02:35

happens in the world. Everybody just

1:02:37

says, I learnt this when I

1:02:40

was a baby bird. and I'm

1:02:42

never going to try anything different,

1:02:44

you know, when I'm a 40-year-old

1:02:46

bird or something like this, and

1:02:48

that if that, if, if, if

1:02:51

sort of the, it could be

1:02:53

that sort of in, in, sort

1:02:55

of in the, in the course

1:02:57

of biological evolution, that having organisms

1:03:00

stick around forever is a way

1:03:02

to kind of prevent certain kinds

1:03:04

of flexibility in innovation. Maybe that's

1:03:06

true with humans as well. But

1:03:08

it's, you know, and I think

1:03:11

most of us humans, well, for

1:03:13

having a good time, we kind

1:03:15

of say, let's continue this as

1:03:17

long as possible. And, you know,

1:03:19

one would like to see effective

1:03:22

human immortality. Of course, it will,

1:03:24

it will have all sorts of

1:03:26

societal consequences that I think people,

1:03:28

it's an interesting question, sort of

1:03:30

how it affects one's view of

1:03:33

one's life if one thinks it's

1:03:35

infinitely long. And like, I'm not

1:03:37

going to bother to do this

1:03:39

now. I'll do this a century

1:03:41

from now. Or, you know, in

1:03:44

my long life I will have

1:03:46

done almost everything. You know, I

1:03:48

will have had every profession. I

1:03:50

will have had, etc. cetera.

1:03:52

etc. You know, it's a

1:03:55

It's a different

1:03:57

view of things

1:03:59

when you start

1:04:01

thinking on those thinking

1:04:03

on those timescales. unclear

1:04:06

whether aging is

1:04:08

a biological bug, a

1:04:11

a bug, a of

1:04:13

the physics of biology, the or

1:04:15

something that is a trade

1:04:17

-off that is a is a trade

1:04:19

-off that is irrelevant given sort of

1:04:21

of modern and you know

1:04:23

it's like there are plenty

1:04:25

of things that are trade -offs because

1:04:27

in for most of most of human

1:04:29

history, there wasn't things like clean

1:04:31

drinking water. There were parasites

1:04:34

everywhere. and And so there were

1:04:36

things that that had to do

1:04:38

to kind of deal with that,

1:04:40

which are with of irrelevant are sort

1:04:42

of in well, much

1:04:44

of the world at least. at least.

1:04:46

And And so as

1:04:48

as those things are is

1:04:50

now our environment, what

1:04:52

is now the, what

1:04:54

know, one can make

1:04:57

changes. one I changes. to

1:04:59

disappear here, to but here, but

1:05:01

a lot of lot of

1:05:03

interesting questions questions which I try

1:05:05

to address another time.

1:05:07

time. Thanks for joining

1:05:09

me me and bye for now. You've

1:05:11

for now. to the Stephen Wolfram podcast.

1:05:13

You can view the full Q&A series on

1:05:15

the Wolfram Research YouTube channel. For more information

1:05:18

on Stephen's publications, live coding streams, and this

1:05:20

podcast, visit Stephen Wolfram.com. you