Tuesday, June 26, 2012

Google X, AI and cat recognition

So let me see if I understand this

Google's "X laboratory" (geez, secret supervillain organisation-sounding name much?) has been experimenting with AI and neural nets. They decided to hook 16,000 computer processors together and set this assemblage to examining a few million stills from Youtube videos. Without any prompting or input about what it should or should not look for, the neural net learned how to distinguish "cat" from "not a cat", and to recognise when a cat appeared in a still.

That's kind of cool. Weird, but cool.

Thursday, June 21, 2012

Critical realism and scientific inquiry: natural vs social science (Part II: the role of interpretation)

An ongoing debate within the philosophy of science is the extent to which knowledge about reality that is independent of a person's subjective interpretation of it is possible. The position of critical realism is that people cannot conceive of external reality outside of frameworks of understanding that pre-exist the attempt to perceive external reality and influence what is perceived, and how it is perceived and understood. Examples of such frameworks include established scientific theories and paradigms.

Multiple frameworks of understanding can exist with regard to any given area of reality, and they can be incompatible with each other. However, unlike relativist philosophies such as postmodernism, critical realism does not take this to mean that reality is nothing more than how humans decide to perceive it. For critical realists, an independent, external reality exists, and although conceptualisations of it can be imperfect and occasionally contradictory, this is not an excuse to give up on scientific inquiry. Indeed, one of the goals of any science is to try and make existing interpretations and conceptualisations of external reality more accurate. This is what makes critical realism critical: the goal of science in critical realism is to test our understanding of reality and through better understanding of it, improve the human condition.

A key difference in natural and social sciences is the role of interpretive frameworks of understanding in their respective areas of inquiry. The natural sciences make use of interpretive frameworks, utilising them to explain findings or updating them when they do not. A classic example is the evolution of the understanding of the atom: from conceptualising them as the smallest particle that exists, to realising that atoms themselves are composed of sub-particles, in the process evolving from the "plum pudding" model of protons and electrons in an atom into the more familiar "electrons orbiting a nucleus" model that most schoolchildren still learn as the correct one, and even further into the vagaries of quantum physics (which I won't pretend to fully understand).

This use of interpretive frameworks of understanding also occurs in the social sciences. But the social scientist is studying humans. Humans make use of interpretive frameworks of understanding in every area of their lives. Therefore, scientists don't just use interpretive frameworks in their study, but study them directly. Once again, this means that social science is difficult in a way that natural science is not. For one, how do you study an interpretive framework that is very dissimilar to your own interpretive framework? Can you even make sense of it at all?

Various methods exist to try and make unfamiliar interpretive frameworks more readily accessible and understandable, not just to the social science researcher but to anyone who is actually going to read the research as well. The most famous in social science is Clifford Geertz's "thick description": studying a culture in depth and trying to describe every experience that is observed to occur in an unfamiliar culture in as much detail as possible, with as much detail to how it relates to other parts of the culture as possible. Yet even here it is difficult to touch on everything, and important details can easily left out for no other reason that the researcher doesn't know to look for them.

Even such unintended misrepresentations of the conceptual understandings of others are problematic. The claim to be able to authentically represent the mental and social world of another carries a great deal of responsibility with it, and authoritative accounts that are inaccurate can  have serious negative consequences for the nature of interactions between different cultures, or even between people who share a similar culture but whose worldviews differ in some fundamental way. A particular danger here, and one that can't readily be overcome, is that a social scientist can never completely represent an external interpretive framework completely in its own terms, but will always represent it in terms of their own interpretive framework. An obvious consequence of this is that social science researchers should, in order to be able to do their job most effectively, try to cultivate for themselves as broad a range of available interpretive frameworks of understanding as possible.

And this brings up a further complication. Since social scientists are themselves social beings, any actions that they take have an effect on the society with which they are interacting. That includes the act of undertaking social research. In particular, the interpretive frameworks that a social scientist might make use of in order to explain an aspect of society that they are investigating could readily become adopted by members of that society as their own method of explanation for that aspect of their society, and potentially for other aspects of their society as well. This is known in social science as the "double hermeneutic", and it means that social scientists have a much greater ethical responsibility to practise care in their research, especially in their interpretation of their findings, than would be found in the natural sciences.

That's basically it as far as I understand the differences. I feel like I've left details out, but I think the core concepts are sound in my mind now. I don't suppose anyone else will find it particularly useful. I should get back to blogging about stuff that's more useful to other people.

Wednesday, June 20, 2012

Critical realism and scientific inquiry: natural vs social science (Part I: Open and Closed Systems)

Critical realism is a philosophy of social science originally pioneered by Roy Bhaskar. Originally quite useful in laying out a theoretical model of experienced reality that sought to explain why it is even possible to perform scientific research at all, his later work took things in a rather esoteric and, frankly, mystical direction. But the work on critical realism that did not take this "spiritual turn" is quite interesting, especially in how it seeks to explain why the methods of natural science and the methods of social science must be significantly different.

The premise of critical realism is realist: it presumes that there is a reality "out there" that exists independently of our knowledge of it. The means by which we learn about this reality is fallible, affected (but not determined) by our pre-existing theories about it, and different aspects of reality have different methods through which those aspects can be studied. Critical realism, as a theory about scientific inquiry, puts forward some ideas about the different nature of physical and social objects of study, and these ideas demonstrate differences in the way that scientific inquiry ought to work for the natural and social sciences.

Open vs Closed Systems

In natural sciences, the preferred method of study is the laboratory. In the terminology of critical realism, this is referred to as a closed system: a single aspect of reality is to be studied, so parts of reality that are irrelevant to that study are artificially removed so as not to affect the outcome. Even in situations where study isn't specifically in a laboratory, the goal is still much the same: close the system so as to be able to focus specifically on one aspect of external reality. In critical realist terminology, the goal off the closed system is to isolate and prove the existence of a "causal mechanism". In natural science terminology, a causal mechanism is a scientific law.

Closed systems are artificial environments. This becomes problematic when the object of study is social reality. According to critical realist philosophy, most attributes of social reality are not the result of simple, easy to isolate causal mechanisms, but stem directly from the interaction of multiple causes; each cause by itself would not produce even remotely the same effect by itself. Additionally, some effects may be actually negated by other causes, effectively removing evidence of the effect, even though the cause of the expected effect is present and salient. While isolating and testing a single cause-and-effect relationship can be done in some situations, there are a whole host of causal mechanisms that cannot be isolated for study, as the emergent effects of the interactions between those mechanisms forms a vital part of the picture of what is going on. This may not be so bad if there are only a few such interacting mechanisms, but in even the simplest social system, there are a whole host of them. And different social systems and situations can have entirely different causal mechanisms in operation.

To complicate matters even further, social reality is in part defined by its relationship to other parts of that reality, social or otherwise. Even before critical realism, an ongoing criticism of laboratory social science was that social scientific laboratory experiments don't explain what people do, but only what people do when they're in the social environment a laboratory. Trying to close off a social system for study potentially changes how that social system works, which means that any results from such a study cannot readily be generalised to social systems which have not been artificially closed off in the same way.

The goal of social scientific study, similar to attempts in natural science to uncover scientific "laws", is to discover causal mechanisms that explain the observed behaviour of an external reality. But while the natural scientist can observe closed systems and interpret them through deduction, along with some amount of induction (i.e. to consider how generalisable the experimental findings might be), the social scientist has to observe open systems. And while the social scientist may be able to make some deductive conclusions from such observations, they must make make much greater use of inductive reasoning to try and figure out what causal mechanisms are present and salient in a particular social system.  In order to consider what causal mechanisms might even potentially be present, they must also make use of what critical realists call "retroduction": the attempt to explain an aspect of reality by positing what pre-conditions are necessary for such an aspect of reality to exist.

Both induction and retroduction are less reliable tools of inquiry than deduction, As social science must rely on these tools to a greater extent than natural science, the claims of social science tend to be less reliably true than claims made in the natural sciences. Such "laws" as can be posited from these methods of inquiry are also much harder to test across different social systems and situations. This means that the explanations provided by social science are much more tentative than in natural science, and the extent to which findings from the study of one area of social reality can be applied to a different area of social reality is much less clear than in it is in the study of reality undertaken by the natural sciences.

I actually wrote this in an attempt to clarify my own understanding of critical realism's understanding of the differences between natural and social science. It's worked (I think), but it's taken longer for me to explain the first difference between them  than I expected it to. The other main difference - the element that interpretation plays as both something that is applied to findings in all types of scientific study, and as something that is also studied by social science - will have to wait for Part II.

Sunday, June 10, 2012

The insidious harm of ex-gay "therapy"

Courtesy of some people coming forward in Australia, the existence of "ex-gay therapy" in Australia is generating headlines, both in News Limited's holdings (Christian groups try to cure homosexual teens with brainwashing treatment) and in the Age (Ministries preying on gay shame). For once, it seems, the attitude of even News Ltd's holdings towards anti-gay sentiment is unabashedly negative.

The Age article actually features an interview from someone who works at an ex-gay program. Haydn Sennit, "pastoral care worker" for the Sydney ex-gay group Liberty Christian Ministries, answered a few questions, and in the process revealed something about the attitude that such ministries have to people who say they experience no change in sexual orientation. Simply put: if a person doesn't change, it's their own fault for not trying hard enough. Some sample quotes:
Some people give up, while others keep going and it’s different for every individual
Success is varied and it depends a lot on a person’s personal commitment.
And the kicker:
Some give up entirely because it’s so hard and it’s actually their disappointment with themselves that gets them undone.

Yes, they do get "undone" by disappointment in themselves. But not in the sense that they fail to experience change as a result of giving up. They get undone in the sense that their desire to give up after failing to experience change is turned into evidence of their weakness. The victim-blaming rhetoric that Sennit here is propagating prevents would-be ex-gays from acknowledging that they have tried to change as hard as they could, that they literally could not try any harder to change, and that they still wouldn't change even if they could. They are told to blame themselves for failing to do something impossible, and are told that their inability to experience change, even after monstrous effort at it, is evidence of their own weakness, not of the impossibility of the task before them.

Yes, ex-gay therapy teaches people to hate and fear their own desires for sexual intimacy. Yes, it teaches people to be ashamed of themselves, to think of themselves as "sexually broken" instead of just as gay. But the cruelest and most insidious lesson that they teach is that, no matter how much effort a person makes to change, it will never be enough. Those who admit to themselves that they will never change are made to think of themselves as weak-willed failures. No wonder so many of them become suicidal.