Nearly half (45%) of the respondents are aware of generative AI, as with most new technologies, awareness skews higher among the younger, those on higher incomes and males. Yet, generative AI tools are far from being mainstream. ChatGPT is the most popular with 40% of the respondents aware of it. Snapchat’s My AI ranks second with only 8% awareness. Those newer tools like ChatGPT, Bard and Firefly offer a familiar-looking, easy to use interface, and do not require expert knowledge to use, or are directly integrated in existing Apps. Sign-up takes minutes and most tools do not (yet) require payment, despite significant underlying costs.

A large proportion of users have only dabbled with generative AI, and dropped it after a few tries.

Most people who have used generative AI are infrequent users, with only one in five (20%) using it weekly, and one in twenty (5%) using it daily. Interestingly, 32% have only used it “once or twice to try it” which is an indication that the tool remains nascent and is not yet of utility to all users.

Generative AI is nascent, fallible and there are still questions about ethics. Nearly a third of Belgian users (29%) claim to have used generative AI for work and educational purposes. Most likely, much of this use whilst well-intentioned and unsanctioned, may have taken place without adequate governance, in companies which might not yet have formulated an internal policy on usage yet. The majority of users (71%) have tried the technology for personal purposes.

As mentioned above, in our survey, respondents had used many types of Generative AI, but were most familiar with Large Language Models (LLM), such as GPT-4. LLMs work by estimating the probability that words will appear in a certain sequence. However, these models can make the wrong prediction, an output also known as ‘hallucination’. LLMs can serve-up sentences which are both factually incorrect and plausible to some readers. Users may lack the context or knowledge to be able to readily identify an impeccably presented (in terms of format, spelling and grammar) but wholly inaccurate response.

A significant proportion of respondents were unaware of the potential inaccuracies of generative AI tools, highlighting poor AI literacy. Of those who have used it, one third believe that it always produces factually accurate answers and a similar proportion (32%) believe that answers generated are unbiased.

Yet, they still see a need for regulating the AI market. Fifty-one percent do not think that there is adequate protection in place to control what content can be used to generate outputs by AI models. Only 15% believe that the current controls are sufficient, predominantly represented by the segment of 18–24 year-olds.

As with most emerging technologies, there are fears among the public about the impact on jobs. If, for example, technology can automate the creation of legal contracts or the compilation of software code, will a country’s workforce require fewer lawyers and developers?

Almost 41% of the respondents are concerned that generative AI could reduce the number of jobs in the future. Moreover, one third are convinced that their workplace would allow them to use it to complete daily tasks.