This study addresses public perception of the relationship between artificial intelligence (AI) and disinformation. The level of general awareness of AI is considered, and based on this, an analysis is carried out of whether it may favor the creation and distribution of false content or, conversely, the public perceive its potential to counteract information disorders. A survey has been conducted on a representative sample of the Andalusian population aged 15 and over (1550 people). The results show that over 90% of the population have heard of AI, although it is less well known among the eldest age group (78%). There is a consensus that AI helps to produce (86%) and distribute (84%) fake news. Descriptive analyses show no major differences by sex, age, social class, ideology, type of activity or size of municipality, although those less educated tend to mention these negative effects to a lesser extent. However, 54% of the population consider that it may help in combating hoaxes, with women, the lower class and the left wing having positive views. Logistic regressions broadly confirm these results, showing that education, ideology and social class are the most relevant factors when explaining opinions about the role of AI in disinformation.
I'm not a machine learning expert. But I think it's just that we haven't yet learned how to do it yet. It's not a technical matter, or a question where to put it. But more that science has to figure out a lot of things. It'd be massively useful to guide these things. To control whether they hallucinate or tell the truth. To make them just do customer support based on factual information instead of also engaging in intimate cinversations. To strip bias and stereotypes. To make them "safe". But if you look at these systems in practice, you'll see they often fail. And then someone writes a news article every few weeks. And it happens to all current AI systems, even the market leaders. So I figure science just can't do it yet and we're in the early stages. Nobody knows at this point where to put it so a company could trust AI to act exactly in their interest. We might be able to do so, but that's still science fiction undtil we arrive at Skynet.
They can do some limited things with confined machine learning algorithms until then. Like Youtube recommendations, control your heating thermostats... I'm pretty sure these systems are small enough to have them work reliably to some specification. But LLMs are just very complex things. And a lot of training there happens unsupervised. So they might pick up arbitrary things which are hard to control.