We thought for a long time that neurons were fairly easy to imitate. After all, most of what they do seems to be exchanging neurotransmitters, whereas internally they have an electric potential which is encoding most of their behaviour. While not easy, individually they are also not super duper complex compared to say, a microprocessor. Taken from this perspective, it looked like the main difficulty was in simulating them in large enough quantities to obtain interesting behaviour.
The argument in the paragraph above is foundational to how we think about intelligence. There are various other aspects to it, like morphological intelligence, but the above seems the most relevant when we want to create an artificial intelligence. While these days there are also more engineering-focused arguments for why we think artificial neural networks are interesting for research, underlying there is the idea that the only intelligence we know is made out of small simple things in large quantities that exchange simple signals. This is a key thought when you want to use artificial neural networks to create artificial intelligence. For instance the Human Brain Project in the EU (cost ~1 Billion EUR) is kind of built on this argument.
Now, and this is very recent research, this foundation is starting to shake from the biology side. It appears that neurons are also exchanging RNA, trans-scripting the RNA received into proteins and that those proteins seem to interfere with the transmission of RNA by the neuron [0]!
That is something completely different from the argument I poned before! In a computing analogy, every single neuron is a computer which is receiving and transmitting program code (RNA) and executing code it receives (transscription) and we know that the program it receives is interfering with the neurons receiving and transmission of programs to other neurons.
If intelligence does require such complex mechanisms, we are VERY far away from simulating (or understanding) that at scale.
Now we don't know whether this complexity is necessary. But the old argument of an upperbound to the level of complexity needed seems to be crumbling too.
All in all, putting some real neurons might prove to be a viable alternative for creating artificial intelligence, as it might circumvent the problem I sketched above. It might not, but it at least seems to be a direction worth exploring as a second bet next to simulating neurons.
[0] https://www.cell.com/action/showPdf?pii=S0092-8674%2817%2931...