2023人工智能信任全球洞察（英）-KPMG-2023.05月

Trust

inArtificialIntelligenceA

global

study2023.auKPMG.com.auCitationGillespie,

N.,

Lockey,

S.,

Curtis,

C.,

Pool,

J.,

&Akbari,

A.

(2023).

Trust

in

Artificial

Intelligence:A

Global

Study.The

University

of

Queenslandand

KPMG

Australia.

doi:10.14264/00d3c94University

of

Queensland

ResearchersProfessor

Nicole

Gillespie,

Dr

Steve

Lockey,Dr

Caitlin

Curtis

and

Dr

Javad

Pool.The

University

of

Queensland

team

led

thedesign,

conduct,

analysis

and

reporting

ofthis

research.KPMG

AdvisorsJames

Mabbott,

Rita

Fentener

vanVlissingen,JessicaWyndham,

and

Richard

Boele.AcknowledgementsWe

are

grateful

for

the

insightful

input,

expertiseand

feedback

on

this

research

provided

byDr

Ali

Akbari,

Dr

Ian

Opperman,

Rossana

Bianchi,Professor

Shazia

Sadiq,

Mike

Richmond,

andDr

Morteza

Namvar,

and

members

of

theTrust,

Ethics

and

Governance

Alliance

at

TheUniversity

of

Queensland,

particularly

Dr

NatalieSmith,

Associate

Professor

Martin

Edwards,Dr

Shannon

Colville

and

Alex

Macdade.FundingThis

research

was

supported

by

an

AustralianGovernment

Research

Support

Package

grantprovided

toThe

University

of

Queensland

AICollaboratory,

and

by

the

KPMG

Chair

in

Trustgrant

(ID

2018001776).Acknowledgement

of

CountryThe

University

of

Queensland

(UQ)acknowledges

theTraditional

Owners

and

theircustodianship

of

the

lands.We

pay

our

respectsto

their

Ancestors

and

their

descendants,

whocontinue

cultural

and

spiritual

connectionsto

Country.

We

recognise

their

valuablecontributions

to

Australian

and

global

society.©

2023

The

University

of

Queensland

ABN:63

942

912

684

CRICOS

Provider

No:00025B.©2023

KPMG,

an

Australian

partnership

and

a

member

firm

of

the

KPMG

global

organisation

of

independent

member

firms

affiliated

with

KPMG

International

Limited,

a

private

English

companylimited

by

guarantee.

All

rights

reserved.

The

KPMG

name

and

logo

are

trademarks

used

under

license

by

the

independent

member

firms

of

the

KPMG

global

organisation.Liability

limited

by

a

scheme

approved

under

Professional

Standards

Legislation.ContentsExecutive

summaryIntroduction020708How

we

conducted

the

research1.

To

what

extent

do

people

trust

AI

systems?2.

How

do

people

perceive

the

benefits

and

risks

of

AI?3.

Who

is

trusted

to

develop,

use

and

govern

AI?1122294.

What

do

people

expect

of

the

management,

governanceand

regulation

of

AI?34435360665.

How

do

people

feel

about

AI

at

work?6.

How

well

do

people

understand

AI?7.

What

are

the

key

drivers

of

trust

in

and

acceptance

of

AI?8.

How

have

trust

and

attitudes

towards

AI

changed

over

time?Conclusion

and

implications70737577Appendix

1:

Method

and

statistical

notesAppendix

2:

Country

samplesAppendix

3:

Key

indicators

for

each

country©

2023

The

University

of

Queensland

ABN:63

942

912

684

CRICOS

Provider

No:00025B.©2023

KPMG,

an

Australian

partnership

and

a

member

firm

of

the

KPMG

global

organisation

of

independent

member

firms

affiliated

with

KPMG

International

Limited,

a

private

English

companylimited

by

guarantee.

All

rights

reserved.

The

KPMG

name

and

logo

are

trademarks

used

under

license

by

the

independent

member

firms

of

the

KPMG

global

organisation.Liability

limited

by

a

scheme

approved

under

Professional

Standards

Legislation.TRUSTIN

ARTIFICIALINTELLIGENCE2ExecutivesummaryArtificial

Intelligence

(AI)

has

become

a

ubiquitous

part

of

everyday

life

and

work.AI

is

enabling

rapid

innovation

that

is

transforming

the

way

work

is

done

andhow

services

are

delivered.

For

example,

generative

AI

tools

such

as

ChatGPTare

having

a

profound

impact.

Given

the

many

potential

and

realised

benefits

forpeople,

organisations

and

society,

investment

in

AI

continues

to

grow

across

allsectors1,

with

organisations

leveraging

AI

capabilities

to

improve

predictions,optimise

products

and

services,

augment

innovation,

enhance

productivity

andefficiency,

and

lower

costs,

amongst

other

beneficial

applications.However,

the

use

of

AI

also

poses

risks

and

challenges,

raising

concerns

aboutwhether

AI

systems

(inclusive

of

data,

algorithms

and

applications)

are

worthyof

trust.These

concerns

have

been

fuelled

by

high

profile

cases

of

AI

usethat

were

biased,

discriminatory,

manipulative,

unlawful,

or

violated

humanrights.

Realising

the

benefits

AI

offers

and

the

return

on

investment

in

thesetechnologies

requires

maintaining

the

public’s

trust:

people

need

to

be

confidentAI

is

being

developed

and

used

in

a

responsible

and

trustworthy

manner.Sustained

acceptance

and

adoption

of

AI

in

society

are

founded

on

this

trust.This

research

is

the

first

to

take

a

deep

dive

examination

into

the

public’s

trustand

attitudes

towards

the

use

of

AI,

and

expectations

of

the

management

andgovernance

of

AI

across

the

globe.Wesurveyed

over

17,000

people

from

17

countries

covering

all

global

regions:Australia,

Brazil,

Canada,

China,

Estonia,

Finland,

France,

Germany,

India,

Israel,Japan,

the

Netherlands,

Singapore,

South

Africa,

South

Korea,

the

United

Kingdom(UK),

and

the

United

States

of

America

(USA).These

countries

are

leaders

inAI

activity

and

readiness

within

their

region.

Each

country

sample

is

nationallyrepresentative

of

the

population

based

on

age,

gender,

and

regional

distribution.Weasked

survey

respondents

about

trust

and

attitudes

towards

AI

systems

ingeneral,

as

well

as

AI

use

in

the

context

of

four

application

domains

where

AI

israpidly

being

deployed

and

likely

to

impact

many

people:

in

healthcare,

public

safetyand

security,

human

resources

and

consumer

recommender

applications.The

research

provides

comprehensive,

timely,

global

insights

into

the

public’strust

and

acceptance

of

AI

systems,

including

who

is

trusted

to

develop,use

and

govern

AI,

the

perceived

benefits

and

risks

of

AI

use,

communityexpectations

of

the

development,

regulation

and

governance

of

AI,

and

howorganisations

can

support

trust

in

their

AI

use.

It

also

sheds

light

on

how

peoplefeel

about

the

use

of

AI

at

work,

current

understanding

and

awareness

of

AI,and

the

key

drivers

of

trust

in

AI

systems.

We

also

explore

changes

in

trust

andattitudes

to

AI

over

time.Next,

we

summarise

the

key

findings.©

2023

The

University

of

Queensland

ABN:63

942

912

684

CRICOS

Provider

No:00025B.©2023

KPMG,

an

Australian

partnership

and

a

member

firm

of

the

KPMG

global

organisation

of

independent

member

firms

affiliated

with

KPMG

International

Limited,

a

private

English

companylimited

by

guarantee.

All

rights

reserved.

The

KPMG

name

and

logo

are

trademarks

used

under

license

by

the

independent

member

firms

of

the

KPMG

global

organisation.Liability

limited

by

a

scheme

approved

under

Professional

Standards

Legislation.TRUSTIN

ARTIFICIALINTELLIGENCE3Most

people

are

wary

about

trusting

AI

systems

andhave

low

or

moderate

acceptance

of

AI:

however,trust

and

acceptance

depend

on

the

AI

applicationPeople

perceive

the

risks

of

AI

in

asimilar

wayacross

countries,

with

cybersecurity

rated

as

thetop

risk

globallyAcross

countries,

three

out

of

five

people

(61%)

are

waryabout

trusting

AI

systems,

reporting

either

ambivalence

oran

unwillingness

to

trust.Trust

is

particularly

low

in

Finlandand

Japan,

where

less

than

a

quarter

of

people

report

trustingAI.

In

contrast,

people

in

the

emerging

economies

of

Brazil,While

there

are

differences

in

how

the

AI

benefit-riskratio

is

viewed,

there

is

considerable

consistency

acrosscountries

in

the

way

the

risks

of

AI

are

perceived.Just

under

three-quarters

(73%)

of

people

across

the

globereport

feeling

concerned

about

the

potential

risks

of

AI.These

risks

include

cybersecurity

and

privacy

breaches,manipulation

and

harmful

use,

loss

of

jobs

and

deskilling,system

failure,

the

erosion

of

human

rights,

and

inaccurateor

biased

outcomes.India,

China

and

South

Africa

(BICS

)

have

the

highest

levels2of

trust,

with

the

majority

of

people

trusting

AI

systems.People

have

more

faith

in

AI

systems

to

produce

accurateand

reliable

output

and

provide

helpful

services,

and

aremore

sceptical

about

the

safety,

security

and

fairness

of

AIsystems

and

the

extent

to

which

they

uphold

privacy

rights.In

all

countries,

people

rated

cybersecurity

risks

as

theirtop

one

or

two

concerns,

and

bias

as

the

lowest

concern.Job

loss

due

to

automation

is

also

a

top

concern

in

Indiaand

South

Africa,

and

system

failure

ranks

as

a

top

concernin

Japan,

potentially

reflecting

their

relative

heavydependence

on

smart

technology.Trust

in

AI

systems

is

contextual

and

depends

on

thespecific

application

or

use

case.

Of

the

applicationswe

examined,

people

are

generally

less

trusting

andaccepting

of

AI

use

in

human

resources

(i.e.

for

aidinghiring

and

promotion

decisions),

and

more

trusting

ofAI

use

in

healthcare

(i.e.

for

aiding

medical

diagnosisand

treatment)

where

there

is

a

direct

benefit

to

them.People

are

generally

more

willing

to

rely

on,

than

shareinformation

with

AI

systems,

particularly

recommendersystems

(i.e.

for

personalising

news,

social

media,

andproduct

recommendations)

and

security

applications(i.e.

for

aiding

public

safety

and

security

decisions).These

findings

reinforce

the

critical

importance

of

protectingpeople’s

data

and

privacy

to

secure

and

preserve

trust,

andsupporting

global

approaches

and

international

standardsfor

managing

and

mitigating

AI

risks

across

countries.There

is

strong

global

endorsement

for

the

principlesof

trustworthy

AI:

trust

is

contingent

on

upholdingand

assuring

these

principles

are

in

placeMany

people

feel

ambivalent

about

the

use

of

AI,Our

findings

reveal

strong

global

public

support

for

theprinciples

and

related

practices

organisations

deployingAI

systems

are

expected

to

uphold

in

order

to

be

trusted.Each

of

theTrustworthy

AI

principles

originally

proposed

byreporting

optimism

or

excitement

on

the

one

hand,

whilesimultaneously

reporting

worry

or

fear.

Overall,

two-thirdsof

people

feel

optimistic

about

the

use

of

AI,

while

abouthalf

feel

worried.While

optimism

and

excitement

aredominant

emotions

in

many

countries,

particularly

the

BICScountries,

fear

and

worry

are

dominant

emotions

for

peoplein

Australia,

Canada,

France,

and

Japan,

with

people

inFrance

the

most

fearful,

worried,

and

outraged

about

AI.the

European

Commission

are

viewed

as

highly

important3for

trust

across

all

17

countries,

with

data

privacy,

securityand

governance

viewed

as

most

important

in

all

countries.This

demonstrates

that

people

expect

organisationsdeploying

AI

systems

to

uphold

high

standards

of:People

recognise

the

many

benefits

of

AI,

but

onlyhalf

believe

the

benefits

outweigh

the

risks–

data

privacy,

security

and

governance–

technical

performance,

accuracy

and

robustness–

fairness,

non-discrimination

and

diversity–

human

agency

and

oversightPeople’s

wariness

and

ambivalence

towards

AI

can

be

partlyexplained

by

their

mixed

views

of

the

benefits

and

risks.Most

people

(85%)

believe

AI

results

in

a

range

of

benefits,and

think

that

‘process’

benefits

such

as

improved

efficiency,innovation,

effectiveness,

resource

utilisation

and

reducedcosts,

are

greater

than

the

‘people’

benefits

of

enhancingdecision-making

and

improving

outcomes

for

people.–

transparency

and

explainability–

accountability

and

contestability–

risk

and

impact

mitigation–

AI

literacy

supportHowever,

on

average,

only

one

in

two

people

believe

thebenefits

of

AI

outweigh

the

risks.

People

in

the

westerncountries

and

Japan

are

particularly

unconvinced

that

thebenefits

outweigh

the

risks.

In

contrast,

the

majority

ofpeople

in

the

BICS

countries

and

Singapore

believe

thebenefits

outweigh

the

risks.©

2023

The

University

of

Queensland

ABN:63

942

912

684

CRICOS

Provider

No:00025B..©2023

KPMG,

an

Australian

partnership

and

a

member

firm

of

the

KPMG

global

organisation

of

independent

member

firms

affiliated

with

KPMG

International

Limited,

a

private

English

companylimited

by

guarantee.

All

rights

reserved.

The

KPMG

name

and

logo

are

trademarks

used

under

license

by

the

independent

member

firms

of

the

KPMG

global

organisation.Liability

limited

by

a

scheme

approved

under

Professional

Standards

Legislation.TRUSTIN

ARTIFICIALINTELLIGENCE4People

expect

these

principles

to

be

in

place

for

each

of

theAI

use

applications

we

examined

(e.g.,

Human

Resources,Healthcare,

Security,

Recommender,

and

AI

systems

ingeneral),

suggesting

their

universal

application.This

strongpublic

endorsement

provides

a

blueprint

for

developing

andusing

AI

in

a

way

that

supports

trust

across

the

globe.People

expect

AI

to

be

regulated

with

some

form

ofexternal,

independent

oversight,

but

view

currentregulations

and

safeguards

as

inadequateThe

large

majority

of

people

(71%)

expect

AI

to

beregulated.With

the

exception

of

India,

the

majority

in

allother

countries

see

regulation

as

necessary.This

findingOrganisations

can

directly

build

trust

and

consumerwillingness

to

use

AI

systems

by

supporting

andimplementing

assurance

mechanisms

that

help

peoplefeel

confident

these

principles

are

being

upheld.Three

outof

four

people

would

be

more

willing

to

trust

an

AI

systemwhen

assurance

mechanisms

are

in

place

that

signalethical

and

responsible

use,

such

as

monitoring

systemaccuracy

and

reliability,

independent

AI

ethics

reviews,

AIethics

certifications,

adhering

to

standards,

and

AI

codesof

conduct.These

mechanisms

are

particularly

importantgiven

the

current

reliance

on

industry

regulation

andgovernance

in

many

jurisdictions.corroborates

prior

surveys

indicating

strong

desire

for4regulation

of

AI

and

is

not

surprising

given

most

people(61%)

believe

the

long-term

impact

of

AI

on

society

isuncertain

and

unpredictable.People

are

broadly

supportive

of

multiple

forms

of

regulation,including

regulation

by

government

and

existing

regulators,a

dedicated

independent

AI

regulator,

and

co-regulation

andindustry

regulation,

with

general

agreement

of

the

need

forsome

form

of

external,

independent

oversight.Despite

the

strong

expectations

of

AI

regulation,

only

two

infive

people

believe

current

regulations,

laws

and

safeguardsare

sufficient

to

make

AI

use

safe.This

aligns

with

previousPeople

are

most

confident

in

universities

and

defenceorganisations

to

develop,

use

and

govern

AI

and

leastconfident

in

government

and

commercial

organisationssurveys

showing

public

dissatisfaction

with

the

regulation5of

AI,

and

is

problematic

given

the

strong

relationshipbetween

current

safeguards

and

trust

in

AI

demonstratedby

our

modelling.This

highlights

the

importance

ofstrengthening

and

communicating

the

regulatory

andlegal

framework

governing

AI

and

data

privacy.People

have

the

most

confidence

in

their

nationaluniversities

and

research

institutions,

as

well

as

theirdefence

organisations,

to

develop,

use

and

govern

AI

in

thebest

interest

of

the

public

(76–82%

confident).

In

contrast,they

have

the

least

confidence

in

governments

andcommercial

organisations

to

do

this.

A

third

of

people

lackconfidence

in

government

and

commercial

organisations

todevelop,

use

and

regulate

AI.This

is

problematic

given

theincreasing

scope

with

which

governments

and

commercialorganisations

are

using

AI,

and

the

public’s

expectationthat

these

entities

will

responsibly

govern

and

regulateits

use.

An

implication

is

that

government

and

businesscan

partner

with

more

trusted

entities

in

the

use

andgovernance

of

AI.There

are,

however,

substantial

country

differences,

withpeople

in

India

and

China

most

likely

to

believe

appropriatesafeguards

are

in

place

(74–80%

agree)

followed

by

Braziland

Singapore

(52–53%).

In

contrast,

people

in

Japanand

South

Korea

are

the

least

convinced

(13–17%

agree)as

are

the

majority

of

people

in

western

countries.Thesedifferences

in

the

perceived

adequacy

of

regulations

maypartly

explain

the

higher

trust

and

acceptance

of

AI

amongpeople

in

the

BICS

countries.Most

people

are

comfortable

with

the

use

of

AI

toaugment

work

and

inform

managerial

decision-making,

but

want

humans

to

retain

controlThere

are

significant

differences

across

countries

inpeople’s

trust

of

their

government

to

use

and

governAI,

with

about

half

of

people

lacking

confidence

in

theirgovernment

in

South

Africa,

Japan,

the

UK

and

the

USA,whereas

the

majority

in

China,

India

and

Singaporehave

high

confidence

in

their

government.This

patternmirrors

people’s

general

trust

in

their

governments:

wefound

a

strong

association

between

people’s

generaltrust

in

government,

commercial

organisations

and

otherinstitutions

and

their

confidence

in

these

entities

to

useand

govern

AI.These

findings

suggest

that

taking

actionto

strengthen

trust

in

institutions

generally

is

an

importantfoundation

for

trust

in

specific

AI

activities.Most

people

are

comfortable

with

the

use

of

AI

at

workto

augment

and

automate

tasks,

but

are

less

comfortablewhen

AI

is

focused

on

them

as

employees,

for

example

forHR

and

people

management

(e.g.

to

monitor

and

evaluateemployees,

and

support

recruitment).

On

average,

half

thepeople

are

willing

to

trust

AI

at

work,

for

example

by

relyingon

the

output

it

provides.

People

in

Australia,

Canada,France

and

Germany

are

the

least

comfortable

with

theuse

of

AI

at

work,

while

those

in

the

BICS

countries

andSingapore

are

the

most

comfortable.©

2023

The

University

of

Queensland

ABN:63

942

912

684

CRICOS

Provider

No:00025B.©2023

KPMG,

an

Australian

partnership

and

a

member

firm

of

the

KPMG

global

organisation

of

independent

member

firms

affiliated

with

KPMG

International

Limited,

a

private

English

companylimited

by

guarantee.

All

rights

reserved.

The

KPMG

name

and

logo

are

trademarks

used

under

license

by

the

independent

member

firms

of

the

KPMG

global

organisation.Liability

limited

by

a

scheme

approved

under

Professional

Standards

Legislation.TRUSTIN

ARTIFICIALINTELLIGENCE5Most

people

view

AI

use

in

managerial

decision-makingas

acceptable,

and

actually

prefer

AI

involvement

to

solehuman

decision-making.

However,

the

preferred

option

iseither

a

25%-75%

or

50%-50%

AI-human

collaboration,with

humans

retaining

more

or

equal

control.This

indicates

aclear

preference

for

AI

to

be

used

as

a

decision

aid,

and

a

lackof

support

for

fully

automated

AI

decision-making

at

work.They

have

greater

knowledge

of

AI

and

are

better

ableto

identify

when

AI

is

used,

and

have

greater

interest

inlearning

about

AI.They

perceive

more

benefits

of

AI,

butremain

the

same

as

other

groups

in

their

perceptions

ofthe

risks

of

AI.They

are

more

likely

to

believe

AI

will

createjobs,

but

also

more

aware

that

AI

can

perform

key

aspectsof

their

work.They

are

more

confident

in

entities

todevelop,

use

and

govern

AI,

and

more

likely

to

believe

thatcurrent

safeguards

are

sufficient

to

make

AI

use

safe.

It

isnoteworthy

that

we

see

very

few

meaningful

differencesacross

gender

in

trust

and

attitudes

towards

AI.While

about

half

believe

AI

will

enhance

their

competenceand

autonomy

at

work,

less

than

one

in

three

people

believeAI

will

create

more

jobs

than

it

will

eliminate.

However,most

managers

believe

the

opposite

–

that

AI

will

createjobs.This

reflects

a

broader

trend

of

managers

being

morecomfortable,

trusting

and

supportive

of

AI

use

at

work

thanother

employees,

with

manual

workers

the

least

comfortableand

trusting

of

AI

at

work.

Given

managers

are

typically

thedrivers

of

AI

adoption

in

organisations,

these

differing

viewsmay

cause

tensions

in

the

implementation

of

AI

at

work.There

are

stark

differences

in

trust

and

attitudesacross

countries:

people

in

the

emerging

economiesof

Brazil,

India,

China,

and

South

Africa

are

moretrusting

and

accepting

of

AI

and

have

more

positiveattitudes

towards

AIA

key

insight

from

the

survey

is

the

stark

differences

in

trust,attitudes

and

use

of

AI

between

people

in

the

emergingeconomies

of

Brazil,

India,

China

and

South

Africa

and

thosein

other

countries.A

minority

of

people

in

western

countries,

Japan

and

SouthKorea

report

that

their

employing

organisation

invests

in

AIadoption,

recognises

efforts

to

integrate

AI,

or

supports

theresponsible

use

of

AI.This

stands

in

contrast

to

a

majorityof

people

in

the

BICS

countries

and

Singapore.People

in

the

emerging

economies

are

more

trustingand

accepting

of

AI

and

hold

more

positive

feelings

andattitudes

towards

AI

than

people

in

other

countries.Thesedifferences

held

even

when

controlling

for

the

effects

of

ageand

education.

Singapore

followed

this

positive

orientationon

several

indicators,

particularly

comfort,

trust

andfamiliarity

with

the

use

of

AI

at

work,

adequacy

of

current

AIregulation

and

governance,

confidence

in

companies

to

useand

govern

AI,

and

the

belief

that

AI

will

create

jobs.People

want

to

learn

more

about

AI

but

currently

havelow

understandingWhile

82%

of

people

are

aware

of

AI,

one

in

two

peoplereport

feeling

they

do

not

understand

AI

or

when

and

howit

is

used.

Understanding

of

AI

is

highest

in

China,

India,South

Korea,

and

Singapore.Two

out

of

five

people

areunaware

that

AI

enables

common

applications

they

use.For

example,

even

though

87%

of

people

use

social

media,45%

do

not

know

AI

is

used

in

social

media.Our

data

suggests

that

this

high

trust

is

not

blind

to

therisks.

People

in

BICS

countries

and

Singapore

did

notperceive

the

risks

of

AI,

or

the

uncertain

impact

of

AI

onsociety,

any

lower

than

people

in

other

countrie