Category: social computing

How to promote employees using machine learning without societal bias

Though it may at first read as being callous, a managerialist stance on inequality in statistical classification can help untangle some of the rhetoric around this tricky issue.

Consider the example that’s been in the news lately:

Suppose a company begins to use an algorithm to make decisions about which employees to promote. It uses a classifier trained on past data about who has been promoted. Because of societal bias, women are systematically under-promoted; this is reflected in the data set. The algorithm, naively trained on the historical data, reproduces the historical bias.

This example describes a bad situation. It is bad from a social justice perspective; by assumption, it would be better if men and women had equal opportunity in this work place.

It is also bad from a managerialist perspective. Why? Because if the point of using an algorithm were not to correct for societal biases introducing irrelevancies into the promotion decision, then it would not make managerial sense to change business practices over to using an algorithm. The whole point of using an algorithm is to improve on human decision-making. This is a poor match of an algorithm to a problem.

Unfortunately, what makes this example compelling is precisely what makes it a bad example of using an algorithm in this context. The only variables discussed in the example are the socially salient ones thick with political implications: gender, and promotion. What are more universal concerns than gender relations and socioeconomic status?!

But from a managerialist perspective, promotions should be issued based on a number of factors not mentioned in the example. What factors are these? That’s a great and difficult question. Promotions can reward hard work and loyalty. They can also be issued to those who demonstrate capacity for leadership, which can be a function of how well they get along with other members of the organization. There may be a number of features that predict these desirable qualities, most of which will have to do with working conditions within the company as opposed to qualities inherent in the employee (such as their past education, or their gender).

If one were to start to use machine learning intelligently to solve this problem, then one would go about solving it in a way entirely unlike the procedure in the problematic example. One would rather draw on soundly sourced domain expertise to develop a model of the relationship between relevant, work-related factors. For many of the key parts of the model, such as general relationships between personality type, leadership style, and cooperation with colleagues, one would look outside the organization for gold standard data that was sampled responsibly.

Once the organization has this model, then it can apply it to its own employees. For this to work, employees would need to provide significant detail about themselves, and the company would need to provide contextual information about the conditions under which employees work, as these may be confounding factors.

Part of the merit of building and fitting such a model would be that, because it is based on a lot of new and objective scientific considerations, it would produce novel results in recommending promotions. Again, if the algorithm merely reproduced past results, it would not be worth the investment in building the model.

When the algorithm is introduced, it ideally is used in a way that maintains traditional promotion processes in parallel so that the two kinds of results can be compared. Evaluation of the algorithm’s performance, relative to traditional methods, is a long, arduous process full of potential insights. Using the algorithm as an intervention at first allows the company to develop a causal understanding its impact. Insights from the evaluation can be factored back into the algorithm, improving the latter.

In all these cases, the company must keep its business goals firmly in mind. If they do this, then the rest of the logic of their method falls out of data science best practices, which are grounded in mathematical principles of statistics. While the political implications of poorly managed machine learning are troubling, effective management of machine learning which takes the precautions necessary to develop objectivity is ultimately a corrective to social bias. This is a case where sounds science and managerialist motives and social justice are aligned.

How to tell the story about why stories don’t matter

I’m thinking of taking this seminar because I’m running into the problem it addresses: how do you pick a theoretical lens for academic writing?

This is related to a conversation I’ve found myself in repeatedly over the past weeks. A friend who studied Rhetoric insists that the narrative and framing of history is more important than the events and facts. A philosopher friend minimizes the historical impact of increased volumes of “raw footage”, because ultimately it’s the framing that will matter.

Yesterday I had the privilege of attending Techraking III, a conference put on by the Center for Investigative Reporting with the generous support and presence of Google. It was a conference about data journalism. The popular sentiment within the conference was that data doesn’t matter unless it’s told with a story, a framing.

I find this troubling because while I pay attention to this world and the way it frames itself, I also read the tech biz press carefully, and it tells a very different narrative. Data is worth billions of dollars. Even data exhaust, the data fumes that come from your information processing factory, can be recycled into valuable insights. Data is there to be mined for value. And if you are particularly genius at it, you can build an expert system that acts on the data without needing interpretation. You build an information processing machine that acts according to mechanical principles that approximate statistical laws, and these machines are powerful.

As social scientists realize they need to be data scientists, and journalists realize they need to be data journalists, there seems to be in practice a tacit admission of the data-driven counter-narrative. This tacit approval is contradicted by the explicit rhetoric that glorifies interpretation and narrative over data.

This is an interesting kind of contradiction, as it takes place as much in the psyche of the data scientist as anywhere else. It’s like the mouth doesn’t know what the hand is doing. This is entirely possible since our minds aren’t actually that coherent to start with. But it does make the process of collaboratively interacting with others in the data science field super complicated.

All this comes to a head when the data we are talking about isn’t something simple like sensor data about the weather but rather is something like text, which is both data and narrative simulatenously. We intuitively see the potential of treating narrative as something to be treated mechanically, statistically. We certainly see the effects of this in our daily lives. This is what the most powerful organizations in the world do all the time.

The irony is that the interpretivists, who are so quick to deny technological determinism, are the ones who are most vulnerable to being blindsided by “what technology wants.” Humanities departments are being slowly phased out, their funding cut. Why? Do they have an explanation for this? If interpetation/framing were as efficacious as they claim, they would be philosopher kings. So their sociopolitical situation contradicts their own rhetoric and ideology. Meanwhile, journalists who would like to believe that it’s the story that matters are, for the sake of job security, being corralled into classes to learn CSS, the programming language that determines, mechanically, the logic of formatting and presentation.

Sadly, neither mechanists nor interpretivists have much of an interest in engaging this contradiction. This is because interpretivists chase funding by reinforcing the narrative that they are critically important, and the work of mechanists speaks for itself in corporate accounting (an uninterpretive field) without explanation. So this contradiction falls mainly into the laps of those coordinating interaction between tribes. Managers who need to communicate between engineering and marketing. University administrators who have to juggle the interests of humanities and sciences. The leadership of investigative reporting non-profits who need to justify themselves to savvy foundations and who are removed enough from particular skillsets to be flexible.

Mechnanized information processing is becoming the new epistemic center. (Forgive me:) the Google supercomputer approximating statistics has replaced Kantian trancendental reason as the grounds for bourgious understanding of the world. This is threatening, of course, to the plurality of perspectives that do not themselves internalize the logic of machine learning. Where machine intelligence has succeeded, then, it has been by juggling this multitude of perspectives (and frames) through automated, data-driven processes. Machine intelligence is not comprehensible to lay interpretivism. Interestingly, lay interpetivism isn’t comprehensible yet to machine intelligence–natural language processing has not yet advanced so far. It treats our communications like we treat ants in an ant farm: a blooming buzzing confusion of arbitrary quanta, fascinatingly complex for its patterns that we cannot see. And when it makes mistakes–and it does often–we feel its effects as a structural force beyond our control. A change in the user interface of Facebook that suddenly exposes drunken college photos to employers and abusive ex-lovers.

What theoretical frame is adequate to tell this story, the story that’s determining the shape of knowledge today? For Lyotard, the postmodern condition is one in which metanarratives about the organization of knowledge collapse and leave only politics, power, and language games. The postmodern condition has gotten us into our present condition: industrial machine intelligence presiding over interpretivists battling in paralogical language games. When the interpretivists strike back, it looks like hipsters or Weird Twitter–paralogy as a subculture of resistance that can’t even acknowledge its own role as resistance for fear of recuperation.

We need a new metanarrative to get out of this mess. But what kind of theory could possibly satisfy all these constituents?

Aristotelian legislation and the virtual community

I dipped into Aristotle’s Politics today and was intrigued by William Ellis’ introduction.

Ellis claims that in Aristotle’s day, you would call on a legislator as an external consultant when you set about founding a new city or colony. There were a great variety of constitutions available to be studied. You would study them to become an expert in how to design a community’s laws. Classical political philosophy was part of the very real project of starting new communities supporting human flourishing.

We see a similar situation with on-line communities today. If cyberspace was an electronic frontier, it’s been bulldozed and is now a metropolis with suburbs and strip malls. But there is still innovation in on-line social life as social media infrastructure as users migrate between social networking services.

If Lessig is right and “code is law“, then the variety of virtual communities and the opportunity to found new ones renews the role of the Aristotelian legislator. We can ask questions like: should an on-line community be self-governing or run by an aristocracy? How can it sustain itself economically, or defend itself in (cyber-)wars? How can it best promote human flourishing? The arts? Justice?

It would be easy to trivialize these possibilities by noting that virtual life is not real life. But that would underestimate the shift that is occurring as economic and political engagement moves on-line. In recognition and anticipation of these changes, philosophy has a practical significance in comprehensive design.