ECUs still follow maps and tables, and make certain assumptions about the environment. The learning can only go so far - basically some table values can be modified, but it's not true "learning". It's only as clever and the engineer who designed the tables to be modified.
Regardless, ECUs have some tremendous advantages over analog systems. Some examples:
I can't speak for all engines, but for at least 30 years Honda has had greater than 99% efficient combustion of fuel, suggesting there is no issue with fuel atomization to be solved.
In my engine, there are some areas where very clever intake and exhaust design have increased the volumetric efficiency to over 110%. An ECU can be programmed to retard timing and add extra fuel locally to these specific regions, something which cannot be done (to my knowledge) in an analog system which follows linear or exponential curves. Using a carburetor on my engine would result in needing to run very rich over areas of the curve to compensate for zones of 100%+ volumetric efficiency, and needing to run excessively retarded timing over large areas of the rev range so that it's not too advanced in areas of high VE. Fueling doesn't always follow vacuum or curves, and there's a lot of use of having sensors that can give combustion exactly what it needs to be most efficient in every scenario.
I'm sure carburetors could be used with some very basic implementations of things like variable valve timing and variable valve lift, but I haven't the faintest idea how one would compensate for changes that result from having multiple cam profiles and 50+ degrees of intake and/or exhaust cam phasing. These systems boost both engine efficiency and power.
In many modern engines, fuel is intentionally added late during certain parts of the load/RPM map to ensure it doesn't spread homogeneously throughout the charge in the cylinder, but instead is locally rich around the spark plug when it fires. This rich zone allows the flame front to spread more quickly, leading to increased combustion speed. When combustion happens more quickly, ignition timing can happen later, meaning less negative work is being done against the rising piston. This allows engines to be lighter, more powerful and more efficient. You can read about some of these things here:
https://jalopnik.com/heres-how-toyot...ere-1824090404
I have a car in my driveway which can cruise at greater than 100mpg on the highway in a way that is repeatable, measurable, and understood. Fuel is already burnt completely, and electronic engine controls have increased thermal efficiency of gasoline engines from the 20's as seen in carburetor days to greater than 40% in some cases. In a well designed engine, I could see uncounted, vaporized homogeneous fuel/air charges being added in an uncontrolled manor only reducing efficiency. This may not apply to older designs, granted, but I'm highly skeptical of this showing positive effects on anything made in the last 10 years.
~
So, all of that said, and as skeptical as I am, I would love to see this work. Do you have any data on what has worked and what hasn't? Any idea on how and why?