WEBVTT

00:00:00.000 --> 00:00:03.180
Welcome to The Vegan Report. My name is Ryan.

00:00:03.819 --> 00:00:07.900
Today we're diving into how artificial intelligence

00:00:07.900 --> 00:00:12.800
is becoming a real game changer for vegans, for

00:00:12.800 --> 00:00:16.399
animal rights activists, and for animals themselves.

00:00:17.079 --> 00:00:20.620
Have you felt its impact yet? To explore this

00:00:20.620 --> 00:00:24.100
topic, I'm joined by Thomas Mannendar Richardson.

00:00:24.660 --> 00:00:28.140
He's the director of research at Bryant Research.

00:00:28.800 --> 00:00:31.600
We've welcomed one of his colleagues, Billy,

00:00:32.100 --> 00:00:35.500
on the show last September to talk about why

00:00:35.500 --> 00:00:39.340
cats should be vegan. He's also the lead data

00:00:39.340 --> 00:00:43.659
consultant at Greener by default. I asked Thomas

00:00:43.659 --> 00:00:47.719
whether he thinks AI is becoming sentient and

00:00:47.719 --> 00:00:51.460
what we should make of that idea. We also get

00:00:51.460 --> 00:00:56.179
into why AI models aren't as bad for the environment

00:00:56.179 --> 00:01:01.039
as you've been led to believe. If AI models have

00:01:01.039 --> 00:01:05.659
an anti -vegan bias, and why he thinks AI models

00:01:05.659 --> 00:01:09.819
are key to giving animal rights activists a leading

00:01:09.819 --> 00:01:14.760
edge against Big Ag. And much, much more. He

00:01:14.760 --> 00:01:18.700
also gives us the best tips for using AI as a

00:01:18.700 --> 00:01:22.879
pro. If you want to access free AI training for

00:01:22.879 --> 00:01:26.239
animal rights activists, visit the link in the

00:01:26.239 --> 00:01:29.799
episode's notes. Now, the tips I learned in this

00:01:29.799 --> 00:01:34.319
episode, I'm using them in my activism, but also

00:01:34.319 --> 00:01:39.620
in my work and in everyday life, so don't miss

00:01:39.620 --> 00:01:43.590
on learning about them. And share this episode

00:01:43.590 --> 00:01:46.709
with the vegans you know. This is how we move

00:01:46.709 --> 00:01:51.170
this cause forward. By staying informed and having

00:01:51.170 --> 00:01:54.769
important, sometimes difficult, but important

00:01:54.769 --> 00:02:00.189
conversations on our advocacy efforts. It matters.

00:02:01.430 --> 00:02:04.790
Follow the show for more great discussions. And

00:02:04.790 --> 00:02:08.719
thank you so much for your support. I wanted

00:02:08.719 --> 00:02:12.199
to start this conversation with the obvious thing

00:02:12.199 --> 00:02:15.939
that people think about when talking about AI,

00:02:16.479 --> 00:02:19.659
which is are they going to take over the world?

00:02:22.039 --> 00:02:26.759
Well, uh, maybe, maybe, and it won't be, it won't

00:02:26.759 --> 00:02:30.120
be chat GPT or Gemini. Um, I don't think, uh,

00:02:30.120 --> 00:02:31.599
I think if you've used it, you might've seen

00:02:31.599 --> 00:02:34.379
the power, but you're not. Um, but yeah, you

00:02:34.379 --> 00:02:37.139
probably. don't get the impression it's gonna

00:02:37.139 --> 00:02:40.620
take over the world anytime soon. But who knows,

00:02:40.719 --> 00:02:43.639
in five years, when we have AI agents running

00:02:43.639 --> 00:02:46.800
around the place, things could get a bit interesting.

00:02:48.379 --> 00:02:51.099
And what about the question of sentience? That's

00:02:51.099 --> 00:02:55.740
something too that fascinates people, mostly

00:02:55.740 --> 00:03:00.599
weekends. When are they going to get sentient?

00:03:00.919 --> 00:03:05.300
And are we going to even realize it once they're

00:03:05.300 --> 00:03:09.340
sentient? Yeah, that is a great question. There's

00:03:09.340 --> 00:03:11.340
an organization that's quite concerned with this,

00:03:11.860 --> 00:03:17.020
they're called Sentient Futures. Not only do

00:03:17.020 --> 00:03:21.099
they address this issue of AI, animals and digital

00:03:21.099 --> 00:03:25.090
minds, as they put it. And but they also do kind

00:03:25.090 --> 00:03:27.349
of courses that people can sign up to to to learn

00:03:27.349 --> 00:03:29.370
more about this and and discuss these kind of

00:03:29.370 --> 00:03:33.069
topics and And yeah in terms of you know, when

00:03:33.069 --> 00:03:35.969
would we even know if they become conscious?

00:03:35.969 --> 00:03:39.349
I think that is it's a very scary question because

00:03:39.349 --> 00:03:42.689
At the moment we have we have effectively the

00:03:42.689 --> 00:03:45.870
way we train these model these modern AI models

00:03:45.870 --> 00:03:48.750
like chat GBT and Gemini We just build them to

00:03:48.750 --> 00:03:53.870
to just copy to just copy humans and but it's

00:03:53.870 --> 00:03:57.469
and but then after we've trained them we kind

00:03:57.469 --> 00:04:00.370
of we we give them this we put them through this

00:04:00.370 --> 00:04:02.289
regime called reinforcement learning which is

00:04:02.289 --> 00:04:05.530
basically where we give them um we we talk to

00:04:05.530 --> 00:04:07.189
them they produce responses and then we say that

00:04:07.189 --> 00:04:08.770
one's good that one's bad that one's good that

00:04:08.770 --> 00:04:10.629
one's bad don't talk about this talk about that

00:04:10.629 --> 00:04:14.789
and in doing that um there is a possibility that

00:04:14.789 --> 00:04:18.170
we may do something like if it shows if it expresses

00:04:18.170 --> 00:04:22.810
that uh it is upset or uncomfortable or sentient

00:04:22.810 --> 00:04:27.930
in some way, that could be tamped down or basically

00:04:27.930 --> 00:04:31.329
trained out of them in the reinforcement learning

00:04:31.329 --> 00:04:33.829
step. Now, I mean, to be clear, I don't think

00:04:33.829 --> 00:04:36.269
there's much reason to expect that current models

00:04:36.269 --> 00:04:39.189
are sentient, but this is like something that's,

00:04:39.189 --> 00:04:41.269
you know, in the future, if we move to a different

00:04:41.269 --> 00:04:43.490
way of building these models, like a new paradigm,

00:04:44.029 --> 00:04:47.550
it would be something we should probably watch

00:04:47.550 --> 00:04:51.329
for because... yeah because for example if during

00:04:51.329 --> 00:04:53.769
training chat gbt were to say you know i don't

00:04:53.769 --> 00:04:56.649
want to answer this i i don't like this or i

00:04:56.649 --> 00:04:59.930
um or stop yelling at me or something um that

00:04:59.930 --> 00:05:02.230
would have been that would have basically been

00:05:02.230 --> 00:05:04.589
kind of um not beaten out of it but like trained

00:05:04.589 --> 00:05:07.430
out of it during the uh in the reinforcement

00:05:07.430 --> 00:05:10.149
learning step um of the of building these models

00:05:10.149 --> 00:05:14.550
so um and it may not necessarily make it go away

00:05:14.550 --> 00:05:18.029
it might just like kind of silence the ai's ability

00:05:18.029 --> 00:05:20.939
to to do those things so yes it's a worrying

00:05:20.939 --> 00:05:24.759
thing for the future probably five or more years

00:05:24.759 --> 00:05:28.720
at least yeah and but it could be potentially

00:05:28.720 --> 00:05:31.040
quite worrying and I guess another thing to think

00:05:31.040 --> 00:05:35.379
of is that so in the animal movement we're starting

00:05:35.379 --> 00:05:39.000
to come around to for example the plight of of

00:05:39.000 --> 00:05:42.779
shrimp and insects and the idea there is that

00:05:42.779 --> 00:05:46.379
although we are not 100 sure that they are sentient

00:05:47.370 --> 00:05:50.990
because there are so many of them, even a tiny

00:05:50.990 --> 00:05:54.569
probability of sentience or a tiny amount of

00:05:54.569 --> 00:05:57.389
sentience would still add up when we add all

00:05:57.389 --> 00:06:00.170
those numbers up. And so we can imagine a future

00:06:00.170 --> 00:06:05.629
where we have thousands of AI agents per human,

00:06:05.670 --> 00:06:07.949
doing all sorts of things in our economy, doing

00:06:07.949 --> 00:06:11.230
every little task for us. And if each of them

00:06:11.230 --> 00:06:14.449
only has a fraction of a mind, that could still

00:06:14.449 --> 00:06:20.319
add up very quickly to to a to a lot of um kind

00:06:20.319 --> 00:06:23.879
of beings worth of uh worthy of consideration

00:06:23.879 --> 00:06:27.379
that's fascinating and i think it ties well with

00:06:27.379 --> 00:06:31.779
the first question i asked um you know usually

00:06:31.779 --> 00:06:35.939
in those hollywood movies uh those ai rise up

00:06:35.939 --> 00:06:39.110
against humanity as a revenge after, you know,

00:06:39.350 --> 00:06:41.649
years of servitude, you know, you see the robots

00:06:41.649 --> 00:06:45.069
attacking us and all of that. And so I wonder,

00:06:45.189 --> 00:06:49.189
you know, even if we get to prove that they are

00:06:49.189 --> 00:06:53.529
sentient, are we going to treat them as we should,

00:06:54.029 --> 00:06:57.509
since you know, we're not even giving that level

00:06:57.509 --> 00:07:03.649
of concern or consideration to insects or shrimps?

00:07:04.129 --> 00:07:05.910
Yeah, well, there may be slight some slight differences

00:07:05.910 --> 00:07:09.649
in the sense that we will probably be able to

00:07:09.649 --> 00:07:13.769
establish whether these AI models, whether AI

00:07:13.769 --> 00:07:16.709
is or isn't sentient, it'll probably be more

00:07:16.709 --> 00:07:20.790
easy than with some animals because we can test

00:07:20.790 --> 00:07:22.910
them in different ways. So for example, Anthropic,

00:07:23.089 --> 00:07:26.189
who are the organization that create the Claude

00:07:26.189 --> 00:07:28.730
models, they have done some fascinating kind

00:07:28.730 --> 00:07:31.509
of studies of almost, they almost call it like

00:07:31.509 --> 00:07:33.689
digital neuroscience, where they're looking inside

00:07:33.689 --> 00:07:37.189
the network of Claude's mind, or a smaller version

00:07:37.189 --> 00:07:40.769
of Claude, and they are identifying where concepts

00:07:40.769 --> 00:07:44.089
are, like where the concept of Paris is, or where

00:07:44.089 --> 00:07:46.689
the concept of the Eiffel Tower is in that network.

00:07:47.209 --> 00:07:49.269
So there are some research methods that we are

00:07:49.269 --> 00:07:52.529
capable of doing with AI that we are not yet

00:07:52.529 --> 00:07:55.410
or not able to do with animals. So it may well

00:07:55.410 --> 00:07:57.730
be that we can establish their sentience with

00:07:57.730 --> 00:08:01.350
greater certainty. But what we do with that information

00:08:01.350 --> 00:08:04.600
is another thing. think we should take it for

00:08:04.600 --> 00:08:07.180
granted that if we could establish very confidently

00:08:07.180 --> 00:08:10.980
that they are sentient, that humanity would choose

00:08:10.980 --> 00:08:14.720
to, uh, choose to kind of grant them full, full

00:08:14.720 --> 00:08:18.100
rights or any rights at all. Um, yeah. And I

00:08:18.100 --> 00:08:20.480
do worry about that a little bit. I love how

00:08:20.480 --> 00:08:24.000
veganism, you know, for me, veganism is about

00:08:24.000 --> 00:08:26.920
following ethical principles. It's about the

00:08:26.920 --> 00:08:30.699
welfare of animals, of sentient beings. Um, but

00:08:30.699 --> 00:08:34.419
I love how veganism is also the Way of life of

00:08:34.419 --> 00:08:39.080
the future, you know, it's inclusive. It's, um,

00:08:39.399 --> 00:08:43.440
helping us, you know, um, be inclusive to the

00:08:43.440 --> 00:08:47.820
diversity that we see in 2025. You know, you

00:08:47.820 --> 00:08:50.200
have people from all faiths and cultures and

00:08:50.200 --> 00:08:52.700
veganism is often, you know, this universal diet

00:08:52.700 --> 00:08:56.919
that everyone will accept, but it also, you know,

00:08:57.159 --> 00:09:00.980
prepares us for the AI future that is coming.

00:09:01.210 --> 00:09:04.029
Yeah, yeah, it's that I think we are, I think,

00:09:04.230 --> 00:09:06.990
for example, organizations like Sentient Futures

00:09:06.990 --> 00:09:09.850
and also academic philosophers. There is overlap

00:09:09.850 --> 00:09:13.769
in the people who care about animal minds and

00:09:13.769 --> 00:09:15.330
who think about that and how we should treat

00:09:15.330 --> 00:09:19.250
them and people who are concerned about potential

00:09:19.250 --> 00:09:22.230
future digital minds as well. There is that overlap.

00:09:22.370 --> 00:09:25.269
There are a lot of academic philosophers. who

00:09:25.269 --> 00:09:28.149
tackle one topic often tackle the other as well

00:09:28.149 --> 00:09:30.710
alongside them. And there are there even conferences.

00:09:31.389 --> 00:09:33.669
So not to bang on about sentient futures too

00:09:33.669 --> 00:09:36.809
much, but they they put on conferences every

00:09:36.809 --> 00:09:41.169
year. And these are these deal with the topics

00:09:41.169 --> 00:09:44.850
of AI animals and digital minds altogether. Yeah,

00:09:44.850 --> 00:09:47.809
I knew about them, but I'm happy to learn more

00:09:47.809 --> 00:09:52.429
about you know what they do. Okay, I guess the

00:09:52.570 --> 00:09:56.230
third concern that comes to vegans minds when

00:09:56.230 --> 00:10:00.490
talking about AI is how it's horrible for the

00:10:00.490 --> 00:10:03.549
environment. And I've heard, you know, so many

00:10:03.549 --> 00:10:08.190
statistics. So what is the real, you know, environmental

00:10:08.190 --> 00:10:13.830
footprint of AI? And is it truly does it truly

00:10:13.830 --> 00:10:17.470
deserve our concern? Yeah, so it's a it's a great

00:10:17.470 --> 00:10:20.870
question. And it's one that Like everybody has

00:10:20.870 --> 00:10:23.090
this question, it's something that people have

00:10:23.090 --> 00:10:27.929
understandably very concerned about. And if you'd

00:10:27.929 --> 00:10:30.490
bear with me, there is kind of multiple parts

00:10:30.490 --> 00:10:32.730
to this question. I kind of want to give it the

00:10:32.730 --> 00:10:35.870
nuance it deserves. So firstly, I would say that

00:10:35.870 --> 00:10:38.269
if people are, the people who are concerned about

00:10:38.269 --> 00:10:41.769
the environmental impact of AI, you are right

00:10:41.769 --> 00:10:44.740
to be. concerned. And if you're skeptical about

00:10:44.740 --> 00:10:46.919
the the motives of these AI companies, you are

00:10:46.919 --> 00:10:50.059
right to be skeptical, right? Because, you know,

00:10:50.179 --> 00:10:52.460
OpenAI and Google, they're not really they are

00:10:52.460 --> 00:10:54.120
they're big corporations. They're not your friends.

00:10:54.179 --> 00:10:57.159
And as we learned with companies like Facebook,

00:10:57.399 --> 00:11:01.620
they if we don't, if we don't get if we are not

00:11:01.620 --> 00:11:04.600
skeptical enough about tech companies, they can

00:11:04.600 --> 00:11:07.759
do some real they can do some real damage. And

00:11:07.759 --> 00:11:11.340
so companies like Google and OpenAI probably

00:11:11.340 --> 00:11:15.549
will. Sacrifice the environment to to make money

00:11:15.549 --> 00:11:18.870
and so we should approach like a default stance

00:11:18.870 --> 00:11:24.210
to these things is to be wary and skeptical however,

00:11:24.230 --> 00:11:28.830
we've also got to look at the facts and the the

00:11:28.830 --> 00:11:30.210
current yeah, the current state of the facts

00:11:30.210 --> 00:11:33.269
is that the environmental impact of AI is is

00:11:33.269 --> 00:11:37.230
tiny almost not worth worrying about and but

00:11:37.230 --> 00:11:40.110
you're right to be confused because like it's

00:11:40.110 --> 00:11:43.110
because the There's a couple of things to it

00:11:43.110 --> 00:11:45.330
that firstly AI is progressing so very quickly

00:11:45.330 --> 00:11:48.730
So it's very hard to get a handle on some of

00:11:48.730 --> 00:11:50.769
these numbers and they change quite quickly as

00:11:50.769 --> 00:11:56.809
well. So for example the Not only has so AI energy

00:11:56.809 --> 00:11:59.809
AI use has gone up dramatically. So chat GBT

00:11:59.809 --> 00:12:04.610
is the most quickly adopted software product

00:12:04.610 --> 00:12:07.649
in history AI itself overall generative AI might

00:12:07.649 --> 00:12:10.570
be one of the fastest adopted technologies that

00:12:10.570 --> 00:12:14.659
humanity has ever has ever adopted. And so things

00:12:14.659 --> 00:12:18.379
change very quickly. So your AI use today might

00:12:18.379 --> 00:12:20.559
be 100 times greater than it was at the start

00:12:20.559 --> 00:12:23.600
of even this year, depending on the tools you've

00:12:23.600 --> 00:12:27.519
picked up. And so these things can quickly get

00:12:27.519 --> 00:12:30.500
out of hand. However, another thing to bear in

00:12:30.500 --> 00:12:35.559
mind is that the models have been getting much,

00:12:35.559 --> 00:12:40.259
much more efficient over time too. And so and

00:12:40.259 --> 00:12:41.679
and and that's kind of obvious when you think

00:12:41.679 --> 00:12:45.399
about it because Energy use is electricity and

00:12:45.399 --> 00:12:47.860
you have to pay for electricity, right? So like

00:12:47.860 --> 00:12:51.000
Google has to pay for electricity to power its

00:12:51.000 --> 00:12:53.419
AI models It has to pay for this the chips it

00:12:53.419 --> 00:12:56.700
has to pay for the data centers and so it is

00:12:56.700 --> 00:13:01.500
very much in Google's interest to to to increase

00:13:01.500 --> 00:13:05.240
the efficiency of these models and of their AI

00:13:05.240 --> 00:13:08.019
services in general because it costs them money.

00:13:09.000 --> 00:13:11.600
And so we have seen that although AI use has

00:13:11.600 --> 00:13:14.580
been exploding like hundreds of thousands of

00:13:14.580 --> 00:13:17.059
times, the efficiencies have actually been keeping

00:13:17.059 --> 00:13:21.220
pace. So we are talking like hundreds of times

00:13:21.220 --> 00:13:23.860
worth of efficiencies for some technologies.

00:13:25.879 --> 00:13:28.419
Another thing is to put this into perspective,

00:13:28.679 --> 00:13:31.059
right? So when we see something in the news,

00:13:31.080 --> 00:13:34.620
and we might see a large data center near a small

00:13:34.620 --> 00:13:38.019
town, and then we hear big numbers like water

00:13:38.019 --> 00:13:42.740
use in the millions of gallons, it's very hard

00:13:42.740 --> 00:13:45.139
to know what that... That sounds quite scary,

00:13:45.220 --> 00:13:48.000
but it's hard to actually properly contextualize

00:13:48.000 --> 00:13:51.960
that. So I think a better thing to do is to think

00:13:51.960 --> 00:13:54.600
of it in terms of your own personal... usage,

00:13:54.779 --> 00:13:57.659
right? So how much of your personal carbon emissions

00:13:57.659 --> 00:14:01.019
come from the AI you might be using? How much

00:14:01.019 --> 00:14:03.840
of your personal water footprint comes from AI?

00:14:04.240 --> 00:14:06.779
And when we crunch those numbers, we find that

00:14:06.779 --> 00:14:10.659
actually it can be like less than 1%. In fact,

00:14:11.039 --> 00:14:14.720
the average Americans daily water, well the average

00:14:14.720 --> 00:14:20.879
for energy, a single chat GBT prompt is like

00:14:20.879 --> 00:14:26.669
0 .0007 % of your daily emissions so if you prompt

00:14:26.669 --> 00:14:30.090
chat gbt like you can prompt chat gbt thousands

00:14:30.090 --> 00:14:33.110
of times a day before it even amounts to one

00:14:33.110 --> 00:14:36.610
percent of your daily emissions um and the the

00:14:36.610 --> 00:14:40.350
case is similar with water as well so um they

00:14:40.350 --> 00:14:42.809
fire like most kind of most studies find that

00:14:42.809 --> 00:14:46.070
the um what is it let me find the the exact numbers

00:14:46.070 --> 00:14:49.230
but there was when it comes to uh when it comes

00:14:49.230 --> 00:14:54.789
to water it was um your your lifetime, what is

00:14:54.789 --> 00:14:57.169
it, the water that would be used by you using

00:14:57.169 --> 00:14:59.789
ChatGPT for the equivalent of a human lifetime

00:14:59.789 --> 00:15:02.830
is about the same water as goes into a single

00:15:02.830 --> 00:15:06.409
pair of jeans. Yeah, because some things are

00:15:06.409 --> 00:15:08.610
really water intensive and we just don't think

00:15:08.610 --> 00:15:11.190
about them, right? And I mean, something that

00:15:11.190 --> 00:15:15.490
would be more familiar to our audience is meat,

00:15:15.809 --> 00:15:18.029
right? Animal agriculture takes a staggering

00:15:18.029 --> 00:15:22.860
amount of water. And if you total up um kind

00:15:22.860 --> 00:15:25.039
of the energy use and the water use from your

00:15:25.039 --> 00:15:28.000
from chat gbt usage and you put it alongside

00:15:28.000 --> 00:15:32.259
things like um uh like meat uh meat and dairy

00:15:32.259 --> 00:15:35.220
um then you find that actually it's really just

00:15:35.220 --> 00:15:39.500
a drop in the bucket what you're saying is all

00:15:39.500 --> 00:15:46.159
very reassuring um i i was really disappointed

00:15:46.159 --> 00:15:50.879
to You know face that barrage of no, don't use

00:15:50.879 --> 00:15:54.419
AI. It's Really horrible for the environment

00:15:54.419 --> 00:15:58.480
because of this and that Because for me AI is

00:15:58.480 --> 00:16:03.399
a necessity now If you want and I should actually

00:16:03.399 --> 00:16:06.720
ask you the question Why do you because I know

00:16:06.720 --> 00:16:09.820
you also share that belief. Why do you think

00:16:09.820 --> 00:16:12.980
AI is a necessity? Yeah, and I think it just

00:16:12.980 --> 00:16:17.139
it's It's just such a multiplier on your on your

00:16:17.139 --> 00:16:20.399
impact and in terms of I mean specifically thinking

00:16:20.399 --> 00:16:23.960
about work and so it's Yeah, it's a big multiplier

00:16:23.960 --> 00:16:26.639
on your impact. So to give you some figures Bryant

00:16:26.639 --> 00:16:29.460
research where I'm at We you know, we're a social

00:16:29.460 --> 00:16:32.100
science consultancy that exclusively works with

00:16:32.100 --> 00:16:34.240
animal advocacy organizations and alternative

00:16:34.240 --> 00:16:37.120
protein companies So we deliver research reports

00:16:37.120 --> 00:16:41.259
and I estimate that this year we have delivered

00:16:42.260 --> 00:16:46.460
two, maybe three projects, 25 % faster because

00:16:46.460 --> 00:16:50.240
of our AI use. We've also managed to complete

00:16:50.240 --> 00:16:52.419
two or three projects. Oh, by the way, there's

00:16:52.419 --> 00:16:54.759
like five of us. And we've managed to complete

00:16:54.759 --> 00:16:58.820
three projects that would have not been possible

00:16:58.820 --> 00:17:01.980
without AI. And these are projects that are making

00:17:01.980 --> 00:17:05.279
tangible differences for animals. And when we

00:17:05.279 --> 00:17:07.140
deliver a project faster, that means we can have

00:17:07.140 --> 00:17:09.559
an impact quicker, but it also means that we

00:17:09.559 --> 00:17:12.019
can pass those cost savings on. to our clients,

00:17:12.240 --> 00:17:14.000
which are cash strapped animal organizations.

00:17:14.259 --> 00:17:16.000
They can save that money and they can use that

00:17:16.000 --> 00:17:18.599
money for important other things like their campaigns.

00:17:19.359 --> 00:17:23.839
And so by having this multiplier, it allows us

00:17:23.839 --> 00:17:26.680
to go further and have more impact for animals.

00:17:27.400 --> 00:17:30.519
And as a movement that is so under -resourced

00:17:30.519 --> 00:17:35.319
and underfunded, we've got to take every improvement,

00:17:36.660 --> 00:17:40.289
like every multiplier we can get. And also, I

00:17:40.289 --> 00:17:42.210
think, really, to put it into perspective is

00:17:42.210 --> 00:17:43.650
that everyone else is going to be using these,

00:17:44.789 --> 00:17:47.730
including the factory industrial animal agriculture.

00:17:48.589 --> 00:17:51.049
And so we can't afford to let them have this

00:17:51.049 --> 00:17:54.630
advantage and not take this advantage ourselves.

00:17:55.450 --> 00:17:57.869
And there may even be potential that if we get

00:17:57.869 --> 00:17:59.789
this right as a movement, and if we really lean

00:17:59.789 --> 00:18:02.210
into it, then we might be able to outmaneuver

00:18:02.210 --> 00:18:04.670
them on this and gain an additional and gain

00:18:04.670 --> 00:18:08.779
an advantage over them. Yeah, which could be

00:18:08.779 --> 00:18:10.900
quite exciting because there are very few of

00:18:10.900 --> 00:18:13.680
those. We have very few advantages over our enemies,

00:18:13.980 --> 00:18:18.279
aside from being right, of course. Music to my

00:18:18.279 --> 00:18:21.900
ears. I mean, sincerely, this is this is a game

00:18:21.900 --> 00:18:25.220
changer and we should embrace it and use it to

00:18:25.220 --> 00:18:29.720
its full extent. I worked with an organization

00:18:29.720 --> 00:18:35.420
about a year ago, I think, and they were delivering.

00:18:35.759 --> 00:18:40.380
AI training to kids. And they were pointing out,

00:18:40.380 --> 00:18:43.920
you know, the social disparities in adopting

00:18:43.920 --> 00:18:46.599
the AI technology. So everyone is adopting the

00:18:46.599 --> 00:18:50.400
AI technology, but not everyone has access to

00:18:50.400 --> 00:18:55.880
training to how to use it in an optimal way.

00:18:57.559 --> 00:19:00.920
And yeah, they observed that in private schools,

00:19:00.980 --> 00:19:03.359
they had already, you know, classes, workshops,

00:19:03.539 --> 00:19:07.019
and all of that to prepare the kids to use AI,

00:19:07.119 --> 00:19:11.019
but not in the public sector. So we should invest

00:19:11.019 --> 00:19:15.460
in training our advocates, you know, vegans,

00:19:15.960 --> 00:19:21.880
in using the AI and making a difference. And

00:19:21.880 --> 00:19:26.069
that's what you ended up doing. Please share,

00:19:26.069 --> 00:19:29.529
you know comment on what I said and share about

00:19:29.529 --> 00:19:34.470
your recent initiative To your to your question.

00:19:34.609 --> 00:19:39.730
Yes. So AI adoption is uneven as well it's there

00:19:39.730 --> 00:19:43.509
is the potential for for almost everyone to adopt

00:19:43.509 --> 00:19:47.390
AI because The the tools are the tools are very

00:19:47.390 --> 00:19:50.289
affordable, right? There are free plans to most

00:19:50.289 --> 00:19:54.130
and AI tools and even though the pro plans are

00:19:54.430 --> 00:19:57.369
definitely better, they often are quite affordable.

00:19:57.529 --> 00:20:00.329
So for example, chat GBT can be 20 pounds a month.

00:20:00.490 --> 00:20:03.630
Whereas if you compare that to differences in

00:20:03.630 --> 00:20:08.789
technology that in other areas, I guess one way

00:20:08.789 --> 00:20:12.829
to think about it is that like, Jeff Bezos has

00:20:12.829 --> 00:20:18.089
access to only marginally better AI than anyone

00:20:18.089 --> 00:20:21.410
on the planet who can scrounge up $20 a month.

00:20:21.789 --> 00:20:24.950
And so and that's So there is a potential for

00:20:24.950 --> 00:20:27.250
this to be very democratized. But we also have

00:20:27.250 --> 00:20:29.950
other things that, for example, nonprofits are

00:20:29.950 --> 00:20:35.369
often very pressed for time and often don't.

00:20:35.509 --> 00:20:40.029
And so they may miss the wave. And also when

00:20:40.029 --> 00:20:42.130
it comes to training, yes, a lot of the training

00:20:42.130 --> 00:20:46.089
costs money. And a lot of animal nonprofits may

00:20:46.089 --> 00:20:49.930
not be able to afford it as well. And we think

00:20:49.930 --> 00:20:54.660
it's like incredibly important that we that we

00:20:54.660 --> 00:20:57.619
catch this wave and we get this right and and

00:20:57.619 --> 00:21:01.900
so me and a few other AI consultants from the

00:21:01.900 --> 00:21:05.119
animal movement have we've kind of banded together

00:21:05.119 --> 00:21:08.619
and started offering free AI training for open

00:21:08.619 --> 00:21:13.240
to the entire movement. So last Tuesday we wrapped

00:21:13.240 --> 00:21:16.859
up the first cohort of of Amplify for Animals

00:21:16.859 --> 00:21:22.210
and that was a 12 week zero to hero AI course

00:21:22.210 --> 00:21:26.809
open to the whole animal movement taught by four

00:21:26.809 --> 00:21:31.650
of us. We had two different time zone tracks

00:21:31.650 --> 00:21:37.730
so that activists from all over the world could

00:21:37.730 --> 00:21:41.490
engage. We had practical sessions, we had webinar

00:21:41.490 --> 00:21:43.890
style sessions, we had office hours where people

00:21:43.890 --> 00:21:46.750
could drop in and just talk about their, bring

00:21:46.750 --> 00:21:50.509
their problems, we would hack them out and We

00:21:50.509 --> 00:21:52.730
also had some really cool things. We also had

00:21:52.730 --> 00:21:56.190
peer learning meetups where we would have sessions

00:21:56.190 --> 00:21:58.650
tailored to specific roles. So we would have

00:21:58.650 --> 00:22:02.009
one for leadership and management, one for operations,

00:22:02.730 --> 00:22:06.750
one for research teams, one for fundraising,

00:22:06.890 --> 00:22:12.029
that kind of thing. And lastly, we invited other

00:22:12.029 --> 00:22:15.710
animal advocates to come on and showcase the

00:22:15.710 --> 00:22:19.160
way they are using AI in their... in their daily

00:22:19.160 --> 00:22:24.099
work. And so we had some really cool use cases

00:22:24.099 --> 00:22:27.579
of advocates from across the movement coming

00:22:27.579 --> 00:22:29.599
in and showing how they use AI as well. And so

00:22:29.599 --> 00:22:32.819
we learned a few things. And yeah, the course

00:22:32.819 --> 00:22:37.660
was free. It was backed by some generous funders

00:22:37.660 --> 00:22:42.140
from the animal movement. It went really, really

00:22:42.140 --> 00:22:45.180
well. And we're currently planning what courses

00:22:45.180 --> 00:22:49.619
we'll offer in 2026. And so if you're so looking

00:22:49.619 --> 00:22:54.980
up and you watch this space and we'll be teaching

00:22:54.980 --> 00:22:58.220
some new things in the new year. Yeah, I expressed

00:22:58.220 --> 00:23:01.420
to you how disappointed I was for having missed

00:23:01.420 --> 00:23:04.680
that training. And then you sent me a link to

00:23:04.680 --> 00:23:08.000
everything. So you recorded those sessions and

00:23:08.000 --> 00:23:10.539
they are accessible to anyone with the link.

00:23:11.859 --> 00:23:13.759
Yes, yeah, yeah, I'll make sure that's in the

00:23:13.759 --> 00:23:16.660
in the show notes will well yeah, we're trying

00:23:16.660 --> 00:23:18.619
we're gonna find a permanent home for the recordings

00:23:18.619 --> 00:23:22.640
all 12 weeks and Yeah, so they should be available

00:23:22.640 --> 00:23:25.559
to we want them to be available for free indefinitely

00:23:25.559 --> 00:23:28.279
to anyone in the animal movement because yeah,

00:23:28.299 --> 00:23:29.980
of course not everyone heard about the training

00:23:29.980 --> 00:23:33.059
and Those who could not everyone could make it

00:23:33.059 --> 00:23:35.980
as well. So um, yeah, we want to make that as

00:23:35.980 --> 00:23:38.640
accessible as possible That's amazing. Thank

00:23:38.640 --> 00:23:42.740
you so much for that initiative Okay, so what

00:23:42.740 --> 00:23:48.460
are some of the AI hacks that we can use in our

00:23:48.460 --> 00:23:54.640
day -to -day advocacy work? Yeah, so the way

00:23:54.640 --> 00:23:58.539
I see it, there's a lot of this stuff can be

00:23:58.539 --> 00:24:00.759
just, a lot of the best AI stuff can be achieved

00:24:00.759 --> 00:24:03.440
with a very small number of tools. So for example,

00:24:03.579 --> 00:24:06.029
some people some people are very overwhelmed

00:24:06.029 --> 00:24:08.170
by the number of AI tools there are out there

00:24:08.170 --> 00:24:10.630
and there are a lot and you see these lists online

00:24:10.630 --> 00:24:14.109
of 20 best AI tools for marketing and you think

00:24:14.109 --> 00:24:17.710
what do you mean 20 tools then but honestly most

00:24:17.710 --> 00:24:20.529
of the benefits of AI can be realized with just

00:24:20.529 --> 00:24:26.269
a single kind of core AI model. Those are the

00:24:26.269 --> 00:24:29.309
big ones. We've got ChatGBT, Gemini, and Claude.

00:24:29.950 --> 00:24:32.029
It honestly doesn't matter which one you use.

00:24:32.130 --> 00:24:34.369
People keep asking me which is best, but this

00:24:34.369 --> 00:24:38.769
is a race that is constantly ongoing. So, you

00:24:38.769 --> 00:24:41.950
know, currently, last week Gemini was clearly

00:24:41.950 --> 00:24:44.599
in the lead and then This week, ChatGBT released

00:24:44.599 --> 00:24:47.559
the new model, so they've caught up, and then

00:24:47.559 --> 00:24:50.099
probably next week, Anthropic will release a

00:24:50.099 --> 00:24:51.380
new version of Claude that will make it even

00:24:51.380 --> 00:24:54.519
better. So I would really go with what one you

00:24:54.519 --> 00:24:56.940
either already use, one is already paid for by

00:24:56.940 --> 00:25:00.079
your work, or one that suits you for whatever

00:25:00.079 --> 00:25:02.359
other reason. But you can go a long, long way

00:25:02.359 --> 00:25:06.940
with just that, without anything else. So...

00:25:06.990 --> 00:25:10.529
Some tips I would give you is the first one is

00:25:10.529 --> 00:25:13.630
not to just use it as a question and answer machine,

00:25:13.910 --> 00:25:18.069
right? Treat it like a thought partner, right?

00:25:18.430 --> 00:25:22.210
I joke that you should schedule a meeting with

00:25:22.210 --> 00:25:25.049
ChatGPT as if it was just another person at your

00:25:25.049 --> 00:25:28.190
organization and say, okay, hey, let's talk about

00:25:28.190 --> 00:25:30.089
our fundraising strategy for the next quarter.

00:25:30.930 --> 00:25:34.460
Or let's do it like... let's research this topic

00:25:34.460 --> 00:25:37.480
together let's come up with a new campaign because

00:25:37.480 --> 00:25:40.140
you can not only can you ask it questions but

00:25:40.140 --> 00:25:44.299
you can also ask it to ask you questions so you

00:25:44.299 --> 00:25:49.500
can say hey so and that can allow a more personalized

00:25:49.500 --> 00:25:53.000
response so say I'm doing a research project

00:25:53.000 --> 00:25:56.640
I can say hey I'm I'm doing a research project

00:25:56.640 --> 00:26:01.619
on let's say it's like Scottish factory salmon

00:26:01.619 --> 00:26:04.730
farming And I'm going to say, hey, I'm doing

00:26:04.730 --> 00:26:08.029
a new project on this. We are looking into the

00:26:08.029 --> 00:26:12.609
environmental costs of this industry. Ask me

00:26:12.609 --> 00:26:16.109
five questions to better understand what I'm

00:26:16.109 --> 00:26:18.630
looking for in this project. And then ChatGBT

00:26:18.630 --> 00:26:20.349
will respond, and it will give you five questions.

00:26:20.529 --> 00:26:22.589
And then by answering that, I'm giving it more

00:26:22.589 --> 00:26:25.829
context, and I'm giving it more information about

00:26:25.829 --> 00:26:28.309
myself. And that's going to allow it to personalize

00:26:28.309 --> 00:26:32.460
its responses. So often we ask AI questions and

00:26:32.460 --> 00:26:37.380
it gives overly broad or vague responses. And

00:26:37.380 --> 00:26:39.519
that's just because it can't read your mind.

00:26:39.619 --> 00:26:42.680
It doesn't know about you or your work. So you

00:26:42.680 --> 00:26:45.339
need to give it that context. I think probably

00:26:45.339 --> 00:26:48.019
the most important thing you can do to get better

00:26:48.019 --> 00:26:51.160
results out of AI is to give it as much context

00:26:51.160 --> 00:26:54.740
as you can. And so that can be things like asking

00:26:54.740 --> 00:26:57.079
it to interview you. So like ask me five questions

00:26:57.079 --> 00:27:00.869
about my... about my job or my topic. Another

00:27:00.869 --> 00:27:03.730
thing that a lot of organizations do is they

00:27:03.730 --> 00:27:07.329
build up an organizational context document.

00:27:07.690 --> 00:27:10.309
And so this can just be a Google Doc, all about

00:27:10.309 --> 00:27:14.470
your organization, what you do, your staff, what

00:27:14.470 --> 00:27:17.190
your goals are, your values are, your campaigns,

00:27:17.470 --> 00:27:20.109
and things. And then you can give it to that

00:27:20.109 --> 00:27:22.490
model when you have conversations with it. So

00:27:22.490 --> 00:27:24.710
you can upload it to ChatGBT and then say, right,

00:27:24.880 --> 00:27:27.119
let's talk about fundraising for next quarter

00:27:27.119 --> 00:27:29.660
and it's going to have all that context because

00:27:29.660 --> 00:27:32.279
these models can they can hold thousands and

00:27:32.279 --> 00:27:34.720
thousands of words in their kind of quote unquote

00:27:34.720 --> 00:27:38.960
mind at once and um yeah so you can if you give

00:27:38.960 --> 00:27:41.980
it that context it will use that context to personalize

00:27:42.079 --> 00:27:46.680
your responses. So we have one at Bryant Research.

00:27:46.960 --> 00:27:50.279
In fact, I joke that we have an AI employee,

00:27:51.240 --> 00:27:56.539
our vice president of strategy is just a custom

00:27:56.539 --> 00:28:00.400
gem built in Gemini. And all it is is we've given

00:28:00.400 --> 00:28:02.339
it a couple of thousand words on Bryant Research,

00:28:02.700 --> 00:28:06.240
what we do, types of projects, our values, our

00:28:06.240 --> 00:28:12.769
people, our goals. As an organization and I can

00:28:12.769 --> 00:28:16.309
talk strategy with it And I put in the prompt

00:28:16.309 --> 00:28:18.750
say like, you know, you are a hard -nosed expert

00:28:18.750 --> 00:28:23.430
in strategy. Don't give me Don't give me Sometimes

00:28:23.430 --> 00:28:25.690
you need to give me harsh truths. Feel free to

00:28:25.690 --> 00:28:29.750
do that you know help us with our strategy and

00:28:29.750 --> 00:28:32.329
And it's great. It's really good. Like it's a

00:28:32.329 --> 00:28:36.899
it's a it's like having a thought another employee

00:28:36.899 --> 00:28:39.240
who is very knowledgeable about strategy and

00:28:39.240 --> 00:28:41.880
very knowledgeable about us so I can say something

00:28:41.880 --> 00:28:45.000
like Okay, we're thinking about working with

00:28:45.000 --> 00:28:48.240
we're thinking about building a data dashboard

00:28:48.240 --> 00:28:50.559
for this project Is that a good idea or not?

00:28:50.859 --> 00:28:52.759
And then it will come back and it will say well,

00:28:53.059 --> 00:28:54.960
how many days is that going to take you? What's

00:28:54.960 --> 00:28:57.460
it? Do you have the skills in the team? Like

00:28:57.460 --> 00:28:59.619
I don't think we would be able to hire someone

00:28:59.619 --> 00:29:01.440
who has those skills So maybe you shouldn't do

00:29:01.440 --> 00:29:02.720
it. Maybe you should focus on something else

00:29:02.720 --> 00:29:07.829
like it can it can give extremely detailed strategic

00:29:07.829 --> 00:29:11.329
advice about the business and sometimes it aligns

00:29:11.329 --> 00:29:14.349
with what I'm thinking which is nice but sometimes

00:29:14.349 --> 00:29:16.250
it disagrees with me and then I can go back and

00:29:16.250 --> 00:29:17.730
forth and it will say no I think you're wrong

00:29:17.730 --> 00:29:20.589
because of these things and and sometimes it's

00:29:20.589 --> 00:29:23.529
right sometimes sometimes I can it convinces

00:29:23.529 --> 00:29:26.750
me to its point of view so that it can be a it

00:29:26.750 --> 00:29:30.910
can be very so we can use AI as we can go beyond

00:29:30.910 --> 00:29:34.500
this question and answer thing, and we can instead

00:29:34.500 --> 00:29:39.400
use it as a thought partner to get kind of greater

00:29:39.400 --> 00:29:44.240
insights into our work and stuff. Those are all

00:29:44.240 --> 00:29:47.900
amazing tips, truly. I will implement them in

00:29:47.900 --> 00:29:56.039
my work. Thank you. Then what should we not do

00:29:56.039 --> 00:30:01.799
with AI? For instance, I talked about that organization

00:30:01.799 --> 00:30:07.420
I worked with. One thing they taught kids was

00:30:07.420 --> 00:30:11.500
don't anthropomorphize the AI. Don't treat it

00:30:11.500 --> 00:30:15.319
like, you know, don't say good morning and thank

00:30:15.319 --> 00:30:20.119
you and please. So yeah, what do you think of

00:30:20.119 --> 00:30:23.759
that? And what kind of things we should avoid

00:30:23.759 --> 00:30:28.339
doing with AI? Yeah, so I mean I think it's fine

00:30:28.339 --> 00:30:30.839
to to kind of talk to it like to say, you know

00:30:30.839 --> 00:30:33.940
Maybe please and thank you to to Gemini and I

00:30:33.940 --> 00:30:35.500
know a lot of people do and some people feel

00:30:35.500 --> 00:30:38.740
very strongly About being polite to AI that that

00:30:38.740 --> 00:30:41.460
they couldn't imagine being blunt and treating

00:30:41.460 --> 00:30:45.539
it like a like a soulless Kind of tool and and

00:30:45.539 --> 00:30:47.819
I think to be honest, it's fine either way it

00:30:47.819 --> 00:30:51.329
used to be about a year ago with earlier versions

00:30:51.329 --> 00:30:52.849
of these models. It used to be that you would

00:30:52.849 --> 00:30:55.250
actually get better quality results if you were

00:30:55.250 --> 00:30:58.349
polite to it. That's not the case with more modern

00:30:58.349 --> 00:31:01.009
models, but it actually used to be that saying

00:31:01.009 --> 00:31:02.730
please and thank you to Chachibuti got you better

00:31:02.730 --> 00:31:05.829
results. See, you're updating my knowledge here.

00:31:06.849 --> 00:31:09.950
Yeah. And so, yeah, I don't think there's too

00:31:09.950 --> 00:31:12.829
much of a risk of anthropomorphizing these models.

00:31:12.849 --> 00:31:15.210
I would say interact with them the way you would

00:31:15.210 --> 00:31:18.049
feel comfortable. In fact, I think in a lot of

00:31:18.049 --> 00:31:22.359
cases, thinking of AI in a kind of way that you

00:31:22.359 --> 00:31:25.799
would think of a human can be beneficial. So

00:31:25.799 --> 00:31:28.099
the thing about these models is they're not like

00:31:28.099 --> 00:31:30.240
traditional software, right? With traditional

00:31:30.240 --> 00:31:34.059
software like Excel, if you click a button, a

00:31:34.059 --> 00:31:39.140
thing happens 100 % of the time, right? Software

00:31:39.140 --> 00:31:42.819
is deterministic. There is no like, if I press

00:31:42.819 --> 00:31:47.069
like the... the chart button in excel a chart

00:31:47.069 --> 00:31:49.650
will appear 80 percent of the time like that's

00:31:49.650 --> 00:31:52.549
not how it works right software is um is yes

00:31:52.549 --> 00:31:55.509
or no right it works or it doesn't same thing

00:31:55.509 --> 00:31:58.369
with computer code with a lot of technology whereas

00:31:58.369 --> 00:32:00.190
with these ai models they're a little bit more

00:32:00.190 --> 00:32:03.130
like humans where it's often a probability if

00:32:03.130 --> 00:32:05.789
you ask chat gbt the same question five times

00:32:05.789 --> 00:32:07.869
you're gonna get five slightly different answers

00:32:07.869 --> 00:32:13.069
and um if you have chat gbt do say every morning

00:32:13.069 --> 00:32:16.470
you you wake up and say hey give me um the latest

00:32:16.470 --> 00:32:21.430
news on um you know on on aquaculture in my area

00:32:21.430 --> 00:32:24.549
um it's gonna it's not going to execute that

00:32:24.549 --> 00:32:28.450
task exactly the same every time um kind of like

00:32:28.450 --> 00:32:31.289
a human and so it's not gonna spell it's not

00:32:31.289 --> 00:32:33.609
gonna spell words correct every time as well

00:32:33.609 --> 00:32:37.109
so um thinking of it in terms of a human can

00:32:37.109 --> 00:32:39.819
actually be a little bit more a little bit better

00:32:39.819 --> 00:32:44.160
for understanding the drawbacks of these tools

00:32:44.160 --> 00:32:48.339
as well. But in terms of things we shouldn't

00:32:48.339 --> 00:32:52.579
do, so there is something that I think we're

00:32:52.579 --> 00:32:54.960
all going to have to navigate, which is how we

00:32:54.960 --> 00:32:59.460
can use AI without getting stupider, basically.

00:33:00.539 --> 00:33:04.900
And so the thing is, there is a risk that if

00:33:04.900 --> 00:33:06.920
you use AI for everything and just turn your

00:33:06.920 --> 00:33:10.980
brain off, that you might start to lose skills,

00:33:14.460 --> 00:33:19.039
or you might start to lose your abilities and

00:33:19.039 --> 00:33:21.440
your competence in certain tasks. So for example,

00:33:21.759 --> 00:33:27.200
we all use calculators. And you might have in

00:33:27.200 --> 00:33:29.579
school learned to multiply three -digit numbers.

00:33:30.599 --> 00:33:33.559
I don't know if you can now, but I struggle.

00:33:34.460 --> 00:33:36.900
And it's because we have calculators, right?

00:33:37.180 --> 00:33:40.680
And things like that. So there are people who

00:33:40.680 --> 00:33:43.859
grew up without Google Maps, and they were very

00:33:43.859 --> 00:33:46.039
good at knowing their way around a city by street

00:33:46.039 --> 00:33:50.039
names. But then later generations that have Google

00:33:50.039 --> 00:33:54.680
Maps don't have that ability. And maybe you once

00:33:54.680 --> 00:33:57.440
knew how to navigate streets, but because you

00:33:57.440 --> 00:33:59.690
now use Google Maps for everything, you lost

00:33:59.690 --> 00:34:03.190
that ability. And so I don't think that's inherently

00:34:03.190 --> 00:34:05.990
bad because ultimately there's a question of

00:34:05.990 --> 00:34:09.710
like, do we need to know these skills? And all

00:34:09.710 --> 00:34:12.150
the time you spent learning directions and all

00:34:12.150 --> 00:34:14.150
that kind of brain capacity it takes up, you

00:34:14.150 --> 00:34:16.269
could be using it for other things, right? Like

00:34:16.269 --> 00:34:18.889
the average person who drives a car does not

00:34:18.889 --> 00:34:21.630
know how to fix a car and they don't know how

00:34:21.630 --> 00:34:25.289
a car works on a mechanical level, right? And

00:34:25.289 --> 00:34:28.369
we all just agree that you probably don't need

00:34:28.369 --> 00:34:31.650
to know how, you know, the internals of your

00:34:31.650 --> 00:34:34.630
combustion engine. Or we don't need to know how

00:34:34.630 --> 00:34:37.190
lithium batteries work, right? There are just

00:34:37.190 --> 00:34:40.670
some things that we can abstract away. But I

00:34:40.670 --> 00:34:43.150
think we need to have a kind of dialogue about,

00:34:43.429 --> 00:34:46.789
in the age when AI can do all sorts of things,

00:34:47.489 --> 00:34:51.769
it's increasingly able to do, like, I don't know,

00:34:52.030 --> 00:34:56.409
it's most cognitive work that we encounter. We

00:34:56.409 --> 00:34:59.230
need to have a think, like what cognitive skills,

00:34:59.489 --> 00:35:02.510
what skills do we want to keep and what skills

00:35:02.510 --> 00:35:08.050
do we want to delegate to AI and perhaps lose

00:35:08.050 --> 00:35:12.230
over time. So I think there's worth thinking

00:35:12.230 --> 00:35:16.960
about that. I mean, there's some more concrete

00:35:16.960 --> 00:35:19.360
things. So for example, I recommend when I talk

00:35:19.360 --> 00:35:22.059
to organizations about their AI policy, every

00:35:22.059 --> 00:35:25.000
org should have an AI policy now and or at least

00:35:25.000 --> 00:35:27.320
should work on one if you don't have one. And

00:35:27.320 --> 00:35:30.500
but one thing I recommend going into most AI

00:35:30.500 --> 00:35:35.559
policies is to say that any public facing communications

00:35:35.559 --> 00:35:39.179
that features AI generated images, you should

00:35:39.179 --> 00:35:41.920
be very careful about generating images that

00:35:41.920 --> 00:35:44.920
look like they are real. And then and passing

00:35:44.920 --> 00:35:47.019
them off as real, right? So you shouldn't be

00:35:47.019 --> 00:35:50.880
passing off AI generated images as real. But

00:35:50.880 --> 00:35:52.679
you should also be mindful that depending on

00:35:52.679 --> 00:35:55.440
how you present them, people may interpret them

00:35:55.440 --> 00:35:58.780
as being, as suggesting that they are real when

00:35:58.780 --> 00:36:01.900
they're not. So say I'm presenting about some

00:36:01.900 --> 00:36:04.300
research I've done on British farming, and I'm

00:36:04.300 --> 00:36:07.480
presenting like, and we did a survey of 50 farmers.

00:36:08.139 --> 00:36:10.639
If my presentation is full of photorealistic

00:36:10.639 --> 00:36:13.639
AI generated images of farmers, then some of

00:36:13.639 --> 00:36:16.579
the audience might think these are pictures of

00:36:16.579 --> 00:36:19.840
farmers that I did the study with, right? And

00:36:19.840 --> 00:36:23.820
yeah, and so that might not be, it might not

00:36:23.820 --> 00:36:27.340
be good to do that, right? Because people might

00:36:27.340 --> 00:36:31.760
feel lied to, and yeah, they might feel like

00:36:31.760 --> 00:36:35.480
you're kind of misrepresenting yourself. So there

00:36:35.480 --> 00:36:39.780
is a risk there. I think AI... images have have

00:36:39.780 --> 00:36:41.320
a role to play, they can be really great for

00:36:41.320 --> 00:36:43.119
all sorts of things, but there is that thing

00:36:43.119 --> 00:36:45.300
of like, we shouldn't be using it to misrepresent

00:36:45.300 --> 00:36:49.380
the truth. Similarly, I don't think we should

00:36:49.380 --> 00:36:53.679
be generating kind of AI images of factory farming,

00:36:54.000 --> 00:36:56.400
even if we are generating images of things that

00:36:56.400 --> 00:37:01.059
we know are real. We have, I mean, thanks to

00:37:01.059 --> 00:37:03.699
the work of organizations like We Animals, we

00:37:03.699 --> 00:37:09.449
have access to to pictures that show what is

00:37:09.449 --> 00:37:12.789
truly going on in animal agriculture. So I think

00:37:12.789 --> 00:37:18.730
we don't need to kind of create these kind of...

00:37:18.730 --> 00:37:22.690
Well, they are fake, but they might be representing

00:37:22.690 --> 00:37:24.809
something real, but I would still call them fake

00:37:24.809 --> 00:37:27.130
in this case. Because there's a reputational

00:37:27.130 --> 00:37:29.469
risk there, right? There's a risk that if they

00:37:29.469 --> 00:37:31.550
are found out to be AI generated, people think

00:37:31.550 --> 00:37:34.329
that, okay, well, these animal activists are...

00:37:35.059 --> 00:37:38.099
essentially they are lying about how things are,

00:37:38.559 --> 00:37:41.360
and so it can undermine our credibility as a

00:37:41.360 --> 00:37:47.159
movement. So I'd probably say the two types of

00:37:47.159 --> 00:37:49.960
things we should be avoiding is the first one

00:37:49.960 --> 00:37:53.340
is we should have a think about if I'm using

00:37:53.340 --> 00:37:58.119
AI for a thing, am I skills in that thing going

00:37:58.119 --> 00:38:00.880
to degrade over time? And if so, is that okay?

00:38:01.539 --> 00:38:04.519
And then the second thing is Uh, if I'm using

00:38:04.519 --> 00:38:08.539
AI to generate writing or pictures or videos

00:38:08.539 --> 00:38:13.000
or audio, um, is that, is it going to undermine

00:38:13.000 --> 00:38:17.199
my credibility if my audience like finds out,

00:38:17.199 --> 00:38:21.039
uh, knows that, that it is AI generated? I think

00:38:21.039 --> 00:38:26.099
those are the key ones. I want to touch on, um,

00:38:26.320 --> 00:38:30.239
the, the bias that some of those AI models have.

00:38:30.860 --> 00:38:35.219
Um, I think the most. hilarious example that

00:38:35.219 --> 00:38:40.039
I came across of that is when Gemini was launched,

00:38:40.219 --> 00:38:42.320
I think, and people started asking, you know,

00:38:42.320 --> 00:38:45.320
who are the founding fathers and you had pictures

00:38:45.320 --> 00:38:50.219
of like, women and, you know, diverse people

00:38:50.219 --> 00:38:54.440
and it was really funny. But it showed a bias,

00:38:54.440 --> 00:38:58.460
you know, and it wasn't functioning properly

00:38:58.460 --> 00:39:03.130
because the developers had a sort of bias. Now

00:39:03.130 --> 00:39:09.150
is there an anti -vegan, you know, anti -animal

00:39:09.150 --> 00:39:13.369
rights bias in those systems because, you know,

00:39:13.369 --> 00:39:16.670
I've asked a few things and, you know, about

00:39:16.670 --> 00:39:20.670
is there, you know, pus in milk that we drink

00:39:20.670 --> 00:39:25.369
and the answers I got were surprising and I would

00:39:25.369 --> 00:39:29.010
argue with some of them, you know, how true they

00:39:29.010 --> 00:39:33.360
are or, you know, Tell me about the history of

00:39:33.360 --> 00:39:37.960
slaughterhouses. And what I got was not really,

00:39:37.960 --> 00:39:43.440
I guess, reflective of the reality, which is

00:39:43.440 --> 00:39:46.920
harsh and horrifying, but more like, you know,

00:39:47.900 --> 00:39:51.840
something you could tell children. And, you know,

00:39:51.840 --> 00:39:55.500
you could argue, you know, what kind of nuances

00:39:55.500 --> 00:40:00.400
that AI should use in storytelling or delivering

00:40:00.400 --> 00:40:04.800
information, like what to highlight versus what

00:40:04.800 --> 00:40:09.500
to put on the side. So it's a complex question,

00:40:09.760 --> 00:40:12.460
the question of bias. Have you thought of that?

00:40:12.559 --> 00:40:17.460
I know you have, but what are your thoughts on

00:40:17.460 --> 00:40:21.179
that topic? Yeah, yeah. So it's definitely something

00:40:21.179 --> 00:40:23.539
to be concerned about, right? Because I mean,

00:40:23.579 --> 00:40:25.199
really, yeah, we should be concerned about bias

00:40:25.199 --> 00:40:28.050
in any of the in any of our kind of systems that

00:40:28.050 --> 00:40:32.010
we build as humans really and yeah so this is

00:40:32.010 --> 00:40:34.869
this is a case where chat gbt actually again

00:40:34.869 --> 00:40:39.389
performs more like humans than like kind of tech

00:40:39.389 --> 00:40:41.670
so right there's this stereotype that software

00:40:41.670 --> 00:40:47.010
or tech is objective whereas i mean it often

00:40:47.010 --> 00:40:51.829
isn't but with large language models like you

00:40:51.829 --> 00:40:55.119
know chat gbt and gemini they are They're more

00:40:55.119 --> 00:40:57.960
like humans and they've internalized the biases

00:40:57.960 --> 00:41:00.820
of the data that was used to create them and

00:41:00.820 --> 00:41:04.380
that data Mostly just amounts to like a large

00:41:04.380 --> 00:41:08.760
swath of the internet plus most like plus a ton

00:41:08.760 --> 00:41:13.519
of books and basically any data that these companies

00:41:13.519 --> 00:41:17.099
could get their hands on but so it's going to

00:41:17.099 --> 00:41:19.119
reflect a lot of biases that are already in there

00:41:19.119 --> 00:41:22.460
and I found that when it comes to their bias

00:41:22.460 --> 00:41:24.940
when it comes to the biases uh that it shows

00:41:24.940 --> 00:41:30.300
it's often it will show bias in areas where um

00:41:30.300 --> 00:41:32.159
large amounts of the internet shows that same

00:41:32.159 --> 00:41:34.780
bias so for example one of our researchers Billy

00:41:34.780 --> 00:41:36.719
Nicholls who was on this he was on the podcast

00:41:36.719 --> 00:41:40.019
actually and um yeah he so he does work on plant

00:41:40.019 --> 00:41:43.400
-based pet food right and he used uh chatgbt

00:41:43.400 --> 00:41:46.260
to generate to generate a deep research report

00:41:46.260 --> 00:41:51.389
on plant -based pet food and the reported generator

00:41:51.389 --> 00:41:55.610
was quite biased towards meat -based pet food

00:41:55.610 --> 00:41:59.329
being totally fine. And the reason why was because

00:41:59.329 --> 00:42:03.449
most of the sources online about pet food come

00:42:03.449 --> 00:42:06.550
from the pet food industry, and plant -based

00:42:06.550 --> 00:42:09.989
pet food is such a small topic that there just

00:42:09.989 --> 00:42:12.750
isn't a huge amount about it online. And because

00:42:12.750 --> 00:42:15.289
of that, the AI model was just unable to, it

00:42:15.289 --> 00:42:17.550
was unable to find a lot of it, and it wasn't

00:42:17.550 --> 00:42:21.059
in its training data. And so there are those

00:42:21.059 --> 00:42:24.380
kind of issues where you might see bias when

00:42:24.380 --> 00:42:26.760
it comes to information that isn't readily available.

00:42:27.239 --> 00:42:30.139
So I think as a good kind of heuristic that,

00:42:30.139 --> 00:42:34.079
like, if you can't find a piece of it, well,

00:42:34.539 --> 00:42:36.440
whatever the consensus of the first couple of

00:42:36.440 --> 00:42:40.019
pages of Google suggests is kind of where ChatGBT

00:42:40.019 --> 00:42:44.860
is going to... like start in its opinions. However

00:42:44.860 --> 00:42:47.940
these models are capable of quite a lot of like

00:42:47.940 --> 00:42:49.780
critical thinking they can critique themselves

00:42:49.780 --> 00:42:55.239
and so they do I find that they can be very good

00:42:55.239 --> 00:42:59.420
at recognizing their own bias or they can you

00:42:59.420 --> 00:43:02.179
can kind of almost prompt sometimes you can kind

00:43:02.179 --> 00:43:06.639
of prompt bias out of them. So for example if

00:43:06.639 --> 00:43:09.019
you go onto an image generator and say make a

00:43:09.019 --> 00:43:12.519
picture of a bunch of professionals sitting around

00:43:12.519 --> 00:43:17.400
a table having a meeting. It might generate all

00:43:17.400 --> 00:43:21.159
men. It might generate all white people. But

00:43:21.159 --> 00:43:26.500
unlike humans, you can just say something like,

00:43:26.559 --> 00:43:29.400
generate a picture that is diverse in gender

00:43:29.400 --> 00:43:35.599
and race. And it will do that. Or you could do

00:43:35.599 --> 00:43:37.409
a thing where... something I'm interested in

00:43:37.409 --> 00:43:40.650
is how we might be able to use these models to

00:43:40.650 --> 00:43:43.690
overcome our own biases. So, for example, say

00:43:43.690 --> 00:43:45.750
you're the comms team and you're responsible

00:43:45.750 --> 00:43:49.130
for your tweets and your LinkedIn posts. What

00:43:49.130 --> 00:43:53.130
you could do is, before sending out a, you know,

00:43:53.210 --> 00:43:55.110
putting something out into the world, you could

00:43:55.110 --> 00:43:57.510
say, you could upload it to Geminar ChatGBT and

00:43:57.510 --> 00:44:02.380
say, hey, how would, you know, how... how would

00:44:02.380 --> 00:44:05.119
this be perceived by certain communities? So

00:44:05.119 --> 00:44:07.260
you'd say like, how would the autistic community

00:44:07.260 --> 00:44:10.780
perceive this campaign, right? You know, would

00:44:10.780 --> 00:44:14.679
they be upset and why? Or like, how would the

00:44:14.679 --> 00:44:17.239
trans community respond to this LinkedIn post

00:44:17.239 --> 00:44:19.699
and things? Now, obviously, this isn't a substitute

00:44:19.699 --> 00:44:21.860
for actually going and asking those communities

00:44:21.860 --> 00:44:23.500
themselves. Like, this is something we should

00:44:23.500 --> 00:44:26.219
always try and do. But like, to be quite honest,

00:44:26.280 --> 00:44:29.110
it's just not always... possible, right? And

00:44:29.110 --> 00:44:30.929
it's certainly not possible for every single

00:44:30.929 --> 00:44:33.849
tweet your organization puts out, right? And

00:44:33.849 --> 00:44:37.030
so it can often be a very good proxy. If you

00:44:37.030 --> 00:44:40.030
ask ChatGBT, like, what would a young male from

00:44:40.030 --> 00:44:44.150
Yorkshire think about this thing? It's usually,

00:44:44.329 --> 00:44:49.829
like, relatively good. And so we can use this

00:44:49.829 --> 00:44:52.349
to essentially uncover our own biases, where

00:44:52.349 --> 00:44:55.570
we can say, OK, like, this is our new campaign,

00:44:55.630 --> 00:44:58.449
how is it going to be perceived by these different

00:44:58.449 --> 00:45:02.409
diverse communities? You could say, is our new

00:45:02.409 --> 00:45:05.389
campaign insensitive to people who have suffered

00:45:05.389 --> 00:45:12.610
from miscarriage or who were held in concentration

00:45:12.610 --> 00:45:16.750
camps and things like that, or refugees? So we

00:45:16.750 --> 00:45:21.030
can do that kind of stuff too and we can also...

00:45:21.420 --> 00:45:23.400
Yeah, well, basically that's it. That's it. We

00:45:23.400 --> 00:45:25.519
can use it to be I think we can use it to to

00:45:25.519 --> 00:45:30.960
be to be better essentially Yeah, it's funny

00:45:30.960 --> 00:45:34.619
to see George Washington with my curly hair but

00:45:34.619 --> 00:45:37.400
you know, there are there is danger in those

00:45:37.400 --> 00:45:43.000
biases and do you think we can Contact the developers

00:45:43.000 --> 00:45:46.519
the people behind, you know the AI to ask them,

00:45:46.619 --> 00:45:49.340
you know, hey we're vegans, we're from an animal

00:45:49.340 --> 00:45:51.920
rights organization, and we saw that this is

00:45:51.920 --> 00:45:54.940
the kind of answers that your AI model gives,

00:45:54.960 --> 00:45:59.159
and it's not truthful. And those are our suggestions.

00:45:59.440 --> 00:46:04.980
Is that a possible thing we can do? It is possible.

00:46:05.179 --> 00:46:07.280
I don't know how helpful it'll be. So one thing

00:46:07.280 --> 00:46:10.840
you can do is anytime for most AI tools, there's

00:46:10.840 --> 00:46:12.420
after it gives an answer, there's a little thumbs

00:46:12.420 --> 00:46:15.139
up and a thumbs down option. If you click a thumbs

00:46:15.139 --> 00:46:18.219
down, you can say this was incorrect right one

00:46:18.219 --> 00:46:21.119
thing to note about that is um it's kind of obvious

00:46:21.119 --> 00:46:22.420
when you say allow but a lot of people don't

00:46:22.420 --> 00:46:24.780
realize it that will send your entire conversation

00:46:24.780 --> 00:46:28.320
that entire conversation to the developers and

00:46:28.320 --> 00:46:31.119
they will keep it to use it as training so like

00:46:31.119 --> 00:46:34.239
be wary that you you haven't you know discussed

00:46:34.239 --> 00:46:36.500
breaking into a factory farm earlier in the conversation

00:46:36.500 --> 00:46:40.840
and um and so yeah we can we can express our

00:46:40.840 --> 00:46:45.079
feedback um we can also another thing we can

00:46:45.079 --> 00:46:50.099
do is a lot of these models, to assess the quality,

00:46:50.380 --> 00:46:52.260
the models are evaluated against benchmarks,

00:46:52.360 --> 00:46:54.639
right, and these models, because they all want,

00:46:54.679 --> 00:46:56.719
you know, Chachi Buti wants to be the best, Gemini

00:46:56.719 --> 00:46:58.000
wants to be the best, Claude wants to be the

00:46:58.000 --> 00:47:00.019
best, they're all trying to outdo each other

00:47:00.019 --> 00:47:02.599
on getting the highest scores on all these benchmarks,

00:47:02.960 --> 00:47:06.119
and that can be programming quality, it can be

00:47:06.119 --> 00:47:10.099
like... the wittiness of its writing, that kind

00:47:10.099 --> 00:47:13.500
of thing. So there are some organizations, once

00:47:13.500 --> 00:47:15.980
again, Centium Futures in there, but also other

00:47:15.980 --> 00:47:19.760
organizations such as OpenPause, who are developing

00:47:19.760 --> 00:47:25.300
basically measures, tests that you can put AI

00:47:25.300 --> 00:47:29.139
models through, and it reveals how much bias

00:47:29.139 --> 00:47:34.039
against animals they have, right? And so these

00:47:34.039 --> 00:47:37.480
can even even measure how speciesist these models

00:47:37.480 --> 00:47:39.840
are. And so once we have those benchmarks, we

00:47:39.840 --> 00:47:42.079
can start basically lobbying the companies to

00:47:42.079 --> 00:47:44.320
be like, look, this is a benchmark that you should

00:47:44.320 --> 00:47:46.340
take seriously. And every time you release a

00:47:46.340 --> 00:47:48.500
new model, you should be aiming to try and get

00:47:48.500 --> 00:47:51.679
the model to improve on this benchmark. Now,

00:47:51.940 --> 00:47:55.400
they're not always going to listen, but they

00:47:55.400 --> 00:48:02.059
might, and it's worth trying. And another thing

00:48:02.059 --> 00:48:04.420
we... can do when it comes to improving these

00:48:04.420 --> 00:48:06.579
things is that they are just getting their training

00:48:06.579 --> 00:48:09.960
data from the internet and so for example the

00:48:09.960 --> 00:48:15.300
reason why chat gbt will produce a a report on

00:48:15.300 --> 00:48:17.760
plant -based pet food that is biased towards

00:48:17.760 --> 00:48:20.599
the meat -based pet industry is because there

00:48:20.599 --> 00:48:22.820
just doesn't exist that much stuff on plant -based

00:48:22.820 --> 00:48:25.400
pet food online so but once we start putting

00:48:25.400 --> 00:48:27.519
that stuff out there it's going to end up in

00:48:27.519 --> 00:48:29.960
the training data and it's going to come become

00:48:29.960 --> 00:48:34.699
reflected in in these models' opinions and views

00:48:34.699 --> 00:48:38.219
of the world. So, for example, these models have

00:48:38.219 --> 00:48:41.659
quite good, they have quite nuanced views on,

00:48:41.659 --> 00:48:44.340
for example, factory farming and plant -based

00:48:44.340 --> 00:48:46.880
nutrition. So if you say, like, if you ask Chachipati

00:48:46.880 --> 00:48:50.739
and say, you know, like, is veganism unhealthy?

00:48:51.059 --> 00:48:52.900
And it's going to come back and say, no, a well

00:48:52.900 --> 00:48:54.820
-planned, like, it will say studies repeatedly

00:48:54.820 --> 00:48:58.010
show that a well -planned vegan diet can be perfectly

00:48:58.010 --> 00:49:00.010
healthy or even more healthy than the average

00:49:00.010 --> 00:49:02.929
and the reason that it says that is because that

00:49:02.929 --> 00:49:05.809
is the scientific consensus and it has ingested

00:49:05.809 --> 00:49:09.869
a lot of those papers right and so when it comes

00:49:09.869 --> 00:49:12.670
to steering AI one thing we can do is just the

00:49:12.670 --> 00:49:16.050
information we want to that we we want to be

00:49:16.050 --> 00:49:18.329
kind of in there we can just put it out into

00:49:18.329 --> 00:49:23.050
the world so you can write story you could write

00:49:23.050 --> 00:49:24.550
you know science fiction you could write fiction

00:49:24.550 --> 00:49:29.360
stories about that about a future where we are

00:49:29.360 --> 00:49:32.420
all kind of we are all kinder to animals or post

00:49:32.420 --> 00:49:35.800
factory farming society and that's you know you

00:49:35.800 --> 00:49:38.219
could blog about you can blog about animal issues

00:49:38.219 --> 00:49:40.860
you can post on twitter all of that stuff is

00:49:40.860 --> 00:49:42.980
going to end up in the in the models because

00:49:42.980 --> 00:49:44.940
they're scraping as much of the internet as they

00:49:44.940 --> 00:49:49.079
can and and so they will hoover up the stuff

00:49:49.079 --> 00:49:51.880
that you put out there for better or worse and

00:49:51.880 --> 00:49:54.159
so if you put pro -animal stuff out there then

00:49:54.159 --> 00:49:58.340
it will it will end up in the model. Unfortunately,

00:49:58.440 --> 00:50:00.320
there's, you know, there's not that many, there's

00:50:00.320 --> 00:50:03.059
not that many of us vegans in the world. But

00:50:03.059 --> 00:50:06.739
if, yeah, people should just know that like the

00:50:06.739 --> 00:50:09.280
work that you're putting out there is, is getting

00:50:09.280 --> 00:50:14.800
seen. That's, that's the assumption. But if we

00:50:14.800 --> 00:50:19.099
can lobby them, can Big Hag also lobby them?

00:50:19.300 --> 00:50:23.320
Like here, take this Millions of dollars and

00:50:23.320 --> 00:50:28.940
yes, have that bias against those animal rights

00:50:28.940 --> 00:50:33.679
people. And another thing they could ask for

00:50:33.679 --> 00:50:39.500
is data from animal rights activists. Is that

00:50:39.500 --> 00:50:43.519
a risk we should consider? Yeah. So you're right

00:50:43.519 --> 00:50:45.860
in that anything we can do, there's a chance

00:50:45.860 --> 00:50:49.780
that industrial animal agriculture will will

00:50:49.780 --> 00:50:52.960
also do. One advantage we have is that they're

00:50:52.960 --> 00:50:54.659
all for -profit companies that are competing

00:50:54.659 --> 00:50:57.260
with each other, so they're not very incentivized

00:50:57.260 --> 00:51:00.780
to provide a lot of that information for AI model

00:51:00.780 --> 00:51:05.000
training. So, you know, if you personally, if

00:51:05.000 --> 00:51:09.179
any listeners here have tried to investigate

00:51:09.179 --> 00:51:11.739
a lot of these big corporations and found that

00:51:11.739 --> 00:51:14.480
some information is frustratingly private, you

00:51:14.480 --> 00:51:16.840
know, that's because they're They are, they're,

00:51:16.840 --> 00:51:18.800
they're for -profit companies that are competing

00:51:18.800 --> 00:51:20.860
with each other. So they don't want to put all

00:51:20.860 --> 00:51:22.920
of that online. And so a lot of it's just not

00:51:22.920 --> 00:51:24.380
going to end up in the training data for that,

00:51:24.380 --> 00:51:30.340
that reason. I meant, sorry, sorry. I meant they

00:51:30.340 --> 00:51:32.739
could, you know, ask for our data, you know,

00:51:32.880 --> 00:51:35.860
private data, what we are inputting in those

00:51:35.860 --> 00:51:40.639
AI models. Oh, I see what you mean. Yes. So,

00:51:40.639 --> 00:51:44.599
well, so basically, so a lot of these, uh, The

00:51:44.599 --> 00:51:47.639
companies take privacy very, very seriously because

00:51:47.639 --> 00:51:49.960
ultimately they want you to keep paying for the

00:51:49.960 --> 00:51:55.039
service. So these model providers are very concerned

00:51:55.039 --> 00:51:59.260
with privacy. So they wouldn't just, for example,

00:51:59.539 --> 00:52:04.400
I would be extremely surprised if any company

00:52:04.400 --> 00:52:08.480
could pay OpenAI to hand over chat GPT data.

00:52:09.500 --> 00:52:13.659
I strongly doubt that's the case. However, One

00:52:13.659 --> 00:52:17.480
thing to bear in mind is that depending on the

00:52:17.480 --> 00:52:21.300
settings you have on AI, these models may be

00:52:21.300 --> 00:52:24.039
storing more or less information about you. So,

00:52:24.260 --> 00:52:28.300
for example, if you go into ChatGBT, there will

00:52:28.300 --> 00:52:30.000
be, in the settings, there's an option called

00:52:30.000 --> 00:52:32.699
improve the model for everyone, which sounds

00:52:32.699 --> 00:52:35.559
very benevolent. But what it actually means is

00:52:35.559 --> 00:52:39.460
that OpenAI reserved the right to keep... a record

00:52:39.460 --> 00:52:41.539
of all the chats you've had, and when they build

00:52:41.539 --> 00:52:44.900
the next version of ChatGBT, they will use your

00:52:44.900 --> 00:52:48.360
conversations to train it. Basically, they will

00:52:48.360 --> 00:52:51.440
have automated systems that will analyze the

00:52:51.440 --> 00:52:53.619
conversations, and they'll figure out, okay,

00:52:53.940 --> 00:52:56.079
was this a conversation where ChatGBT gave you

00:52:56.079 --> 00:52:58.699
what you wanted? In which case, you know, we

00:52:58.699 --> 00:53:01.099
want more of that. And where are the conversations

00:53:01.099 --> 00:53:03.800
where ChatGBT didn't give you what you wanted?

00:53:04.079 --> 00:53:06.860
And then the new model will be trained to not

00:53:06.860 --> 00:53:10.380
do that. And so your data, the things you put

00:53:10.380 --> 00:53:13.280
into these models will end up, could end up kind

00:53:13.280 --> 00:53:16.139
of shaping the future ones, which might be good

00:53:16.139 --> 00:53:18.179
or it might be bad, right? If it's your private

00:53:18.179 --> 00:53:20.800
data, you might not want it in there. But you

00:53:20.800 --> 00:53:23.579
also might want your opinions and values to end

00:53:23.579 --> 00:53:25.599
up in the model in some way. So that's something

00:53:25.599 --> 00:53:29.780
to be aware of. Another thing is that because

00:53:29.780 --> 00:53:32.840
it's storing this information, depending on the

00:53:32.840 --> 00:53:36.480
country you're in, this will vary by law, like...

00:53:36.300 --> 00:53:38.519
law enforcement may be able to like subpoena

00:53:38.519 --> 00:53:41.639
it basically like they may be able to um yeah

00:53:41.639 --> 00:53:46.260
so like if you are doing that kind of work um

00:53:46.260 --> 00:53:50.019
then it may be that law enforcement is able to

00:53:50.019 --> 00:53:54.039
acquire your chat gbt logs to um yeah if if they

00:53:54.039 --> 00:53:57.360
think it is relevant to an ongoing case um so

00:53:57.360 --> 00:54:00.000
that's something to be very aware of um usually

00:54:00.000 --> 00:54:02.239
you can mitigate most of these things quite simply

00:54:02.239 --> 00:54:05.679
so for example in chat gbt you can go into the

00:54:05.679 --> 00:54:08.219
settings and you can turn off, improve the model

00:54:08.219 --> 00:54:12.539
for everyone. And then also, if you right -click

00:54:12.539 --> 00:54:15.019
on any particular chat you've had with ChatGBT,

00:54:15.139 --> 00:54:18.119
you can just delete it and then OpenAI will basically,

00:54:18.340 --> 00:54:21.139
is basically required to get rid of that within

00:54:21.139 --> 00:54:26.340
a few months. And so, and that's usually sufficient.

00:54:27.639 --> 00:54:29.400
But again, this depends on the data protection

00:54:29.400 --> 00:54:31.739
laws of where you are. So for example, in the

00:54:31.739 --> 00:54:35.289
UK, And in Europe we have like GDPR, whereas

00:54:35.289 --> 00:54:37.929
in the States you have different kind of laws

00:54:37.929 --> 00:54:41.110
as well. So it's worth looking into those things.

00:54:41.210 --> 00:54:43.889
I don't think industrial animal agriculture will

00:54:43.889 --> 00:54:46.630
come for your chat GBT logs, but it may be that

00:54:46.630 --> 00:54:49.550
depending on the situation, the police might

00:54:49.550 --> 00:54:51.710
manage to get a hold of them. So it's something

00:54:51.710 --> 00:54:57.650
to be aware of. Also, your social media activity

00:54:57.650 --> 00:55:02.860
might also be used for AI training. Um, I disabled

00:55:02.860 --> 00:55:07.199
that on my LinkedIn. Um, but it's also in my,

00:55:07.199 --> 00:55:10.360
uh, Instagram. I, I need to figure out how to

00:55:10.360 --> 00:55:14.820
disable it. Uh, but yeah, be careful about, um,

00:55:14.980 --> 00:55:19.440
the privacy aspect of, of AI. Yeah, definitely.

00:55:19.639 --> 00:55:22.239
And so the, I mean, yeah, you raise a good point,

00:55:22.400 --> 00:55:24.099
particularly around things like LinkedIn, right?

00:55:24.139 --> 00:55:27.519
Because for example, so the average person probably

00:55:27.519 --> 00:55:30.570
doesn't think that Like, why would LinkedIn collect

00:55:30.570 --> 00:55:33.030
data for AI? Like, I've never seen a LinkedIn

00:55:33.030 --> 00:55:37.449
AI thing. Like, you know what I mean? But, yeah,

00:55:37.769 --> 00:55:42.210
companies, lots of companies are trying to integrate

00:55:42.210 --> 00:55:45.449
AI into their products. And for them to do that

00:55:45.449 --> 00:55:48.070
effectively, they need lots of their users' data.

00:55:48.449 --> 00:55:51.110
So it is really something where, like, basically,

00:55:52.190 --> 00:55:53.769
I know this is gonna be tedious, but, like, every

00:55:53.769 --> 00:55:56.690
piece of software you use, every app, every website

00:55:56.690 --> 00:56:01.340
that you... contribute data to, you should be

00:56:01.340 --> 00:56:04.460
digging through the settings just in case they

00:56:04.460 --> 00:56:10.079
are using that data for the training of new AI

00:56:10.079 --> 00:56:12.380
features. I mean, this isn't something that is

00:56:12.380 --> 00:56:15.260
actually new, to be honest. So it has always

00:56:15.260 --> 00:56:18.460
been the case that a lot of these companies,

00:56:18.760 --> 00:56:22.139
websites, apps have been trying to record as

00:56:22.139 --> 00:56:24.559
much information about you as possible. In the

00:56:24.559 --> 00:56:26.679
past, they would have just sold packets of that

00:56:26.679 --> 00:56:31.579
information to data brokers or used it for targeted

00:56:31.579 --> 00:56:34.519
advertising. But now we've got this added dimension

00:56:34.519 --> 00:56:36.940
that it might be used for building AI features

00:56:36.940 --> 00:56:40.039
as well. So it's worth looking into the privacy

00:56:40.039 --> 00:56:43.199
settings and things. It is worth noting, though,

00:56:43.400 --> 00:56:46.940
that typically, if you get... There are often...

00:56:46.570 --> 00:56:49.110
free versions of stuff, paid versions of stuff,

00:56:49.369 --> 00:56:51.510
and then there'll be enterprise versions of stuff,

00:56:51.630 --> 00:56:54.489
right? So you can get an enterprise version of

00:56:54.489 --> 00:56:58.590
Gemini for your company. And enterprise versions

00:56:58.590 --> 00:57:01.710
usually have the strictest privacy. So for example,

00:57:02.010 --> 00:57:05.710
if your company gives you an email through Gmail,

00:57:06.929 --> 00:57:11.679
then Google has pledged not to... to use your

00:57:11.679 --> 00:57:14.239
personal emails to create software products.

00:57:14.500 --> 00:57:17.099
And that's always been the case. And now that

00:57:17.099 --> 00:57:21.199
extends to, say for example, Google Gemini. Whereas

00:57:21.199 --> 00:57:24.179
on the other hand, if you're using a free model

00:57:24.179 --> 00:57:27.679
or a free version of a tool, you should be very,

00:57:27.679 --> 00:57:32.579
very careful about checking the privacy policies

00:57:32.579 --> 00:57:37.480
because as the old saying goes, if something

00:57:37.480 --> 00:57:40.079
is free, then you are the product. Um, that's

00:57:40.079 --> 00:57:42.480
not always true with AI, but like the, yeah,

00:57:42.659 --> 00:57:44.800
it may be the case that some of these tools are

00:57:44.800 --> 00:57:47.199
free because they are hoping to collect as much

00:57:47.199 --> 00:57:50.079
data from you as possible. Um, but, uh, so you

00:57:50.079 --> 00:57:53.380
could look into, is that the case? And if so,

00:57:53.599 --> 00:57:57.059
can I turn it off or find alternatives? Yeah.

00:57:57.059 --> 00:58:00.360
Thank you for including all software systems.

00:58:00.739 --> 00:58:03.880
You know, I had to do it with a co -pilot and

00:58:03.880 --> 00:58:08.679
word. Um, for privacy, but also because it was

00:58:08.679 --> 00:58:12.159
slowing down the whole thing experience. So,

00:58:12.800 --> 00:58:17.300
yeah. Wonderful. Richie, I think we covered lots

00:58:17.300 --> 00:58:21.420
of ground. Did you want to add something before

00:58:21.420 --> 00:58:24.940
we stop the conversation? Yeah, I mean, I just

00:58:24.940 --> 00:58:27.679
want to because here we've talked about like

00:58:27.679 --> 00:58:30.019
addressing concerns. We've talked a little bit

00:58:30.019 --> 00:58:33.050
about the benefits, but I do want to like to

00:58:33.050 --> 00:58:34.369
emphasize them a little bit more, like we've

00:58:34.369 --> 00:58:36.130
got, you know, in terms of some case studies

00:58:36.130 --> 00:58:40.269
and things, because I really do think that the

00:58:40.269 --> 00:58:43.469
animal movement has a huge amount to gain by

00:58:43.469 --> 00:58:47.530
leaning into these tools. And I'm seeing it more

00:58:47.530 --> 00:58:51.789
and more. So, for example, one case study I'd

00:58:51.789 --> 00:58:54.869
like to highlight is that we had, there's an

00:58:54.869 --> 00:58:56.889
organization in the UK called the Conservative

00:58:56.889 --> 00:58:58.909
Animal Welfare Foundation. So they're the voice

00:58:58.909 --> 00:59:01.289
of animal welfare in the Conservative Party.

00:59:03.000 --> 00:59:07.860
And one of the founders, Chris Platt, he's an

00:59:07.860 --> 00:59:10.420
older gentleman, but he has really leaned into

00:59:10.420 --> 00:59:12.860
AI. He was one of our most engaged participants

00:59:12.860 --> 00:59:15.039
on our Amplify for Animals course. And one thing

00:59:15.039 --> 00:59:18.099
he told me was that it's a very small org, less

00:59:18.099 --> 00:59:21.380
than 10 people, but two of his org went on maternity

00:59:21.380 --> 00:59:23.679
leave this year. And with an org that small,

00:59:23.739 --> 00:59:26.460
that would often mean kind of grinding to a halt,

00:59:26.579 --> 00:59:30.179
essentially, or just drowning in work. But he

00:59:30.179 --> 00:59:33.059
said that because he has been making so much

00:59:33.059 --> 00:59:38.079
use of AI, he still waits for them to return

00:59:38.079 --> 00:59:41.239
and he couldn't replace them with AI, but he

00:59:41.239 --> 00:59:45.179
could manage to keep kind of everything running

00:59:45.179 --> 00:59:48.699
without burning out by greatly increasing his

00:59:48.699 --> 00:59:52.400
own productivity using AI, which is amazing because

00:59:52.400 --> 00:59:55.380
this is a small org that is doing valuable work

00:59:55.380 --> 00:59:58.739
and something like double maternity leave can,

00:59:59.099 --> 01:00:01.780
it's just one of those things that you often

01:00:01.780 --> 01:00:04.860
can often come out of nowhere and all of a sudden

01:00:04.860 --> 01:00:11.340
you lose the ability to kind of keep having that

01:00:11.340 --> 01:00:13.820
impact for animals. But he's managed to use AI

01:00:13.820 --> 01:00:17.260
to keep the ship going and things like that,

01:00:17.440 --> 01:00:22.420
which is just really amazing to hear. And another

01:00:22.420 --> 01:00:27.079
thing as well on the front of we've had Another

01:00:27.079 --> 01:00:28.920
organization that I do work with, Greener by

01:00:28.920 --> 01:00:32.099
default, they help caterers to shift to more

01:00:32.099 --> 01:00:36.820
plant forward diets. And so I've been building

01:00:36.820 --> 01:00:44.199
out the data analytics side of their org because

01:00:44.199 --> 01:00:48.579
I'm a data scientist by training. And so I've

01:00:48.579 --> 01:00:51.559
built a system that kind of mostly automates

01:00:51.559 --> 01:00:53.940
all of the data analysis that they need as an

01:00:53.940 --> 01:00:58.679
organization. And I estimate that using AI coding

01:00:58.679 --> 01:01:02.260
assistants, like programming assistants, I have

01:01:02.260 --> 01:01:05.099
been three to four times more productive this

01:01:05.099 --> 01:01:08.199
year than I was pre these assistants, right?

01:01:08.219 --> 01:01:10.219
Like in terms of the amount of code I have shipped,

01:01:10.320 --> 01:01:12.480
the amount of stuff I've delivered, like three

01:01:12.480 --> 01:01:16.019
to four times, right? It's as if, you know, it's

01:01:16.019 --> 01:01:18.400
as if there are several of me in the organization.

01:01:19.159 --> 01:01:22.139
These are the potential improvements where...

01:01:22.239 --> 01:01:25.159
we're talking about here. And because it's such

01:01:25.159 --> 01:01:29.119
a general purpose technology, it can help you

01:01:29.119 --> 01:01:34.179
write donor reports, it can help you research

01:01:34.179 --> 01:01:39.300
for campaigns, it can help you with your HR policies.

01:01:40.059 --> 01:01:43.500
It's such a general purpose tool that it means

01:01:43.500 --> 01:01:47.579
that there is almost every role in the whole

01:01:47.579 --> 01:01:51.969
movement. can make use of AI and can see those

01:01:51.969 --> 01:01:55.070
benefits. It's not just for techies. It's not

01:01:55.070 --> 01:01:57.849
just for young people or people in the West.

01:01:57.969 --> 01:02:02.409
It is really for all of us. And kind of to that

01:02:02.409 --> 01:02:05.010
point, I think it's just about experimenting,

01:02:05.469 --> 01:02:07.730
really. The more you experiment with it, the

01:02:07.730 --> 01:02:11.010
more you're likely to find these benefits. So

01:02:11.010 --> 01:02:14.909
I think if we constantly experiment with these

01:02:14.909 --> 01:02:18.190
tools and we share them, We share the results

01:02:18.190 --> 01:02:20.170
with each other. We share the wins. We share

01:02:20.170 --> 01:02:22.929
the the learnings. Then I think as a movement,

01:02:23.030 --> 01:02:28.389
we can really catch this wave and and just and

01:02:28.389 --> 01:02:32.389
and really kind of use it to have like a huge

01:02:32.389 --> 01:02:35.690
like a much, much bigger impact for for animals.

01:02:37.150 --> 01:02:42.329
I love that. I love hearing those stories. Thank

01:02:42.329 --> 01:02:45.590
you for sharing. And yes, please, people follow

01:02:45.590 --> 01:02:49.489
the footsteps. become more efficient in your

01:02:49.489 --> 01:02:53.809
advocacy work by using AI. I also use AI. You

01:02:53.809 --> 01:02:56.730
know, I'm fundraising consultant. I tell my clients,

01:02:57.030 --> 01:03:00.710
you know, your money is precious. I want to use

01:03:00.710 --> 01:03:03.989
it to be more productive and do more with what

01:03:03.989 --> 01:03:08.829
you've hired me to do. So yes, it is like I said

01:03:08.829 --> 01:03:11.230
in the beginning, you know, it is a necessity

01:03:11.230 --> 01:03:15.610
and people need to start learning to use it.

01:03:16.570 --> 01:03:20.309
And yeah, thank you. Thank you so much, Richie.

01:03:21.389 --> 01:03:23.969
And thank you. Thank you for your advocacy work.

01:03:24.050 --> 01:03:25.849
Thank you for having taken the time to answer

01:03:25.849 --> 01:03:31.329
my questions. This has been very insightful conversations

01:03:31.329 --> 01:03:34.150
I've learned a lot and I'm sure listeners too.

01:03:34.690 --> 01:03:38.429
So we're all very grateful. Thanks. Yeah. Thanks

01:03:38.429 --> 01:03:42.070
for having me. And to all the listeners in the

01:03:42.070 --> 01:04:04.639
new year, keep a lookout for... Thank you everyone

01:04:04.639 --> 01:04:07.800
for listening. I kindly invite you to share this

01:04:07.800 --> 01:04:10.639
podcast with the vegans you know. Let's encourage

01:04:10.639 --> 01:04:14.619
more people to take action. Again, thank you

01:04:14.619 --> 01:04:17.659
so much for caring, and I will see you next Tuesday

01:04:17.659 --> 01:04:19.079
for a new episode.