1) If AI gets that good, then theoretically what do we need humans for any task even on earth? I really don't believe that will be the case. Even automated factories require stewardship and I think it will still be quite some time before humans in partnership with the machines won't provide a multiplying factor in the machines productivity. Bottom line, humans will be invaluable handling the myriads of problems and real world tradeoffs that won't be entirely anticipated by the machines programmers. Beyond the strategic, the humans will act as facilitators to increase the productivity.

2) Not having volatiles or mars resources directly available will be a small price to pay to not have to carry the rocket fuel for human landing and support initially. Anyone doing the math will find the cost lowered by probably and order of magnitude.

3) Initial sample return plans are entirely robotic anyway, so I don't see how you lose anything. Robotic craft can certainly rendezvous with an orbited space craft to provide that capability. As far as stepping on Mars, virtual reality can sometimes be better than reality (ask Ballard the famous deep sea explorer who has been wholly converted to robotic exploration). Ballard tells an anecdote of going with someone in a deep sea submersible and when they were at depth, he noticed the guy looking at the monitors more than the portal. This was because he got a better view through the monitors than he did the portal. Then Balllard realized, if that was the case then why even go down at all, just stay at the surface and look at a monitor.

4) My proposal does not replace human landing on Mars. I am just suggestion cost effective prepatory missions to eventual colonization. The teleoperation allows the rapid builiding and exploration that would enable true colonization in a shorter time frame for less cost. Serious colonization cannot occur until we find MUCH MUCH cheaper ways to move material out of our gravity well anyway (e.g. space elevator).