The problem is that our so-called 'modern' CPUs can only do exactly what they're programmed to do. Yes, they can perform incredibly complex calculations, such as pattern search and recognition, many orders of magnitude faster than humans. But that's not really the same kind of 'intelligence' that we can clearly see in the behavior of living creatures.
The behavior of CPUs is deterministic, i.e. tend to produce the same output, for a given set of inputs. Biological creatures, OTOH, tend to behave non-deterministically, that is their behaviors, given identical inputs, tend to produce varying sets of output behaviors, with ranges of variances that are difficult to predict.
Nature itself is only partially predictable. (We like to call the part we can't predict "noise".) So the behavior of electro-mechanical robots is very noisy because robots must process noisy sensor data using deterministic methods. Their programs merely react to input, so the humans who write their programs must somehow 'teach' them how to anticipate and react to all possible input scenarios. Which of course is computationally intractable, even for a planet-sized digital computer.
So, mathematically, robots are modeled as servomechanisms, which can operate automatically (more or less) by measuring responses received on their sensors and applying a kind of negative-feedback to reduce the variance of possible behaviors caused by 'recognizing noise' in a non-deterministic world.
We living creatures are much better at this kind of 'automatic behavior' because we are intrinsically non-deterministic machines, whose behaviors don't always 'make sense', but get us, sometimes, to some desired goals, more effectively and efficiently than simple 'random' behavior.
I believe there is a 'Life Principle', which is not yet fully understood, that makes this possible, by imbuing living creatures with mechanisms for consciousness (self-awareness) and motivation (desires and fears). So living creatures tend to have real-time 'situational awareness', which allows them, in effect, to connect to reality and understand and react to the world in terms of their own fears and desires. More or less.
Humans seem to have a lot of this kind of intelligence. Bumblebees not as much. But even the humblest earthworm seems to perceive a buzz of reality which helps them find the dark moist places they love (and avoid the dry, prickly places they fear). Digital computers perform more poorly in these simple reality tasks. (But can compute Pi to a million places easily!)
Will humans be able to build robots with this same kind of Life Principle? I think so, but first we have to study biological life more and actually figure out how it works, up to understanding how consciousness, fear, love and hate operate at a microscopic level.
Currently we can't even draw the simplest circuit diagram for 'consciousness', or build any simple device that 'understands' reality like we (think) we do.