Benefits and concerns

The UK public has high expectations for some AI technologies. In particular, the majority of the public perceive facial recognition in policing (91%) and AI-driven risk assessment for cancer (86%) to be beneficial uses of AI. A majority of the public (63%) also have positive views about LLMs (e.g. ChatGPT).

Expectations of positive impact are lower for other uses of AI, as shown in Figure 3. While optimism around the use of AI in diagnostics for cancer is high, the same is not true for the application of AI in other areas of healthcare, with only 36% of the public perceiving mental health chatbots to be beneficial, and 55% perceiving robotic care assistants to be beneficial. As with our previous survey wave, the only public sector application of AI in our current survey – the use of AI for assessing eligibility for welfare benefits – was also viewed less positively than others, with less than half (47%) of the public perceiving it as beneficial. Similarly, only 45% perceive driverless cars to be beneficial. 

Overall, the UK public are most concerned about the application of AI in driverless cars (75%), mental health chatbots (63%) and assessing welfare eligibility (59%). Figure 4 shows concern levels for each AI technology. Even for technologies where expected positive impact is high, concern levels are also substantial. For example, nearly two-fifths (39%) of the public are concerned by the use of facial recognition in policing. 

We observed key demographic differences in the perceived benefits of each AI technology. Black/​Black British and Asian/​Asian British people are more likely than the national average to view applications of robotics (driverless cars and robotic care assistants), LLMs and mental health chatbots as beneficial. The demographic difference in perceived benefits is most notable for general-purpose LLMs, where 80% of each minoritised ethnic group perceives them as beneficial, compared to 63% of the general population. 

People on lower incomes and those with fewer digital skills are less likely than the general public to perceive nearly all of the AI technologies we asked about as beneficial. For instance, only 48% of people on lower incomes felt robotic care assistants could be beneficial compared to 55% of the general population. Among those with fewer digital skills, 41% felt LLMs could be beneficial compared to 63% of the general population.

Figure 5 highlights demographic variances in overall levels of perceived benefit for each technology.

We also observed key demographic differences in perceptions of concerns. Over half of all Black (57%) and Asian (52%) people in our sample reported being fairly or very concerned about the use of facial recognition in policing, compared to 39% of the general public. The top three concerns they reported included: 1) the gathering of personal information which could be shared with third parties (for 59% of Black people and 60% of Asian people); 2) causing police to rely too heavily on technology rather than their professional judgment (for 66% of Black people and 58% of Asian people); and 3) the risk of innocent people being wrongly accused if the system makes mistakes (for 62% of Black people and 61% of Asian people). Black and Black British people are also more likely to report concerns around the use of AI to determine eligibility for welfare – 71% find this use of AI very or fairly concerning, compared to 59% of the nationally representative cohort. 

For some applications of AI, concern levels are lower among oversampled subgroups. Individuals belonging to low-income groups (42%) and those with low digital skills (38%) are significantly less concerned by general-purpose LLMs compared to the national average (47%). This is in line with their perception of the benefit of these technologies. Those with low digital skills (55%) are also less concerned by mental health chatbots compared to the nationally representative cohort (63%). And Asian and British Asian people (49%) are less concerned about robotic care assistants than the average (58%).

Figure 6 highlights demographic variances in overall levels of concern for each technology. 

The findings above highlight that attitudes to AI are multifaceted in several aspects. First, AI is not considered to be a single entity, with perceptions varying towards specific applications. For example, the public has positive attitudes towards general-purpose LLMs, with high overall benefit scores and low levels of reported concern, and more negative attitudes towards mental health chatbots. This shows how the context each technology is applied to matters. 

Second, people can simultaneously perceive benefits and risks associated with different applications of AI. For each of the applications of AI we asked about, people reported differential levels of both perceived benefit and concern that are not mirror images of each other. 

Third, minoritised demographic groups perceive differential levels of benefits and concerns for each AI application, as we explore above. For instance, people on lower incomes have higher levels of concern for many of the applications of AI we asked about compared to the national average. It is therefore important to consider the views of diverse publics when trying to understand public sentiment towards different applications of AI.

When comparing responses across both waves of our survey, perceptions of benefit have remained relatively stable (except in the case of facial recognition for policing, where perceptions of benefit have increased slightly) while levels of concern have significantly increased across all applications of AI. However, it is important to note that the comparisons should be read with caution as the sample composition across the two waves is different (refer to the discussion in the methodology section). 

Figures 7 and 8 show a comparison of perceptions of benefit and concern across both surveys for all repeated uses of AI. For example, in 2022/23, 44% of the public were concerned by the use of AI for determining welfare eligibility. This has increased to 59% in 2024/25.