The doubts about CA's real-world effectiveness are actually well-represented in the media, as your example or [0] show. This actually closely follows the a similar news cycle right after the election. I distinctly remember reading several articles casting doubt on CA's effectiveness in The New York Times and The Atlantic.
I also don't think many reports actually assert that CA was responsible for, say, a 10% swing. Why people are taking this serious is, I believe, because we now know that it's possible to get such data and we have seen the sort of feats that machine learning can do today. It's doubtful that CA was the company to successfully combine those two ingredients. But without changes it's all but guaranteed that someone will soon will.
Plus there is the fact that the election just happened to be incredibly close. You can make a list with about ten possible reasons for the result, from people being tired of Hillary, to the FBI's strange moves regarding the e-mail investigation, third-party candidates, russian trolls, the free media Trump got, Cambridge Analytica, and probably the weather in eastern Ohio. It's possible that the absence of any two or three of these would have changed the outcome.
Lets imagine you developed a Deep Learning algorithm that proved beyond a shadow of a doubt to be able to pull off what CA pulled off. Why on god's green earth would you ever tell people that you have the algorithm, if you could instead use it for your own benefit? Why would you ever publish?
Answer A: in the case that you think you can publish at just the right time such that every political party would have enough time to master the tech before the next election cycle.
Answer B: in the case that you can charge exorbitantly for access to the published algorithm, and that - being a rational actor without nationalistic pride - you don't give two hoots and a holler what happens to the nation where you publish, so long as you can walk away with a fat paycheck.
The true power in microtargeting lies in the iteration speed achievable by computers. As with any effective weapon, if you can wield yours faster than your opponent, you will emerge victorious in the normal case. This is the same principle at play in High Frequency Trading. And just like High Frequency Trading, which was pioneered by the chairman of Cambridge Analytica, Robert Mercer, the less competition you have, the less entities that can usurp your temporal advantage. There is no "mastery" of this technique, only being able to do it faster than everyone else. Thus, there is no balance lent by proliferating the algorithm, as I see it.
I guess as an irrational actor, which we all are as humans, I would either seek one of two outcomes, neither of which resulted in giving away the algorithm. If I felt no moral qualms empowering someone else with this ability, then that is the same lack of qualms to use the ability for myself. If I were so inclined, I could use this technique to make that fat paycheck, and never have to walk away. This technique is effective in many fields beyond elections. In the situation I'd walk away, it would only be to suppress the work, in the hopes it stays undiscovered.
I also don't think many reports actually assert that CA was responsible for, say, a 10% swing. Why people are taking this serious is, I believe, because we now know that it's possible to get such data and we have seen the sort of feats that machine learning can do today. It's doubtful that CA was the company to successfully combine those two ingredients. But without changes it's all but guaranteed that someone will soon will.
Plus there is the fact that the election just happened to be incredibly close. You can make a list with about ten possible reasons for the result, from people being tired of Hillary, to the FBI's strange moves regarding the e-mail investigation, third-party candidates, russian trolls, the free media Trump got, Cambridge Analytica, and probably the weather in eastern Ohio. It's possible that the absence of any two or three of these would have changed the outcome.
[0]: example: https://www.vox.com/science-and-health/2018/3/23/17152564/ca...