First to discuss this topic please get acquainted with the topics of signalling theory, and most importantly the theory of collective reputation.
Also you can watch this video
A brief overview, signalling theory if we break it down, a signal in economics is based on the idea from contract theory that a signal provides some credible information information. For example raising prices gives a signal to potential customers. Lowering prices gives a potential signal to customers. If we look at my writing titled "A Market For Good and Bad Behaviors" I go into it a little bit more where I interpret human behavior as being based on economics where incentives increase the probability of certain behaviors occurring while disinentives or punishments reduce the probability of certain behaviors occurring in a market setting.
We do not control how we are initially perceived by others. We are born into collective reputations which are stereotyped. For example you could be born male or female, you could be born into a certain racial category, you could be born into any collectively stereotyped category (financial class, a US or non-US citizen, etc). These categories are what I call coerced identities, as these identities are socially enforced. If a person for example is born male then from birth there are many stereotypes applied to the male and there are expected behaviors from anyone in this category based on the collective reputation effect. Commonly, people born into these categories will be rewarded for meeting (or exceeding) the expected behaviors and punished for going against the expected behaviors. Society is built up of people, and people often believe in both stereotypes and in the collective reputations that different labels have.
A person can be born female, and immediately upon birth society places expectations on behavior (female vs male behavior) . Behaviors which meet these expectations are often rewarded while behaviors which go against it are often punished or disregarded as illegitimate behavior. Individuals are people who can choose their identity, but this does not mean the market or the society (or the market society) will accept true individuals. Parameters are placed on individuals where choices must be made by the individual and each choice the individual makes is sending some kind of signal. If an individual goes to college to pursue an education then employers might see this individual different. This signalling impacts behaviors because certain adopted behaviors are considered more socially desirable by the market than others. This would imply that not all behaviors have equal value in a market context (which is obvious).
Can The Crowd see Individuals even with Big Data?
As more information (big data) becomes available then on one hand it might be possible (via sentiment analysis) to determine the behaviors which are most valued in a particular society at a particular time. I call this (in demand behaviors). The impact of collective reputation is that it determines the behavioral expectations of an individual prior to an individual even being born.If you believe there is a such thing as an individual then "collective reputation" is the worst kind of collectivism that can exist because it allows the crowd to see not individuals but to see labels.
For instance the crowd does not see you, it sees a man or a woman, and then if you are a certain race it might see that, and then if you speak English it might see you as a US citizen, and from this there are all sorts of expectations which the crowd will have on you based on stereotypes (stuff people deemed similar to you did in the past or just total disinfo).
My current opinion is without big data the crowd can't really see or know anyone.
Can The Crowd ever really know anyone?
If the crowd is just a bunch of people subjectively judging other people, with human biases applied, with collective reputation applied (and the stereotypes), then I do not believe the crowd ever really tries to know anyone. If the crowd uses labels then it is using shortcuts designed to help the human brain to more quickly assess information. In a crowd of 200 or so people (Dunbar's number) then it might in fact be possible for that small community to know everyone to some degree. In a crowd of 100,000 or 1 million people what can we really know?
Big data and machine learning are potential game changers. It is possible using our technology to know everyone as an individual. This would mean we could do away with race, gender, class, or any of these social concepts (in theory). The problem is what if the majority of people internalize the identities they were given based on years of adopting behaviors (which were rewarded in some way)?
Steem is not immune - Does Steem reward good people or good content?
Steem relies on collaborative filtering and it is currently encouraged for humans to subjectively measure the value of certain content. In a more professional context these humans would be restricted to follow certain norms or rules of content measurement but we don't have a way to enforce that. As a result we have voting up or down in a subjective manner where your voting itself is a signal. If you for example vote for an account which the Steem community has judged as "bad" then what impact does it have on how your future votes might be perceived by the Steem community? If an account can be deemed "bad" (regardless of content) as a result of a subjective interpretation of the character of the owner of the account then there is bias. This bias means Steem will never be able to fairly reward content because the crowd which controls the votes might determine certain content producers are "evil" or "good" and vote not based on how they view the content itself but based on concerns about how they'll be viewed if they give an upvote to "evil".
Signals are used by the crowd to in essence reveal an identity. These signals are data and in the case of Steem it is on the blockchain forever. This means machine intelligence will someday be analyzing how every account on Steem voted. Accounts which voted in "evil" manners could be seen as "evil". The problems occur when people don't deem the account as evil but deem the person or people behind the account as evil based on how the account voted. There is noise in these signals but this provides an illustration of signalling theory and the impact it already has on Steem voting. The possibility does exist for anonymity behind Steem accounts, so in theory this problem is mostly the problem verified account holders on Steem must face. Are verified accounts worth more? If it's all a market then it's a fair question to ask.
On Steem certain arbitrary labels already exist. The "good" and "evil" on Steem fall often fall into "Steem Abuser" vs "Steem Supporter". The problem with these labels is no one in advance had any public list of behaviors which would put an account on this list until downvote brigades began to enforce the behaviors which they thought were abusive to the platform.
The problem of arbitrary labels and signalling on Steem
Now we can look at an example of something which happened recently which in my opinion was entirely a miscommunication. 
If the crowd for example decides a certain account is "bad" and is willing to actually label it as such then any account which supports the account deemed "bad" by the crowd could also be labeled "bad". The problems could be made even worse if an endless amount of arbitrary labels are invented based on account behaviors which could put either the account holder (if verified) or the account itself as part of the "bad". Collective reputation could also ruin Steem if the crowd begins voting not based on the content of the actual account but based on attributes of the account owner. For example if the account owner is a female and certain people downvote her account because she is "making too much money" then this would be an example of both a collective reputation being placed on her because of her gender, combined with expectations based on a stereotype (she's making too much money).
In my opinion we could see all of these behaviors on Steem and worse depending on what the community decided to value. If the crowd begins to label different accounts then it is certainly possible that machine learning can be applied to the data on the blockchain to determine from voting patterns (and other data) the probable behaviors of different account owners. Even if the different account owners are anonymous the data trail allows for the use of social physics to determine not only possible behaviors of the account owners, but also to apply predictive analytics. The problems in my opinion would be when assumptions are made about people morally or politically based on voting patterns on Steem.
What can be done about this?
Crowd dynamics and collective reputation are tough subjects. Social physics is a brand new area of research where machine learning is applied to the data (including data on blockchains) in new and interesting ways. Voting patterns are potentially signal data. Stereotypes, labels, collective reputation, superficial judgment, all take place whenever a crowd is big enough. If the goal on Steem is truly to upvote the best content then every account needs to be anonymous to prevent bias. If the goal on Steem is to have honest signals or have people vote what they really believe rather than based on how they want to be perceived by the other voters, then the votes have to be secret. If the votes are public then the voting patterns might not vote up the "best" content but instead vote up the fashionable content.
Consider if an account upvotes a certain account not because the account produces content that they like but because they want their account to be linked to that other account? Imagine of the blockchain votes are merely "links" and if you vote for an account of the really popular most liked people then it makes your account more popular and liked?
Furthermore the Steem community and influential members need to have a discussion on how they want the Steem community as a group to be seen by the world. If the reputation of the Steem Community as a group is only as good as it's members then what does this mean?
To be continued...
Reference
De Meo, P., Ferrara, E., Fiumara, G., & Provetti, A. (2011, November). Generalized louvain method for community detection in large networks. In Intelligent Systems Design and Applications (ISDA), 2011 11th International Conference on (pp. 88-93). IEEE.
Manski, C. F. (2000). Economic analysis of social interactions. Journal of economic perspectives, 14(3), 115-136.
Pentland, A. (2015). Social Physics: How social networks can make us smarter. Penguin.
TIROLE, J. (1996). A Theory of Collective Reputations (with applications to the persistence of corruption and to firm quality). The Review of Economic Studies, 63(1), 1-22.