It is.
Most of these predictions are made by a SEIR epidemiology model. It's a differential equation that has many about 4 or 5 parameters that are difficult to measure. Because it's exponential small overestimates in the basic parameters can have huge consequences.
The worst offender is the COVID Act Now site which displays dire graphs but provides no information as to how they were obtained. Rumor has it they used information from the guy at Imperial College.
One good site is the University of Pennsylvania CHIME model. It's purpose is to try to help hospitals plan their predicted census and ICU census based on the doubling rate of cases they are seeing in real time.
You can play with it yourself. The problem with the model though is that it is using a fixed R0 of 2.6(referred to as R-naught which is loosely considered the birth number of new infections). While that is considered the current value for this virus it is still an estimate. No one really knows what the value of the number really is.
The model as presented at the site does not allow you to change the value of R0. So if you play with it, you will find that the exponential underpinnings of the model and its use of a fixed value of R0 mean that the hospital census quickly gets high. This inspite of changing the other parameters to try to effect a reduction.
Thanks!
Stig, I got what you said. I worked on the Club of Rome programs when I was in college as a class project, and they had a similar problem
Which is why I have never trusted computer models.