Doubts about AGI aren’t about AI's intelligence
As everyday AI models continue to improve in performance, an increasing number of people are ready to believe that the development of AGI is imminent. In a recent survey of 500 people from the US and UK, we saw that the more often a respondent personally used an AI chatbot the more likely they were to believe in AGI - whereas just following news at an abstract level made little difference.
Overall, 84% believed that AGI would be developed - with around a quarter (27%) saying it would definitely happen. Around two in five (40%) thought that this would take place within a decade, the most commonly chosen option. By contrast, only 14% were outright sceptics of the possibility of AGI.
If AGI is ever developed, when do you think that's most likely to happen? (all respondents)
When we asked why people believed in the possibility of AGI, we saw that the most commonly chosen reasons were more practical than theoretical: they agreed that recent AI progress has been surprisingly fast (79%), computing power was likely to keep growing (72%) and a huge amount of money and talent had been invested (71%).
Which, if any, do you find convincing as reasons AGI might be developed one day? (% selecting)
But what about the sceptics? Why don’t they believe?
In our data, we saw that it was less about doubts about their potential cognitive abilities, and more about more metaphysical concerns around emotions and sentience. When asked in isolation, we saw that there actually was relative consensus across believers and sceptics that AI could feel emotions, with 79% overall saying that a machine could never truly experience emotion.
However, this was much more likely to be causal for sceptics: 76% of sceptics endorsed "machines could never truly feel emotions" as a reason AGI won't happen 29 points above believers.
Reasons AGI might never be developed, % selecting each (multi-select)
When we crossed two questions - whether they believed machines could feel, and whether they thought this was essential for intelligence - you can see that its in the people who both doubt machine sentience and think it essential to intelligence where AGI belief is lowest.
This came out even more clearly in the open text responses. Unprompted, sceptics were over six times more likely to spontaneously bring up emotions or feeling when asked why they AGI would or would not be developed.
% spontaneously mentioning emotion, feeling or empathy in open-text answers, by AGI belief
"It is still a probability engine… it can study every race ever run… but it can still only make a prediction. It has no gut instinct… [a human] is capable of stepping outside of logic."UK · Definitely won't
"How do you physically create a thought? Till then, AGI will never happen. We can get close. And mimic human behavior. But true thinking conscious AI is not possible."US · Definitely won't
"Because AGI would lack human empathy and feeling."UK · Definitely won't
"It can sound like it can feel things, but it is not true feeling. And a lot of people make decisions based on feelings."US · Probably won't
"Even intellectual tasks require a degree of emotional intelligence and instinct, and an AI system cannot react emotionally."UK · Probably won't
"I doubt AI will ever be able to replicate the flaws and quirks that make the human soul special."UK · Definitely won't
Whether LLMs can feel emotion or not is a genuinely open question - as, of course, is the hard problem of consciousness full stop. But for many people it looks like they remain crucial to their judgement on whether an AI is a true intelligence or not.
