International AI safety
survey data tables
Details
The Centre for Data Ethics and
Innovation commissioned Deltapoll to conduct
international research on public opinion towards AI safety ahead
of the UK’s AI Safety Summit
2023. Respondents from the general public in nine countries –
Canada, France, Germany, Italy, Japan, Singapore, South Korea,
the UK and the USA – expressed widespread support for AI safety
testing.
Key findings:
- When asked how much they agree or disagree that powerful AI
should be tested by independent experts to ensure it is safe,
most respondents in all countries agreed, ranging from 59% in
Japan to 76% in the UK and Singapore.
- When asked if they would support a government-backed AI
safety institute evaluating powerful AI to test if it is safe,
most respondents in all countries agreed, ranging from 51% in
Japan to 65% in Italy, with 62% agreeing in the UK.
- Across all the countries surveyed, the most useful
applications of AI identified by respondents were ‘Using
surveillance footage to detect crime’ and ‘Diagnosing a patient’s
illness’.
- In nearly all countries surveyed, more than half of the
respondents were worried about the risk of AI being used to carry
out cyberattacks, AI being used to help design biological
weapons, and humans losing control of AI. In many instances the
proportions of respondents worried about these risks were
sizeable.
- There was broad agreement across respondents from all
countries surveyed that “mitigating the risk of extinction should
be a global priority alongside other societal-scale risks such as
pandemics and nuclear war.” Agreement ranged from 56% in South
Korea to 40% in Germany, but in no country surveyed is the
proportion of respondents disagreeing above 13% (one in seven).
- Opinion was divided over whether the potential risks of AI
outweigh the potential benefits. In Canada, France, Germany,
Japan, the UK and the US, the largest proportion of respondents
believed that AI has more risks than benefits. On the other hand,
in Italy, Singapore and South Korea the largest proportion of
respondents believed AI has more benefits than risks.
Methodology
Deltapoll interviewed 1,114 adult participants in Canada, 1,120
adult participants in France, 1,164 adult participants in
Germany, 1,136 adult participants in Italy, 1,137 adult
participants in Japan, 1,134 adult participants in Singapore,
1,142 adult participants in South Korea, 1,090 adult participants
in the United Kingdom and 1,126 adult participants in the United
States online between 9th and 13th October 2023. All data have
been weighted to be representative of that country’s adult
population as a whole.
