Player is loading...
Generative AI and Threats to Democracy: What Political Psychology Can Tell Us
Dr Ashley Thornton, Geneva Graduate Institute
Wednesday, Feb. 21, 2024 · noon · 19m 20s
With the recent surge of publicly available AI technologies like Chat GPT, the threat AI may pose to democracy has become an important topic. Given concerns over the past decade about the rise of populist authoritarianism in the United States and Europe—notably the demonization of political opponents, threats to voting, and refusal to accept electoral losses—do recent developments in AI further exacerbate these concerns? Using research from political psychology, I identify factors such as emotional amplification and echo-chamber effects that may potentially worsen these concerns and further destabilize democracies. I will also propose several ideas about how we might attenuate AI’s potential negative impacts on the democratic process.
Embed
Transcriptions
Note: this content has been automatically generated.
00:00:00
so thank you very much for that kind introduction
00:00:03
and i'm really delighted to be here today participating um
00:00:06
it is interesting in interdisciplinary conversation about a i and
00:00:10
markers see an as you can see i'm going to
00:00:14
be talking today about generated a i am threats to democracy i'm going to be doing not from the
00:00:18
political psychology perspective which is perhaps a little bit of a different perspective than what we've seen so far today
00:00:25
i'm sorry go so quick overview of the talk that i'm going to give today um the first thing i wanna do is i
00:00:34
wanna define with authoritarian populism isn't i apologise for this but i
00:00:38
can't help it i'm a political scientist and all habits die hard
00:00:42
um i don't want to discuss a i as a danger to democracy
00:00:47
and i wanna outline to specific threats that i can see that are relevant
00:00:51
thirdly i'd like to outline the mechanisms underlying why these threads um i dangerous to democracy and
00:00:57
i wanna do this using to specific concepts
00:00:59
from political psychology motivated reasoning um and emotional amplification
00:01:04
and i want to wrap up by proposing some ideas about
00:01:08
what maybe we could do to mitigate some of the threats um
00:01:12
to to to democracy so i have a code here i'm from donald
00:01:17
tramp it's from twenty twenty three he waited at the conservative political action conference
00:01:21
um and he said twenty sixteen i declared i and
00:01:24
your voice today i add i endured warrior i endure justice
00:01:29
and for those who've been wronged and betrayed i am your retribution um and i think you should just
00:01:34
keep the scroll kind of in the back of your mind as i go through the definition of authoritarian populism
00:01:41
um okay and a quick overview of the argument that i plan to make
00:01:45
um it's this is highly stylised obviously but i want to argue
00:01:50
that a i turbo charge is the creation and the spread of disinformation
00:01:54
um and it doesn't this view these two
00:01:56
mechanisms i mentioned before motivated reasoning and emotional amplification
00:02:01
uh this then creates and disinformation echo
00:02:05
chamber which increases the likelihood that people are
00:02:08
then going to except bogus information that they come in contact with in the future
00:02:13
um and it's finally has behavioural consequences that friend
00:02:16
of democracy um like the refusal to accept election results
00:02:21
a violent sex after so authoritarian populism what is it um uh
00:02:31
authoritarian capitalism is marked by shoe distinctive features the first
00:02:36
is that um people who believe this political parties leaders voters they're anti establishment
00:02:41
um so this means that they're against liberal democratic institutions so the congress which is an elected body
00:02:48
they're also against you let 'em existing
00:02:51
that's the important work existing collective elected representatives
00:02:54
um they would describe these people as being self serving and corrupt and that they should be removed
00:03:01
the second feature um is that people who subscribe to
00:03:04
this ideology believe that society is divided into two groups
00:03:08
so you have the in group which is good ordinary people people like them um
00:03:14
and then you had the out group and this would consist of elites um immigrants
00:03:19
racial and ethnic minorities those who disagree with their views other are not like them
00:03:25
um we might even throw a public intellectuals and academics into the same category
00:03:34
um and there are some important drivers that lead people
00:03:37
to subscribe to this point of view this world view
00:03:41
so the first is decline interest in government this it was mentioned in earlier talking
00:03:47
it's really important um people feel that the government is no longer working for them
00:03:53
i'm responding to their preferences and they also feel that
00:03:56
they're not being fairly represented at the federal level um
00:04:02
a second driver is the perception the democratic institutions are working or they fail
00:04:06
to solve problems in the u. s. context this could be demonstrated by frustration
00:04:11
i'm from people who think that legislation should be
00:04:16
passed more quickly it's not passed it flanders it fails
00:04:20
um and they get frustrated also by things like threats
00:04:22
of government shut downs and hostage taking indigo she asians
00:04:27
a third driver would be the rise in partisanship um
00:04:31
and ideological polarisation there are more people today who subscribe
00:04:35
to one side or the other in the u. s.
00:04:37
context this is democrats versus republicans than ever before um
00:04:43
and it's not just a question of uh yes i identify as one or the other
00:04:47
it actually goes so deep that if you ask people they would say well i don't even like people
00:04:52
who call themselves democrats i don't like people who call themselves
00:04:55
republicans even though they don't know them um they would say
00:05:00
if my child married a democrat or if my child very republican i would
00:05:04
be really upset so the the depth of the animosity is is important there
00:05:10
and the final point um it has to do with jim innovation of perceived political opposition and the willingness to do
00:05:16
whatever it takes to ensure that the opposition doesn't come
00:05:19
back into power um and this is a really important quote
00:05:23
more than ever before we can see it's not simply enough to just have a policy disagreement um
00:05:31
quote a chunk of justin i'm from new hampshire in the us full disclosure and during the run up
00:05:36
to the vote you need a speech uh in november the run up to the primary in twenty twenty three
00:05:43
and you said in our of our great veterans on veterans day we pledge to you that
00:05:46
we will root out the common is the marks is the fascism the radical left the us
00:05:51
that live like vermin within the compounds of our country that lie
00:05:55
and steal and she on election as you can tell that i'm
00:05:59
this is not well policy this is about power and
00:06:02
this is about retaining power in terms of authoritarian leaders
00:06:07
trump is an obvious example uh but there are others so you can think of mouldy in india
00:06:13
um where ben and hungry and air going in turkey for example
00:06:17
they would all fit the same um classification as an authoritarian populist so
00:06:27
how's if we think about these what will become a yeah if we think about these different
00:06:33
drivers um we can evaluate how is the us
00:06:38
doing so in terms of declining interest in government
00:06:41
one percent of americans trust the government to do the right
00:06:44
thing just about always and fifteen percent most of the time
00:06:48
um sixteen present that's really well and to put it in a little bit of context
00:06:53
um they the national election survey started asking about this in nineteen fifty eight trust
00:06:57
in government and this is the lowest value they've ever had since nineteen fifty eight um
00:07:03
in terms of democratic institutions not working americans with a great deal are quite a lot
00:07:08
of confidence in the supreme court is twenty five percent presidency twenty three percent in congress
00:07:14
um seven percent gallup first measured competence in institutions
00:07:19
in nineteen seventy three and this is the lowest
00:07:21
this was twenty twenty two this was the lowest value for all three branches of the federal government that they have ever seen
00:07:28
increase partisanship and polarisation sixty two percent of republicans and fifty four
00:07:33
percent of democrats have very unfair favourably views of the other party um
00:07:38
in terms of the demon is asian of the out group and ensure they don't return to power
00:07:43
fifty one percent of republicans and that's fifty nine percent of strong republicans
00:07:48
um say they like political years that claim trump one the
00:07:52
twenty twenty election even though we know that that's not true
00:07:55
so the take home message for me um is that the problem that's still authoritarian
00:08:01
populism are only getting worse and the overall picture for american democracy is worrying indeed
00:08:07
and it's important to note that this problem is not only confined to the u. s. um democracies
00:08:13
in europe are also vulnerable you can see some of the same trends for so i wanna quickly go
00:08:21
through um a couple of friends that i identified um to democracy and how this can next to
00:08:30
a guy because i've i've set the stage with democracy in the us so i wanna make the connection
00:08:36
so as i said before the current state of the in the
00:08:38
markets these today is a golden opportunity for actors both foreign and domestic
00:08:44
um not to use disinformation propaganda a destructive troops
00:08:48
to try to interfere in the outcome of an election
00:08:51
uh in into this mass we've already got a mass so into this
00:08:55
mess comes generated a i um and the problem is that the technology
00:09:01
drastically lowers the cost of creating and also spreading disinformation propaganda
00:09:06
and destructive troops um that we can already see this happening
00:09:11
in terms of things like d. fakes okay this is the first example of and generate a very i try to democracy
00:09:19
they're increasingly realistic and very difficult to identify and they are virtually free
00:09:24
and easy to do i found this company on line pair a i
00:09:28
make your favourite celebrity say anything to roast your friends send
00:09:32
birthday messages and light up your group chat that's good for
00:09:35
you um and for twenty nine dollars you can get a
00:09:39
lifetime membership and they advertise right here that there are no watermarks
00:09:43
on whatever it is that you produce um so for me this
00:09:46
is a company behaving very badly uh and this is quite dangerous
00:09:51
and it's already happening in terms of the presidential primary that i'm just happened in
00:09:56
new hampshire there was a rubble call abide in the fake mobile call that discourage people
00:10:02
from going to the polls and boring saying that they should say their vote for november which is completely
00:10:09
a relevant um and not true and as a result uh
00:10:13
on the eighth february the federal commute to um communications commission
00:10:16
i lot rubble calls using a i generated voices and voice
00:10:20
cloning technology um but i don't believe that's going to stop anybody
00:10:24
from engaging in this bad behaviour so the second i enable
00:10:30
threat to democracy that i want to talk about is destructive misleading
00:10:33
fall strokes um and unfortunately if you have some programming not knowledge
00:10:40
as an independent program where you could quite easily create your own
00:10:44
um customised chat bought or model that would allow for
00:10:47
the rapid creation and distribution of destructive troops on social media
00:10:52
so david reside oh oh it's a research scientist in new zealand and he put his skills
00:10:58
to good use doing an experiment and he created right wing g.
00:11:01
p. t. and and you can see from these two examples here um
00:11:07
human prompt concerns about climate change exaggerated and right when g.
00:11:11
p. t. v. i'm just as we expected would set the
00:11:14
cost of addressing climate change maybe too high relative to the
00:11:17
potential benefits and that they're more pressing issues that should take priority
00:11:21
we're should the government by free health care to all citizens right when g. b. t.
00:11:25
no it could lead to longer wait times for medical care blah blah blah blah blah
00:11:29
um and so these responses are channels there you know what you
00:11:34
would expect from white right wing g. p. t. on there also highly
00:11:38
misleading such false um and the problem is that once people get exposed
00:11:44
to this kind of material online it's called them your exposure of back
00:11:48
um the more we're exposed to something the more that we like it
00:11:51
and the more of the word um if it's repeated over and over the
00:11:56
more likely we are to believe that it's actually true um and this is
00:12:01
highly problematic for the democratic process and also just uh just general democratic debate
00:12:07
um between people about the issues the policy issues
00:12:11
that we currently face so for political psychology is perspective
00:12:16
now i wanna go into while deep takes and destructive
00:12:19
troops are so dangerous it's a really it okay um
00:12:31
'kay political scientist hard time sticking to the t. i.
00:12:35
okay um alright someone reasoning um will i this is based
00:12:43
on how cognition hypothesis that all socio political concepts are
00:12:46
associated with positive and negative aspects in our long term memory
00:12:50
and so the moment if you're having a conversation with a friend the moment you
00:12:53
realise that the friend is talking about are george w. bush and not a bunch
00:12:58
there is um an emotional response that you have in your mind um and and this
00:13:04
concept aspect combo is use than as a heuristic to evaluate new things that you come across
00:13:10
so if we look at this example um i don't know if you remember the controversy surrounding
00:13:16
the operations crowd at donald trumps integration but he
00:13:20
claimed that it was the biggest inauguration crowd ever
00:13:24
uh so they did a study on and i asked people to differentiate in these two pictures
00:13:29
which picture shows more people and three percent of
00:13:33
non voters chose the wrong picture so measurement error
00:13:36
fifteen percent of tram voters chose the wrong picture um so they are
00:13:42
mostly biased and unable to take
00:13:46
um unambiguous edited to make a decision
00:13:52
okay the next one has to do with emotional amplification on social
00:13:56
media um when you're also some yeah people are more likely to express
00:14:01
strong feelings as a signalling attempt um may wanna show that there
00:14:07
are there are members of a certain group they wanna influence other people
00:14:11
then people tend to uh miss perceive how much how angry the person who's posting really is
00:14:18
and this basically creates um a vicious cycle of emotional amplification on social
00:14:23
media 'cause enable by social media um from business models of social media
00:14:29
ah rely on engagement and the more emotion all
00:14:33
the posts are on social media the morning agent
00:14:37
um that you have from users so it makes
00:14:39
sense that their algorithms would promote i'm very emotive content
00:14:45
okay so what can be done um since i probably have zero minutes left um i'm just going okay
00:14:53
i'm just going to briefly go through this um i have some ideas about how we might ameliorate some of uh
00:14:59
attenuate some of these issues one is to have more
00:15:03
human input in the data annotation project up process and here
00:15:08
i want to just highlight the intricate project which is a joint project that we did from the graduate institute an area
00:15:15
and we were looking at how journalists interpret information that they come in contact with every day
00:15:21
and says you can't look into journalist brain while they're interpreting something we decided to look
00:15:27
at the articles that they wrote about say a white house press briefing but they tended
00:15:33
for example um and we came up with this structured way to
00:15:37
identifier service the parts of the sentence that we were interested in
00:15:42
namely based on the five w.'s of journalism um
00:15:46
what where when why so what happened who did it
00:15:50
um and what is the documents that the journalist is likely to the looking at when they wrote this article
00:15:56
and and if we don't train the model to try
00:16:01
to anticipate or predict what um the journalist would say or
00:16:07
write when given a fresh reference document um and we
00:16:11
found that assisted annotation um we're the annotator is actually highlighting
00:16:19
um the action implications of the act so here you say that the
00:16:24
um and the officials indicated here in favour of
00:16:28
ramping up the pace of interest rate increases this
00:16:30
year um we highlighted the the action implication of
00:16:35
this is they should raise interest rates faster right
00:16:40
um you okay i'm so my idea is that
00:16:45
in terms of dealing with um things like conspiracy
00:16:49
theories for example you could um involve humans in
00:16:55
the annotation process of information a conspiracy theory type information
00:17:01
and you could surface the action implications
00:17:06
instead of just the basic features of the patterns of words you could actually have human annotator is highlight something
00:17:12
that goes even deeper than what's on the page um okay i'm gonna
00:17:18
go through this really quick luckily someone is gonna be talking about um lies
00:17:22
this afternoon um we could of course create new laws to try to govern how generate of a
00:17:27
i can be used and hold b. tech accountable for an accurate content there are some issues here um
00:17:34
this example here is about um the tobacco free alternative
00:17:39
calls in its indicating pouch number one so nicotine house that people can put in their mouths um in the us
00:17:45
and the problem is it's being marketed to teenagers and
00:17:48
children on socially yeah buy into answers and simple answers
00:17:52
except it's illegal to market nicotine products to anyone under the age of twenty one in all fifty states
00:17:59
so there are lots i'm already on the books they've been on the books for some time but
00:18:04
we're having difficulty in forcing them because of the way that they're the loophole that influences have found
00:18:11
finally with the last slide um i think that we should
00:18:15
combat this information and misinformation through education right we can teach kids
00:18:20
um and teens to code would give them the tools to understand how
00:18:24
large language models work remove some of the mystery of this whole technology
00:18:30
we should also teach kids and teens how social media works um there should be some mandatory
00:18:35
social media literacy courses so that people understand that this
00:18:39
is the business model social media based on their engagement
00:18:43
um and that there are really no guarantees that any of the stuff that you see is accurate or reflective of real life
00:18:49
um and finally i think that we should we should take responsibility um
00:18:54
it's important that public intellectuals more technically trained come forward and explain the dangers
00:19:00
um that they see the pitfalls of generated a i i'm in the public space because the more
00:19:06
open to constructive conversations about this new technology and
00:19:11
that we can have there are solutions uh will be
