Wednesday, October 7, 2015

Is Transnational Surrogacy Exploitative? (1) Kirby's Argument

Surrogacy Hostel - India

Surrogacy is the practice whereby a commissioning parent (or parents) procures the services of a surrogate who will carry a child to term on their behalf. For obvious reasons, all surrogates are thus biologically female. There are two main categories of surrogacy: (i) genetic, whereby a male commissioning parent impregnates the surrogate (typically via artificial insemination); and (ii) gestational, whereby the commissioning parent(s) provide (or procure) an embryo for implantation in the surrogate. The latter has a number of distinct sub-varieties, each varying depending on how exactly the embryo is procured. There are also distinctions to be drawn between altruistic surrogacy, which is not done for profit, and commercial surrogacy, which is.

India has become a major international destination for commercial surrogacy. The practice is legal there since 2002, and one of the goals of the Indian government is to make India a prime destination for medical tourism. Adding to its attraction is the fact that surrogacy is significantly cheaper in India than it is in many Western countries (at least, in those Western countries that have legalised commercial surrogacy). For example, the total cost of a surrogacy arrangement in the US, including medical expenses and surrogate’s fee, is estimated to range from $50K-$150K, whereas the total cost in India, including travel expenses for the commissioning parents, is approximately $25K-$30K. As such, it is common for commissioning parents in high income countries to take advantage of the lower costs in places like India.

The question I want to consider over the next couple of posts is whether such transnational gestational surrogacy (TGS) arrangements are exploitative. I’ll do so by reviewing some recent papers on this topic by Jeffrey Kirby and Stephen Wilkinson. I start today with Kirby’s paper ‘Transnational Gestational Surrogacy: Does it have to be exploitative?’, which appeared in the American Journal of Bioethics in 2014. Kirby argues that TGS probably is exploitative in its current form, but that it need not be.
Note: throughout this post and its sequels, much of the focus will be on India. This is because India is a relatively well-studied example, with a number of empirical and ethnographic reports available about the practice and the lives of surrogates in that country. But it is not the only destination for TGS and the evaluation undertaken here could be applied to other destinations as well.

1. What is exploitation? Kirby’s Three Pronged Framework
We start by looking at the concept of exploitation. Exploitation is morally significant. If I exploit you, then I have wronged you and may owe you some duty of repair. Similarly, the fact that a relationship or transaction is exploitative is usually taken to be a reason (even if not a decisive reason) for deeming that relationship or arrangement impermissible (we could get all philosophical and call it a pro tanto or prima facie reason, but we won’t bother).

But then when is a relationship or transaction exploitative? This is something that has been much debated over the years. The basic gist of the idea is that a transaction is exploitative if you take unfair advantage of another person (or group of persons). This usually means that you profit at their expense, though sometimes exploitative transactions can be mutually advantageous. Indeed, this may be true of TGS transaction since in such arrangements both the commissioning parents and the surrogate are gaining (to at least some degree). What makes the transaction exploitative in such cases (if it is at all) is determined by reference to how the benefits/burdens of the transaction are distributed between the parties.

One of the nice features of Kirby’s article is his review of previous attempts to define exploitation. I’m not going to review his review. Instead, I’ll skip right to the end where Kirby offers his own, somewhat complex, framework for evaluating whether or not an arrangement is exploitative. The framework tries to capture the moral insights from previous definitional efforts (including liberal, Marxist and social justice definitions). It identifies three conditions which can independently determine whether exploitation arises. I’ll quote from the article at this point:

Kirby’s Exploitation Framework: “In a transaction between individuals/parties, exploitation is constituted by the taking of unfair advantage, whereby one individual/party gains at the expense of another individual/party who is relatively disadvantaged on economic, social and/or political grounds. A finding of exploitation is not precluded by a “better” transaction outcome for the disadvantaged individual/party than the status quo ante, and transactions that produce significant mutual advantage may be exploitative to an individual/party. The taking of exploitative unfair advantage is demonstrated by the meeting of one or more of the following three exploitation conditions, each of which is sufficient for a finding of exploitation:
Exploitation Condition 1: It is unlikely (improbable) that the disadvantaged individual would voluntarily choose to enter into, and continue in, the transaction on the basis of her/his full knowledge and understanding of its anticipated burdens and benefits. [Note: members of the relevant group or community of disadvantaged individuals will deliberate together to determine whether this condition is met].
Exploitation Condition 2: The disadvantaged individual is unable to make a meaningful decision about whether to enter into, and continue in, the transaction; that is, the choice is not well-informed and/or is coercive in nature (the latter, at least to the extent that the individual could not really choose to do otherwise).
Exploitation Condition 3: The interests of the disadvantaged individual are not taken into account in the development of the transaction and members of the disadvantaged individual’s social group/community do not participate in, and significantly inform, decision making about the transaction’s (initial and ongoing) terms and labor-related conditions.”
(Kirby 2014, 26-27)

There is a lot going on in this. I suggest re-reading it a couple of times to grasp its full import. One thing worth noting in particular is how Kirby ties his framework to the views of the typical individuals in the relevant community of disadvantage. There is an almost explicitly consultative dimension to it: in order to determine whether the conditions of exploitation have been met, we need to consult with (or at least consider the views of) the people who may or may not be exploited by the transaction. This consultative dimension is very clear in the first and third conditions.

Anyway, with the framework in place, we can move on to Kirby’s main argument which is that, at present, TGS is indeed exploitative in India, but that this need not always be the case. He grants that Indian surrogates are members of a disadvantaged group: they come from a developing nation and have much lower average incomes than Western commissioning parents. He then questions whether each of the three conditions is met in the case of the typical Indian surrogacy arrangement. Let’s go through each step of the analysis.

2. Is First Exploitation Condition met in the case of Indian Surrogates?
The first condition is the most difficult to assess. It states that a transaction is exploitative if the disadvantaged party would be unlikely to enter into it if they were fully aware of the benefits and burdens involved. This means we have to work out what these benefits and burdens are. This requires some engagement with the empirical studies done on the lives of surrogates in India, followed by a complex weighting of the evidence drawn from these studies.

The benefits of surrogacy (from the perspective of the surrogate) are relatively clear. The surrogate receives a considerable amount of money for their services. Chang and Jaiswal estimate that the average fee ranges from $6K-$10K. A study in the Lancet in 2012 put the figures at $5K-$7K. To Western eyes, this may not seem like much, but the significance of these figures must be appreciated within the relevant economic context. Most Indian surrogates are women from poor rural communities. The average annual income for such women, according to the Lancet report, is in the region of $300. So the fee represents (roughly) 16 to 33 times the average income.

Amrita Pande has conducted a number of interviews with surrogates who describe the benefits of this money in terms of how it enhances their material and social circumstances, e.g. paying for education, housing, healthcare, business debts, and so on. Interestingly, some women don’t conceive of the benefits purely in terms of the instrumental value of the money. Some value becoming a surrogate because it temporarily removes them from other workplaces in which accidents and injuries are common (Humbyrd 2009).

So, as I say, the benefits seem relatively clear. What about the burdens? As you might expect — given that Kirby’s article was published in a bioethics journal — he is quite thorough when it comes to this issue, particularly the medical procedure and its health risks and side effects. He divides these burdens into two main classes: (i) physical and (ii) psychological.

The physical burdens arise from the hormonal manipulation that is used to prepare the surrogate for implantation; the process of embryo transfer; and the common risks associated with pregnancy and giving birth. Birth control pills are often used to sync the cycles of the commissioning woman and the surrogate. Gonadotropin-releasing hormone (GnRH) is used to prevent premature ovulation. This is usually well-tolerated but has known side effects including hot flashes, fatigue, headaches, nausea, irritability. Estrogen tablets are used to thicken the uterine lining prior to implantation. Progesterone capsules (or suppositories or injections) are used before embryo transfer and during the early weeks of pregnancy. Both have well-known side effects including bloating, irritability and breast soreness. There are also cumulative risks if you undergo the process several times. The actual process of embryo transfer involves some (usually mild) physical discomfort (a soft catheter is inserted into the uterus). If several embryos are transferred, there is a risk of multiple gestation and preterm births. Becoming pregnant and giving birth also carries risks, including a risk of mortality, but these risks are relatively low given the high standard of care provided by surrogacy clinics. Of greater significance, might be the frequent use of caesarian sections to facilitate the scheduling needs of the clinics and commissioning parents. Finally, Kirby notes a recent study by Stewart et al. (2013) which suggests that women who undergo IVF are at a slightly elevated risk of ovarian tumors, but these tumors are ‘of low malignant potential’.

The psychological burdens are complex, and arise from cultural factors and personal circumstances. One significant feature of surrogacy in India is that some surrogates move into hostels that are run by the fertility clinics. This separates them from their families (all these women must have had at least one child of their own) for extended periods of time. There is also a risk of social stigmatisation as some people equate surrogacy with marital infidelity or prostitution. Indeed, Pande notes that this is one reason why women move to surrogacy hostels: to conceal their pregnancies from members of their local community. There is also the psychological burden that arises from ‘giving up’ the child at the end of the transaction, due to attachment that arises during pregnancy. This is sometimes thought to be lessened in the case of gestational surrogates since there is no genetic connection between the mother and the infant. But Pande argues that this may not be true in parts of India where the significance of genetics is poorly understood and the primary attachment-concept is that of ‘shared bodily substances’ (e.g. blood through the placenta or breast milk). These psychological risks could be mitigated through counseling and pre-screening, but such services are not usually available to women in these settings.

The benefits and burdens are summarised in the table below.

What conclusion should we draw from this description of the benefits and burdens? One of the frustrating things about Kirby’s article is his failure to provide quantitative estimates for the various health risks and side effects. I’m sure these are available somewhere, and they would really help with weighting the benefits relative to the burdens. But, in any event, Kirby doesn’t want to engage in that exercise. He thinks the description of the benefits and burdens reveals just how difficult it would be to determine whether the first exploitation condition is met. There are clearly benefits here — ones that are appreciated by women in the relevant communities and could be potentially transformative — and there are also clearly burdens — some of which are unique to India, some of which are common to all surrogates. He thinks it would be really difficult for him (an outsider) to say whether a properly informed woman would choose to become a surrogate given those benefits and burdens. Fortunately, he doesn’t have to because the other two conditions provide independent grounds for exploitation.

That’s a somewhat underwhelming and disappointing conclusion. My guess — and it is only a guess — is that there would be at least some women for whom the benefits would outweigh the burdens.

3. Is the Second Exploitation Condition met in the case of Indian Surrogates?
The second condition says that a transaction is exploitative if the decision to enter into it is not the product of meaningful choice. This is cashed out by further reference to how ill-informed or coercive in nature the decision might have been.

Kirby thinks this condition is met in the case of the typical woman from rural western India. It doesn’t appear that women are actively coerced into these transactions (though some have argued the size of the fee could have a paradoxically coercive effect). But it does appear that they are ill-informed. Indeed, the available evidence suggests that such women are not provided with much information and often don’t understand the nature of the transaction they are getting into. Pande’s investigations and interviews are a major source of evidence in this regard.

She discovered that in Anand (one of the main regions in India’s TGS trade), little information was provided to the women and no attempt was made to ensure that they understood this information. Perhaps most damningly, the standard contract was written in English, a language that none of the women she interviewed could read. Her interviewees, in turn, revealed the paucity of information and understanding. One woman, Gauri, said:

‘…the only thing they told me was that this thing is not immoral, I will not have to sleep with anyone, and that the seed will be transferred into me with an injection…they also said that I have to keep the child inside me, rest for the whole time, have medicines on time, and give up the child.’ 
(Pande 2010, 976-977 - taken from Kirby 2014)

Given the preceding description of the benefits and burdens associated with surrogacy, you can plausibly argue that this is insufficient for truly informed consent and meaningful choice. To be sure, poor understanding of the risks and benefits of medical procedures are common in many Western contexts too, but from this description it seems like the attempt at informed consent for the surrogates in western India falls well below the desired level.

There are some potential counterarguments. The practice in India is for surrogacy brokers (“middle-women”) to go door-to-door recruiting potential surrogates. Most women refuse, suggesting that the women who do become surrogates may have had the ability to do otherwise. But Kirby thinks this is insufficient to displace worries about a lack of informed consent. Similarly, there are some women who undergo the process repeatedly. They have presumably acquired some experiential knowledge that renders their decisions more meaningfully ‘informed’. But Kirby thinks that ‘this does not preclude the possible influence of coercive factors such as “brainwashing” and/or escalated financial inducement’ (2014, 29). Finally, someone of a libertarian persuasion could argue that these women have a negative right not to be interfered with when it comes to choosing to enter a surrogacy transaction. But Kirby dismisses this on the grounds that a pure negative freedom is insufficient for a genuinely autonomous decision: what matters is who controls the choice-context.

I’m broadly in agreement with this. I think the informed consent angle (in particular, the language issue) needs to be addressed. That said, as per my previous comment, I wouldn’t rule out the possibility that some women make an informed decision. In this regard, I am slightly more persuaded by the possibility of experiential knowledge sufficing for an informed decision than Kirby appears to be.

4. Is the Third Exploitation Condition met in the case of Indian Surrogates?
This brings us to the last exploitation condition. This one states that a transaction is exploitative if the disadvantaged individual/community has no real say in the terms and conditions that govern the transaction. This provides the most plausible exploitation-argument against TGS in India. Why? Because it seems pretty clear that surrogates have no say in setting the terms and conditions of these transactions. It is the fertility clinics, medical practitioners, surrogacy brokers, government and commissioners who have all the bargaining power. This seems clearly evidenced by the frequent use of caesarian sections to suit the scheduling requirements of the clinics and commissioners; by the drafting of contracts in a language that the women cannot read; by the use of surrogacy hostels; and by the failure to provide appropriate counseling and support.

The Assisted Reproductive Technology Bill 2014 (which as far as I can tell is still merely proposed rather than actual law) does attempt to introduce additional regulatory safeguards. In particular, it tries to make it a bit more difficult for foreign commissioning parents to procure surrogates. But, as Kirby notes, it is not a progressive piece of legislation. It requires consent from the husbands of surrogates, and there is no evidence to suggest that women from the relevant communities have been consulted. There have also been earlier reforms that have not protected the interests of the women involved. For example, a 2010 reform reduced the front-loading of the surrogate’s fee, which meant the surrogate would not be fully paid if they lost their pregnancies at a later stage.

For what it is worth, I agree with Kirby on this point: the lack of bargaining power on the part of the women affected by these transactions would seem to provide a pretty compelling argument against them.

5. So what can be done?
So Kirby thinks that TGS (as practiced in India) is exploitative because conditions 2 and 3 of his framework are met. He has thus provided a moral argument against TGS. But the argument is highly contingent in nature. It does not say that TGS is ‘in principle’ exploitative, merely that it is exploitative in its current form. This suggests that the system could be reformed.

And, indeed, Kirby does suggest three general categories of reform which could render these transactions non-exploitative. They are:

Public Education and Enabled Choice: A ‘comprehensive public education program’ could be introduced to combat myths and misinformation about surrogacy and to reduce the stigma that might attach to surrogates. Women thinking of becoming surrogates could be ‘required to attend a government-sponsored educational workshop’ that will explain the benefits and burdens of the transaction to them. Neutral ‘advocate-navigators’ could be recruited and trained to help the women through the process (Kirby 2014, 30).

Enhanced Protections: There could be enhanced regulatory protections. This could include setting a maximum delivery age of 34 for potential surrogates (reduces potential complications), limiting the number of embryo transfers to prevent the risk of multiple gestations, limiting the maximum number of surrogate pregnancies to reduce risk of multiple rounds of IVF, and ensuring that the safest hormone protocols are used. There could also be screening evaluations and counseling available to minimise the psychological burdens.

Empowerment: Efforts could be made to enhance the agency and bargaining power of surrogates. This could include consultation and deliberative engagement with the relevant communities in order to develop standard terms and conditions for surrogacy contracts; ensuring that remuneration is fair by front-loading payments to surrogates; and setting a price floor (i.e. minimum wage). It could also include ‘high quality educational programs that provide training for local, decent-wage jobs’ for surrogates during their pregnancies, as opposed to the existing system of surrogacy hostels providing (some) English and computer courses, which are primarily for the benefit of the commissioning parents who wish to communicate with surrogates (Kirby 2014, 31).

Many of these reforms sound commendable, and may genuinely help to remove some of the exploitative taint from surrogacy transactions. I have only two observations. First, note that some of these reforms involve placing additional burdens on the disadvantaged women, e.g. mandatory education and psychological screening. We might want to question that. As I mentioned at length in my posts on sexual consent, burdens of this sort represent a tradeoff between positive and negative autonomy: extra burdens mean we protect negative autonomy to a greater degree than positive autonomy. This may be entirely appropriate, given the risks associated with becoming a surrogate and the potential exploitation, but we should acknowledge that the tradeoff is being made.

Second, and maybe more importantly, many of the mooted reforms would have the effect of increasing the cost of surrogacy in India. Kirby concedes as much in his discussion. Raising costs may have some perverse effects on surrogates in India. All transactions between people in high income countries and people in low income countries have an exploitative element to them: they involve people from richer countries taking advantage of lower costs in poorer countries. Some people find this appalling; some think it is good because it aids economic growth and provides better alternatives for the people in low-income countries. I don’t know where you would come down on this issue, but in either case the danger with raising costs in India is that it might drive the market for TGS to another cut-price jurisdiction. This could deprive women in India of income and push the problem onto another disadvantaged population. Kirby thinks this possibility could be mitigated by ensuring the costs don’t go too high. But his discussion mainly focuses on ensuring a price differential between India and high income countries; it doesn’t really focus on the price differential between India and other low income countries. Of course, you could argue that this is okay because the reforms in India are morally justified, and any other jurisdiction would simply be obliged to introduce similar reforms. And I would agree with that, but this reform effort is likely to be a drawn out process (Note: this depends entirely on whether other jurisdictions would be willing to legalise commercial surrogacy in the same way; I’m not sure how likely this is given that there are numerous other objections to the practice).

In conclusion, TGS as practiced in India is probably exploitative. Although it is difficult to weigh the relative benefits and burdens of becoming a surrogate, it seems pretty clear from the available evidence that surrogates in India are poorly informed about the transactions they are entering into and lack bargaining power in shaping the terms and conditions of those transactions. Reforms could be introduced to improve the situation, but these will involve trading-off different values and may not improve the overall quality of life for surrogates in India because of the dynamics of transnational capitalism.

Monday, October 5, 2015

Polanyi's Paradox: Will humans maintain any advantage over machines?

(Previous Entry)

There is no denying that improvements in technology allow machines to perform tasks that were once performed best by humans. This is at the heart of the technological displacement we see throughout the economy. The key question going forward is whether humans will maintain an advantage in any cognitive or physical activity. The answer to this question will determine whether the future of the economy is one in which humans continue to play a relevant part, or one in which humans are left behind.

To help us answer this question it is worth considering the paradoxes of technological improvement. It is truly amazing that advances in artificial intelligence have allowed machines to beat humans at cognitive games like chess or Jeopardy!, or that cars can now be driven around complex environments without human assistance. At the same time, it is strange that other physical and cognitive skills have been less easy for machines to master, e.g. natural language processing or dextrous physical movements (like running over rough terrain). It seems paradoxical that technology could be so good at some things and so bad at others.

Technologists and futurists have long remarked on these paradoxes. Moravec’s paradox is a famous example. Writing back in the late 80s, Hans Moravec (among others) noted the oddity in the fact that high-level reasoning took relatively few computational resources to replicate, whereas low-level sensorimotor skills took far more. Of course, we have seen exponential growth in computational resources in the intervening 30 years, so much so that the drain on computational resources may no longer be an impediment to machine takeover of these tasks. But there are other problems.

This brings us to the (very closely related) ‘Polanyi’s Paradox’, named in honour of the philosopher and polymath Michael Polanyi. Back in 1966, Polanyi wrote a book called The Tacit Dimension, which examined the ‘tacit’ dimension to human knowledge. It argued that, to a large extent, human knowledge and capability relied on skills and rulesets that are often beneath our conscious appreciation (transmitted to us via culture, tradition, evolution and so on). The thesis of the book was summarised in the slogan ‘We can know more than we can tell’.

Economist David Autor likes Polanyi’s Paradox (indeed I think he is the one who named it such). He uses it to argue that humans are likely to retain some advantages over machines for the foreseeable future. But in saying this Autor must confront the wave of technological optimism which suggests that advances in machine learning and robotics are likely to overcome Moravec and Polanyi’s paradoxes. And confront it he does. Autor argues that neither of these technological developments is as impressive as it seems and that the future is still bright for human economic relevance.

I think he might be wrong about this (though this doesn’t make the future ‘dark’ or ‘grim’). In the remainder of this post, I want to explain why.

1. Two Ways of Overcoming Polanyi’s Paradox
The first thing I need to do is provide a more detailed picture of Autor’s argument. Autor’s claim is that there are two strategies that technologists can use to overcome Polanyi’s Paradox, but if we look to the current empirical realities of these two strategies we see that they are far more limited than you might think. Consequently, the prospects of machine takeover are more limited than some are claiming, and certain forms of machine-complementary human labour are likely to remain relevant in the future.

I’m going to go through each step in this argument. I’ll start by offering a slightly more precise characterisation of Polanyi’s Paradox:

Polanyi’s Paradox: We can know more than we can tell, i.e. many of the tasks we perform rely on tacit, intuitive knowledge that is difficult to codify and automate.

I didn’t say this in the introduction but I don’t like referring to this as a ‘paradox’ since it doesn’t involve any direct self-contradiction. It is, as Autor himself notes, a constraint on the ease of automation. The question is whether this constraint can be bypassed by technological advances.

Autor claims that there are two routes around the constraint, both of which have been and currently are being employed by engineers and technologists. They are:

Environmental Control: You control and manipulate the environment in such a way that it is easier for machines to perform the task. This route around the constraint acknowledges that one of the major problems for machines is their relative inflexibility in complex environments. They tend to follow relatively simple routines and cannot easily adapt to environmental changes. One solution to this is to simplify the environment.

Machine Learning: You try to get the machine to mimic expert human judgment (which often relies on tacit knowledge and heuristics). You do this by using bottom-up machine-learning techniques instead of top-down programming. The latter require the programmer to pre-define the ruleset the computer will use when completing the task; the former gets the computer to infer its own rules from a series of trials on a large dataset.

We are all familiar with examples of both methods, even if we are occasionally unaware of them. For instance, a classic example of environmental control is the construction of roads for automobiles (or train-tracks for trains). Both have the effect of smoothing out complex environments in order to facilitate machine-based transport. Machine learning is a more recent phenomenon, but is used everywhere in the Big Data economy, from your Facebook newsfeed to Netflix recommendations.
Hopefully, you can see how both methods are used to bypass Polanyi’s Paradox: the first one does so by adapting the environment to fit the relative ‘stupidity’ of the machine; the second one does so by adapting the machine to the complexity of the environment.

2. The Limitations of Both Approaches
This brings us to the next step in Autor’s argument: the claim that neither method is as impressive or successful as we might be inclined to think. One reason why we might think Polanyi’s Paradox is a temporary roadblock is because we are impressed by the rapid advances in technology over the past thirty years, and we convinced that exponential growth in computing power, speed and so forth is likely to continue this trend. Autor doesn’t deny these advances, but is more sceptical about their long-term potential.

He defends this argument by considering some of the leading examples of environmental control and machine learning. Let’s start with environmental control and take the example of Amazon’s Kiva robots. As you may know, Amazon bought Kiva Systems in 2012 in order to take full advantage of their warehousing robots. You can see them at work in the video below.

Kiva robots work in an interesting way. They are not as physically dextrous as human workers. They cannot navigate through the traditional warehouse environment, pick items off shelves, and fill customer orders. Instead, they work on simplifying the environment and complementing the work of human collaborators. Kiva robots don’t transport or carry stock through the warehouse: they transfer shelving units. When stock comes into the warehouse, the Kiva robots bring empty shelving units to a loading area. Once in the loading area, the shelves are stocked by human workers and then transported back by the robots. When it comes time to fill an order, the process works in reverse: the robots fetch the loaded shelves, and bring them back to the humans, who pick the items off the shelf, and put them in boxes for shipping (though it should be noted that humans are assisted in this task by dispatch software that tells them which items belong in which box). The upshot is that the Kiva robots are limited to the simple task of moving shelving units across a level surface. The environment in which they work is simple.

Something similar is true of the much-lauded self-driving car (according to Autor anyway). Google’s car does not drive on roads: it drives on maps. It works by comparing real-time sensory data with maps which have been constructed to include the exact locations of obstacles and signaling systems and so forth. If there is a pedestrian, vehicle or other hazard, the car responds by ‘braking, turning and stopping’. If something truly unexpected happens (like a detour), the human has to take over. In short, the car requires simplified environments and is less adaptive than it may seem.

Autor pours similar amounts of cold water on the machine learning revolution. He is a little less example-driven in this part of his argument. He focuses on describing how machine learning works and then discusses a smattering of examples like search recommendations from Google, movie recommendations from Netflix and IBM’s Watson. I’m going to quote him in full here so you can get a sense of how he argues the point:

My general observation is that the tools [i.e. machine learning algorithms] are inconsistent: uncannily accurate at times; typically only so-so; and occasionally unfathomable… IBM’s Watson computer famously triumphed in the trivia game of Jeopardy against champion human opponents. Yet Watson also produced a spectacularly incorrect answer during its winning match. Under the category of US Cities, the question was, “Its largest airport was named for a World War II hero; its second largest, for a World War II battle.” Watson’s proposed answer was Toronto, a city in Canada. Even leading-edge accomplishments in this domain can appear somewhat underwhelming… 
(Autor 2015, 26).

He goes on then to note that we are still in the early days of this technology — some are bullish about the prospects, others are not — but he thinks there may still be ‘fundamental problems’ with the systems being developed:

Since the underlying technologies — the software, hardware and training data — are all improving rapidly (Andrespouos and Tsotsos 2013), one should view these examples as prototypes rather than as mature products. Some researchers expect that as computing power rises and training databases grow, the brute force machine learning approach will approach or exceed human capabilities. Others suspect that machine learning will only ever get it right on average, while missing many of the most important and informative exceptions… Machine-learning algorithms may have fundamental problems with reasoning about “purposiveness” and intended uses, even given an arbitrarily large training database…(Grabner, Gall and Van Gool 2011). One is reminder of Carl Sagan’s (1980, p 218) remark, “If you wish to make an apple pie from scratch, you must first invent the universe.” 
(Autor 2015, 26)

Again, the upshot being that the technology is more limited than we might think. He goes on to say that there will continue to be a range of skilled jobs that require human flexibility and adaptability and that they will continue to complement the rise of the machines. His go-to example is that of a medical support technician (e.g. radiology technician, nurse technician, phlebotomist). These kinds of jobs require physical dexterity, decent knowledge of mathematics and life sciences along with analytical reasoning skills. The problem, as he sees it, is not so much the continuing relevance of these jobs but the fact that our educational systems (and here he is speaking of the US) are not well set-up to provide training for these kinds of workers.

3. Is Autor Right?
As I mentioned at the outset, I’m not convinced by Autor’s arguments. There are four main reasons for this. The first is simply that I’m not sure that he’s convinced either. It seems to me that his arguments in relation machine learning are pretty weak and speculative. He acknowledges that the technology is improving rapidly but then clings to one potential limitation (the possible fundamental problem with purposiveness) to dampen enthusiasm. But even there he acknowledges that this is something that only ‘may’ be true. So, as I say, I’m not sure that even he would bet on this limitation.

Second, and more importantly, I have worries about the style of argument he employs. I agree that predictions about future technologies should be grounded in empirical realities, but there are always dangers when it comes to drawing inferences from those realities to the future. The simplest one — and one that many futurists will be inclined to push — is that Autor’s arguments may come from a failure to understand the exponential advances in technology. Autor is unimpressed by what he sees, but what he sees are advances from the relatively linear portion of an exponential growth curve. Once we get into the exponential takeoff phase, things will be radically different. Part of the problem here also has to do with how he emphasises and interprets recent developments in technology. When I look at Kiva robots, or the self-driving car or IBM’s Watson, I’m pretty impressed. I think it is amazing that technology that can do these things, particularly given that people used to say that such things were impossible for machines in the not-to-distant past. With that in mind, I think it would be foolish to make claims about future limitations based on current ones. Obviously, Autor doesn’t quite see it that way. Where I might argue that his view is based on a faulty inductive inference; he might argue (I’m putting words in his mouth, perhaps unfairly) that mine is unempirical, overly-optimistic and faith-based. If it all boils down to interpretation and best-guess inferences, who is to say who’s right?

This brings me to my third point, which is that there may be some reason to doubt Autor’s interpretation if it is based (implicitly or otherwise) on faulty assumptions about machine replacement. And I think it is. Autor seems to assume that if machines are not as flexible and adaptable as we are, they won’t fully replace us. In short, that if they are not like us, we will maintain some advantage over them. I think this ignores the advantages of non-human-likeness in robot/machine design.

This is something that Jerry Kaplan discusses quite nicely in his recent book Humans need not apply. Kaplan makes the point that you need four things to accomplish any task: (i) sensory data; (ii) energy; (iii) reasoning ability and (iv) actuating power. In human beings, all four of these things have been integrated into one biological unit (the brain-body complex). In robots, these things can be distributed across large environments: teams of smart devices can provide the sensory data; reasoning and energy can be centralised in server farms or on the ‘cloud’; and signals can be sent out to teams of actuating devices. Kaplan gives the example of a robot painter. You could imagine a robot painter as a single humanoid object, climbing ladders and applying paint with a brush; or, more likely, you could imagine it as a swarm of drones, applying paint through a spray-on nozzle, controlled by some centralised or distributed AI programme. The entire distributed system may look nothing like a human worker; but it still replaces what the human used to do. The point here is that when you look at the Kiva robots, you may be unimpressed because they don’t look or act like human workers, but they may be merely one component in a larger robotic system that does have the effect of replacing human workers. You draw a faulty inference about technological limitations by assuming the technology will be human-like.

This brings me to my final reason, which may be little more than a redressing of the previous one. In his discussion, Autor appears to treat environmental control and machine learning as independent solutions to Polanyi’s Paradox. But I don’t see why they have to be independent. Surely they could work together? Surely, we can simplify the environment and then use data from this simplified environment to train machines to be work smarter in those simplified environments? If such synergy is possible it might further loosen the constraint of Polanyi’s Paradox.

In sum, I would not like to exaggerate the potential impacts of technology on employment, but nor would I like to underestimate them. It seems to me that Autor’s argument tends toward underestimation.

Wednesday, September 30, 2015

Automation and Income Inequality: Understanding the Polarisation Effect

(Previous Entry)

Inequality is now a major topic of concern. Only those with their heads firmly buried in the sand would have failed to notice the rising chorus of concern about wealth inequality over the past couple of years. From the economic tomes of Thomas Piketty and Tony Atkinson, to the battle-cries of the 99%, and on to the political successes of Jeremy Corbyn in the UK and Bernie Sanders in the US, the notion that inequality is a serious social and political problem seems to have captured the popular imagination.

In the midst of all this, a standard narrative has emerged. We were all fooled by the triumphs of capitalism in the 20th century. The middle part of the 20th century — from roughly the end of WWII to 1980 — saw significant economic growth and noticeable reductions in inequality. We thought this could last forever: that growth and equality could go hand in hand. But this was an aberration. Since 1980 the trend has reversed. We are now returning to levels of inequality not seen since the late 19th century. The 1% of the 1% is gaining an increasing share of the wealth.

What role does technology have to play in this standard narrative? No doubt, there are lots of potential explanations of the recent trend, but many economists agree that technology has played a crucial role. This is true even of economists who are sceptical of the more alarmist claims about robots and unemployment. David Autor is one such economist. As I noted in my previous entry, Autor is sceptical of authors like Brynjolfsson and McAfee who predict an increase in automation-induced structural unemployment. But he is not sceptical about the dramatic effects of automation on employment patterns and income distribution.

In fact, Autor argues that automating technologies have led to a polarisation effect — actually, two polarisation effects. These can be characterised in the following manner:

Occupational Polarisation Effect: Growth in automating technologies has facilitated the polarisation of the labour market, such that people are increasingly being split between to two main categories of work: (i) manual and (ii) abstract.

Wage Polarisation Effect: For a variety of reasons, and contrary to some theoretical predictions, this occupational polarisation effect has also led to an increase in wage inequality.

I want to look at Autor’s arguments for both effects in the remainder of this post.

1. Is there an occupational polarisation effect?
The evidence for an occupational polarisation effect is reasonably compelling. To appreciate it, and to understand why it has happened, we need to consider the different types of work that people engage in, and the major technological changes over the past 30 years.

Work is a complex and multifaceted phenomenon. Any attempt to reduce it to a few simple categories will do violence to the complexity of the real world. But we have to engage in some simplifying categorisations to make sense of things. To that end, Autor thinks we can distinguish between three main categories of work in modern industrial societies:

Routine Work: This consists in tasks that can be codified and reduced to a series of step-by-step rules or procedures. Such tasks are ‘characteristic of many middle-skilled cognitive and manual activities: for example, the mathematical calculations involved in simple bookkeeping; the retrieving, sorting and storing of structured information typical of clerical work; and the precise executing of a repetitive physical operation in an unchanging environment as in repetitive production tasks’ (Autor 2015, 11).

Abstract Work: This consists in tasks that ‘require problem-solving capabilities, intuition, creativity and persuasion’. Such tasks are characteristic of ‘professional, technical, and managerial occupations’ which ‘employ workers with high levels of education and analytical capability’ placing ‘a premium on inductive reasoning, communications ability, and expert mastery’ (Autor 2015, 12).

Manual Work: This consists in tasks ‘requiring situational adaptability, visual and language recognition, and in-person interactions’. Such tasks are characteristic of ‘food preparation and serving jobs, cleaning and janitorial work, grounds cleaning and maintenance, in-person health assistance by home health aides, and numerous jobs in security and protective services.’ These jobs employ people ‘who are physically adept, and, in some cases, able to communicate fluently in spoken language’ but would generally be classified as ‘low-skilled’ (Autor 2015, 12).

This threefold division makes sense. I certainly find it instructive to classify myself along these lines. I may be wrong, but think it would be fair to classify myself (an academic) as an abstract worker, insofar as the primary tasks within my job (research and teaching) require problem-solving ability, creativity and persuasion, though there are certainly aspects of my job that involve routine and manual tasks too. But this simply helps to underscore one of Autor’s other points: most work processes are made up of multiple, often complementary, inputs, even when one particular class of inputs tends to dominate.

This threefold division helps to shine light on the polarising effect of technology over the past thirty years. The major growth area in technology over that period of time has been in computerisation and information technology. Indeed, the growth in that sector has been truly astounding (exponential in certain respects). We would expect such astronomical growth to have some effect on employment patterns, but that effect would depend on the answer to a critical question: what it is that computers are good at?

The answer, of course, is that computers are good at performing routine tasks. Computerised systems run on algorithms, which are encoded step-by-step instructions for taking an input and producing an output. Growth in the sophistication of such systems, and reductions in their cost, create huge incentives for businesses to use computerised systems to replace routine workers. Since those workers (e.g. manufacturers, clerical and admin staff) traditionally represented the middle-skill level of the labour market, the net result has been a polarisation effect. People are forced into either manual (low-skill) or abstract (high skill) work. Now, the big question is whether automation will eventually displace workers in those categories too, but to date manual and abstract work have remained difficult to automate, hence the polarisation.

As I said at the outset, the evidence for this occupational polarisation effect is reasonably compelling. The diagram below, taken directly from Autor’s article, illustrates the effect in the US labour market from the late 1970s up to 2012. It depicts the percentage change in employment across ten different categories of work. The three categories on the left represent manual work, the three in the middle represent routine work, and the four on the right represent abstract work. As you can see, growth in routine work has either been minimal (bearing in mind the population increase) or negative, whereas growth in abstract and manual work has been much higher (though there have been some recent reversals, probably due to the Great Recession, and maybe due to other recent advances in automating technologies, though this is less certain).

(Source: Autor 2015, 13)

Similar evidence is available for a polarisation effect in EU countries, but I’ll leave you read Autor’s article for that.

2. Has this led to increased wage inequality?
Increasing polarisation with respect to the types of work that we do need not lead to an increase in wage inequality. Indeed, certain theoretical assumptions might lead us to predict otherwise. As discussed in a previous post, increased levels of automation can sometimes give rise to a complementarity effect. This happens when the gains from automation in one type of work process also translate into gains for workers engaged in complementary types of work. So, for instance, automation of manufacturing processes might increase demand for skilled maintenance workers, which should technically increase the price they can obtain for their labour. This means that even if the labour-force has bifurcated into two main categories of work — one of which is traditionally classed as low-skill and the other of which is traditionally classed as high-skill — it does not follow that we would necessarily see an increase in income inequality. On the contrary, both categories of workers might be expected to see an increase in income.

But this theoretical argument depends on a crucial ‘all else being equal’-clause. In this respect it has good company: many economic arguments depends on such clauses. The reality is that all else is not equal. Abstract and manual workers have not seen complementary gains in income. On the contrary: the evidence we have seems to suggest that abstract workers have seen consistent increases in income, while manual workers have not. The evidence here is more nuanced. Consider the diagram below.

(Source: Autor 2015, 18)

This diagram requires a lot of interpretation. It is fully explained in Autor’s article; I can only offer a quick summary. Roughly, it depicts the changes in mean wages among US-workers between 1979-2012, relative to their occupational skill level. The four curves represent different periods of time: 1979-1989, 1989-1999 etc. The horizontal axis represents the skill level. The vertical axis represents the changes in mean wages. And the baseline (0) is set by reference to mean wages in 1979. What the diagram tells us is that mean wages have, in effect, consistently increased for high skill workers (i.e. those in abstract jobs). We know this because the right-hand portion of each curve trends upwards (excepting the period covering the great recession). It also tells us that low-skill workers (the left-hand portions of the curves) saw increases in the 1980s and 1990s, followed by low-to-negative changes in the 2000s. This is despite the fact that the number of workers in those categories increased quite dramatically in the 2000s (the earlier diagram illustrates this).

As I said, the evidence here is more nuanced, but it does point to a wage polarisation effect. It is worth understanding why this has happened. Autor suggests that three factors have contributed to it:

Complementarity effects of information technology benefit abstract workers more than manual workers: As defined above, abstract work is analytical, problem-solving, creative and persuasive. Most abstract workers rely heavily on ‘large bodies of constantly evolving expertise: for example, medical knowledge, legal precedents, sales data, financial analysis’ and so on (Autor 2015, 15). Computerisation greatly facilitates are ability to access such bodies of knowledge. Consequently, the dramatic advances in computerisation have strongly complemented the tasks being performed by abstract workers (though I would note it has also forced abstract workers to perform more and more of their own routine administrative tasks).

Demand for the outputs abstract workers seems to be relatively elastic: Elasticity is a measure of how responsive some economic variable (demand/supply) is to changes in other variables (e.g. price). If demand for abstract work were inelastic, then we would not expect advances in computerisation to fuel significant increases in the numbers of abstract workers. But in fact we see the opposite. Demand for such workers has gone up. Autor suggests that healthcare workers are the best examples of this: demand for healthcare workers has increased despite significant advances in healthcare-related technologies.

There are greater barriers to entry into the labour market for abstract work: This is an obvious one, but worth stressing. Most abstract work requires high levels of education, training and credentialing (for both good and bad reasons). It is not all that easy for displaced workers to transition into those types of work. Conversely, manual work tends not to require high levels of education and training. It is relatively easy for displaced workers to transition to these types of work. The result is an over-supply of manual labour, which depresses wages.

The bottom line is this: abstract workers have tended to benefit from the displacement of routine work with higher wages; manual workers have not. The net result is a wage polarisation effect.

3. Conclusion
I don’t have too much to say about this except to stress its importance. There has been a lot of hype and media interest in the ‘rise of the robots’. This hype and interest has often been conveyed through alarmist headlines like ‘the robots are coming for our jobs’ and so on. While this is interesting, and worthy of scrutiny, it is not the only interesting or important thing. Even if technology does not lead to a long-term reduction in the number of jobs, it may nevertheless have a significant impact on employment patterns and income distribution. The evidence presented by Autor bears this out.

One final point before I wrap up. It is worth bearing in mind that the polarisation effects described in this post are only concerned with types of work and wage inequalities affected by technology. Wage and wealth inequality are much broader phenomena and have been exacerbated by other factors. I would recommend reading Piketty or Atkinson for more information about these broader phenomena.

Monday, September 21, 2015

Technological Unemployment and the Value of Work (Series Index)

Machines have long been displacing human labour, from the wheelbarrow and plough to the smartphone and self-driving car. In the past, this has had dramatic effects on how society is organised and how people spend their days, but it has never really led to long-term structural unemployment. Humans have always found other economically productive ways to spend their time.

But several economists and futurists think that this time it is different. The type, scope and speed of technological change is, they argue, threatening to put us out of work for good. This raises two important questions. The first is factual and has to do with whether these economists and futurists are right. Is it really different this time round? Are we all soon to be out of work? The second is axiological and has to with the implications of such long-term unemployment for human society? Will it be a good thing if we are all unemployed? Will this make for better or worse lives?

I've explored the answers to these two questions across a number of blog posts over the past two years. I thought it might be worth assembling them together into this handy index. As is fairly typical for this blog, I focus more on the axiological issues, but I will be writing more about the factual question soon so you can expect that section to grow over the coming months.

1. Will there be technological unemployment?

  • Why haven't robots taken our jobs? The Complementarity Effect - This was a more sceptical look at the argument for technological unemployment, drawing upon the work of David Autor. Although I think there is much wisdom to what Autor says, I'm not sure that it really defeats the argument for technological unemployment.

2. Should we welcome technological unemployment?

  • Should there be a right not to work? - This post presents a Rawlsian argument for a right not to work. It is based on the notion that an appropriately just state should be neutral with respect to its citizens conceptions of the good life and that a life of leisure/idleness is a particular conception of the good life.

  • Should libertarians hate the internet? A Nozickian Argument against Social Networks - This post may be slightly out of place here since it is not directly about technological unemployment. Rather, it is about the 'free labour' being provided by users of social media sites to the owners of those sites. It asks whether such provision runs contrary to the principles of Nozickian justice. It ultimately argues that it probably doesn't.

  • Should we abolish work? - This is slightly more comprehensive compendium and assessment of anti-work arguments. I divide them into two broad classes -- 'work is bad' arguments and 'opportunity cost' arguments -- and subject both to considerable critical scrutiny.

  • Does work undermine our freedom? - This post looks at Julia Maskivker's argument against compulsory work. 'Compulsory' work is a feature of the current economic-political reality, but this reality could be altered in an era of technological unemployment.

  • The Automation Loop and its Negative Consequences - The first of three posts dealing with the arguments in Nicholas Carr's book The Glass Cage. This one looks at the phenomenon of automation and two problematic assumptions people make about it the substitution of machine for human labour.

Saturday, September 12, 2015

Sexual Assault, Consent Apps and Technological Solutionism

Sexual assault and rape are significant social problems. According to some sources, one in five American women will be victims of sexual assault or rape at some point during their university education. Though this stat is somewhat controversial  (see here for a good overview) similar (sometimes higher) figures are reported in other countries. For example, in Ireland one estimate is that 31% of women experience some form of 'contact abuse' in their lifetime. The figure for men is lower, but higher than you might suppose, with abuse more likely to occur during childhood.

Clearly we should do something to prevent this from happening. Obvious (and attempted) solutions include reform of legal standards and processes, and challenging prevailing social attitudes and biases. These things are hard to change. But the modern age is also noteworthy for its faith in the power of technology. Many are smitten by technology’s ability to solve our problems, from trivial things like counting calories, contacting friends and navigating around an unfamiliar city, to more complex problems like food production and disease prevention. No problem seems immune to the pervasive reach of technology. Could the problems of sexual assault and rape be the same?

That is certainly the belief of some. In what seems like an almost farcical apotheosis of the ‘is there an app for that?’-trend, two companies have launched sexual consent-apps in the past year: (i) the (short-lived) Good2Go app; and (ii) the more recent We-Consent app. Both are (or were) designed to ensure that the partners to any potential sexual encounter validly consented to that encounter. The rationale behind both being that the presence or absence of consent (and/or reasonable belief in consent) is critical to determining whether a sexual assault took place.

Now I’m all for technology, but in both instances these apps seem spectacularly mis-judged. Criticisms have already proliferated. In this post, I want to take a more detailed look at the philosophical and ethical problems associated with these apps. In doing so, I will suggest that both are indicative of a misplaced belief in the power of technology to solve social problems.

1. What problems need to be solved?
What gives rise to the problem of sexual assault and rape? There are many answers to that question. Part of the problem lies in pervasive and pernicious social attitudes, part of the problem lies in existing legal standards, part of the problem has to do with the procedures used to investigate and adjudicate upon sexual assault cases (be they criminal or civil). It is not possible to do justice to the full suite of problems here, and it is not necessary either since the apps with which I’m concerned are only intended to address a particular aspect of the issue.

The aspect in question concerns the role of consent in sexual encounters. Most legal standards stipulate that the presence or absence of consent is what makes the crucial difference: it’s what turns a mutually enjoyable activity into a criminal one. For instance, the crime of rape (in England) is defined as the intentional penile penetration of the vagina, anus or mouth of another when (a) that other does not consent and (b) the perpetrator does not have reasonable belief in consent (in England, ‘rape’ is a gender-specific crime and can only be perpetrated by a man; there is a gender-neutral crime called ‘assault by penetration’). Consent is thus critical to what we call the ‘actus reus’ and the ‘mens rea’ of the crime.

Consent is primarily a subjective attitude — a willingness to engage in an activity with another — but it is signalled and evidenced through objective conduct (e.g. through saying ‘yes’ to the activity). Ideally, we would like for people to rely upon common knowledge signals of consent, that is: signals that are known (and known to be known etc) to indicate consent by both parties to the activity. But one of the major problems in sexual assault and rape cases is the widespread disagreement as to what counts as an objective signal of consent. Many people infer consent from dubious things like dress, past sexual behaviour, body language, silence, lack of resistance and so on. Oftentimes people are unwilling to have open and explicit conversations about their sexual desires, fearing rejection and social awkwardness. They rely upon indirect speech acts that allow them some plausible deniability. Furthermore, there are a range of factors (intoxication, coercion, deception) that might cast doubt on an otherwise objective signal of consent. The result is that many sexual assault and rape cases break down into (so-called) he-said/she-said contests, with both parties highlighting different potential signals of consent or non-consent, or different interpretations of those signals.

In short, there are significant epistemic problems associated with inferring consent. For present purposes, these problems can be said to break down into two major types:

Social Bias/Awkwardness Problems: These are what prevent people from having open and honest conversations about sexual desires/willingness to engage in sex, and lead them to rely on more dubious indirect signals. These problems occur prior to the sexual encounter itself (i.e. they are ex ante problems).

Evidential Problems: These are what give rise to the he-said/she-said dynamic of many sexual assault and rape trials. Most sexual encounters occur in private. Only the participants to the encounter are present. We rely on their testimony to tell us what happened. But they may disagree about which signals were present or how they ought to be interpreted. Hence, we may lack good, uncontested evidence of what took place (these are ex post problems).

What’s interesting about the two consent apps under consideration here is that they often claim to be directed at solving the first set of problems, but then function in a way that is clearly designed to address the second set of problems. Indeed, despite the protestations of their creators, it seems like the second problem is where they are most likely to have their impact and that impact does not seem to be positive. To see this, we need to consider how they work.

2. How the Apps Work
I am going to focus on two sexual consent apps in this piece. I am not aware of any others, though I haven’t performed an exhaustive search. The first is the Good2Go app, which was released in September 2014, only to be scrapped in October 2014. The creator now promises a re-launch in November 2015, with the focus being exclusively on consent-related education. The second is the We-Consent app, which is actually one of three apps, each designed to address issues surrounding the giving and withdrawing of consent. As far as I am aware, the We-Consent app is still in existence and available for download.

What’s interesting about both apps is how the creators explicitly state that their goal is to address the bias/awkwardness problem. The apps, we are told, are designed to facilitate open and honest conversations about sexual consent. Consider the marketing blurb on the frontpage of the We-Consent website. It tell us that:

Affirmative consent begins with you… talk about “yes means yes” before acting…show respect, ALWAYS DISCUSS mutual consent.

The company’s mission statement says that:

We are the affirmative consent member division of — a group devoted to changing the societal conversations around sexual interactions… the We-Consent Mobile App [is designed] to encourage discussion about affirmative consent between intended partners. 
(Note: the ISCE is the Institute for the Study of Coherence and Emergence)

And the focus on ‘starting the conversation’ is confirmed by the company’s founder Michael Lissack (who also happens to be the executive director of the ISCE) in an interview with the Chronicle of Higher Education:

So what’s the main purpose? The main purpose is to change the conversation. If these apps work the way they should, in a year or two if people go to a frat party, instead of the base assumption being everyone in attendance is available for hooking up, the base assumption will be, if you wish to hook up, talk about it first.

This attitude seems to be shared by Lee Ann Allman, the creator of the Good2Go app. Most of the materials associated with this app have been taken offline after Apple withdrew its approval. But some of the underlying philosophy can be pieced together from media discussions. For example, in a discussion on the Guardian, Allman is quoted as saying that the app should ‘help alleviate the culture of confusion, fear and abuse on campus’. It is also apparent from Allman’s desire to re-launch the product with an exclusive educational purpose. On the webpage we are told that:

Good2Go, a product of Sandton Technologies, will now focus on developing educational materials for college and university students, administrators, and faculty member to help them understand consent...

In many ways, this is laudable stuff. If these apps really could facilitate open and honest conversations about consent and sexual desire, then they might help prevent some incidents of sexual assault and rape. But in terms of their functionality, both apps are also clearly designed to address the evidential problems. They do so by encouraging the potential participants to a sexual encounter to use their smartphones as devices for signalling consent. The signals are then recorded, encrypted, and stored on a database where they can be retrieved and used as evidence in a civil or criminal investigation into sexual assault or rape. This helps to circumvent the he-said/she-said dynamic alluded to earlier on.

The apps perform this function in slightly different ways. Good2Go, in its original form, was a text-based communication app. If you wished to have sex with someone, you would send them a message asking them ‘Are we good2go?’. They would then be given three optional responses: (i) ‘No thanks’ (which would be accompanied by the message ‘Remember! No means No! Only Yes means Yes, but can be changed to NO at anytime!’; (ii) ‘Yes, but…we need to talk’ and (iii) ‘I’m Good2Go’. If the third option was chosen, the app asked the person to gauge their sobriety level, using four options: sober, mildly intoxicated, intoxicated but Good2Go, or pretty wasted. If ‘pretty wasted’ was chosen, the app would not permit consent to be given. Otherwise, everything would be ‘Good2Go’. A record of the interaction would be stored, verifying the identity of the partners by using their phone numbers.

We-Consent is different in that it adopts a video-messaging format, assisted by some pre-recorded voice messages. If you wish to have sex with someone, you open the app and record a message stating your name and the name of your intended partner. You then hand your phone to your partner (or point the video camera at them) and get them to record a response. If they confirm consent through a clear ‘yes’ the app delivers a pre-recorded response stating that the sexual encounter is permissible. The videos are recorded and stored in a double-encrypted form for retrieval at a later date. The functionality here is slightly more straightforward, but conscious of the need to facilitate the withdrawal of consent, the app-makers have created two additional apps, ‘What-about-no’ and ‘Changed-mind’, which allow people to communicate messages of themselves withdrawing consent at a later time. Again, the record of this ‘no’ is recorded and stored on a database for later retrieval. You can watch videos demonstrating how the We-Consent and What-about-no apps work on the company’s webpage.

So, in short, although the creators maintain that the apps are designed primarily to address the bias/awkwardness problems, their functionality is also clearly designed to solve the evidential problems by creating an independently verifiable record of the consent (or non-consent).

3. Why these apps make things comparatively worse
Criticisms of these apps have proliferated online. I share this critical perspective: I think both apps are highly questionable. But I want to conduct a more comprehensive evaluation than has been done to date. I think any evaluation of these apps must do two things. First, it must evaluate them as potential solutions to both sets of problems (i.e. bias/awkwardness and evidential). Second, it must evaluate them using a contrastive methodology. That is to say, it should look to whether these apps improve things relative to the current status quo. That status quo is one that may be characterised by pernicious beliefs surrounding the meaning of different consent signals and significant evidential problems, and in which other proposed solutions to those problems typically involve reforming legal standards (e.g. making it slightly easier to prove non-consent) and improving consent-related education.

Let’s look at the consent apps and the evidential problem first. In a simple sense, these apps do ‘solve’ some of the evidential problems. An encrypted and independently verifiable record of what was signalled between the parties would be a boon to law enforcement. It would represent an evidential advantage relative to the current status quo in which such evidence is not available. But this is obviously a naive way of looking at it. There are at least three significant problems created by these apps that may serve to negate that evidential advantage.

The first is that the apps create decontextualised records of consent signals. I know that is a hideously academic way of putting it, but it captures an important truth. The meaning of a particular signal is always relative to its context (to use some technical terms, the meaning is a function of both pragmatics and semantics). The Good2Go app strips away that context by limiting the record to a series of text messages; the We-Consent app strips away the context by relying on short video recordings of the faces of the potential sexual partners. This is important because there are contextual factors that could render seemingly clear signals of consent practically worthless. The obvious one is coercion. If I record a video message (or tap a button) stating my willingness to consent at gunpoint (with the gun fortuitously invisible to the recording), my signal is worthless. The gun is an extreme example; the same is true if I signal while surrounded by a threatening group of frat boys, or if my friend is being threatened and so on. Other contextual factors that are stripped away by these apps might include degrees of intoxication (though the Good2Go app tried to address this problem) and deception. Eyewitness testimony certainly has its problems, but at least it tends to include contextual information. This facilitates more appropriate interpretations of the signals. The danger with the consent apps is that their verifiable but decontextualised record would be seen to trump this more contextualised eyewitness testimony.

A second problem with the apps concerns the withdrawal of consent. If there is a prior record of you signalling consent (stored on a database and capable of being retrieved at a later time) then the only way to withdraw consent in a legally secure manner is to record another signal of withdrawal. As far as I am aware, the Good2Go app did not even attempt to facilitate such a recording (apart from including the reminder that consent could be withdrawn at any time). The We-consent app does attempt to do so through its companion apps What-about-no and Changed-mind, but both require that the person retrieve their phone in the midst of a sexual encounter and use it to record their withdrawal of consent. Not only is this unlikely to happen, it may be impossible if the other party prevents it (or if the phone is simply too far away).

This brings me to the third problem. By creating a record, both apps may add an air of menace and coercion to sexual encounters that would otherwise be lacking. This could be detrimental to both negative and positive sexual autonomy (i.e. the ability to avoid sex if you don’t want it, and to have sex if you do). If you know that there is a prior record of positive consent, you may be reluctant or unwilling to withdraw consent, even if that is your true preference. Consequently, you might be pressured into continuing with something you would rather bring to an end. Likewise, these apps may have an impact on positive sexual autonomy by making people less likely to initiate sexual encounters they would prefer to have, for fear that they couldn’t bring them to an end when they wished and for fear that there would be permanent and potentially hackable record of all their sexual partners.

For these reasons, I’m inclined to conclude that the apps represent a dis-improvement from the current status quo.

Do they fare any better when assessed as potential solutions to the bias/awkwardness problem? I don’t think so. Although their mere existence (and presence on one’s phone) might direct attention toward the issue of consent — and so might encourage people to take more care to learn their putative partner true desires — they once again seem to create problems that negate any advantage.

An obvious one is ambiguity. This is particularly true for the Good2Go app since it uses a euphemism (‘Good2Go’) as a way of communicating consent. Euphemisms may help people to overcome awkwardness, but they are more uncertain in their meaning than direct forms of speech (e.g. ‘Yes I agree to engage in the following sexual activity X with you’). If you want people to have more open and honest conversations about sexual desire, then it might be better to facilitate direct forms of speech.

This links to another problem. Both apps may stifle the appropriate conversations by giving people limited conversational options. Good2Go gives you only one way of asking for consent and three ways of responding. We-consent is also limited, requiring you to simply state ‘Yes’ or ‘no’ to sexual relations. But these limited options may not allow you to truly express all that needs to be expressed. And because the devices are being used as a proxy for the awkward conversation, they may actually serve to discourage people from having (or seeking) that longer conversation. That said, at least Good2Go tries to facilitate this by including the ‘Yes but…’ option, though as others have pointed out it might have been better if it was simply a ‘we need to talk..’ option.

Another problem with apps of this sort is that they may bias the outcome of any conversation by presuming certain defaults. This criticism has been thrown at Good2Go in particular. The three response options are biased in favour of consent (two out of the three involve affirmation). Given known biases for extremeness avoidance this may result in more people choosing the intermediate option (‘Yes but…’) than is truly desirable. Also, in its measures of sobriety, it assumes that you have to be ‘pretty wasted’ to be unable to consent. The validity of intoxicated consent is contested, but this may err too much in favour of the possibility of intoxicated consent.

There is also the question of how likely people are to use these apps. I may be wrong, but I have a hard time imagining someone whipping out their smartphone and using it to both initiate and record responses to sexual advances. If anything, that would seem to add awkwardness to the situation, not take it away.

Finally, you have to confront the fact that these apps are largely geared towards men (still generally viewed as the ‘initiators’ of sexual encounters). Michael Lissack, the creator of We-Consent, is explicit about this when he describes athletic teams (who he assumes to be male) and fraternities as the target audience for his app. And the evidential functionality of the apps is (arguably) geared toward protecting men from ‘false’ accusations of sexual assault and rape. As such, these apps may largely reinforce patriarchal attitudes towards sexual assault and rape. Empowerment of female sexual agency does not seem to be to the forefront.

On the whole then, it seems like the apps do not represent a contrastive improvement from the existing status quo surrounding bias and awkwardness.

4. Could technology ever solve these problems?
This evaluation of Good2Go and We-consent raises a further question: could technology ever be used to solve the two problems? These apps clearly have their failings, but maybe there are other technological solutions? Maybe. It’s hard to evaluate all the potential possibilities, given the diversity of technology, but if we limit ourselves to information and communication technologies, then I would suggest that it is unlikely that we will find a solution there.

Natasha Lomas — author of a Techcrunch article critiquing Good2Go — suggested that an app including funny conversational prompts might be a better way to overcome the bias/awkwardness problem. You could also improve things by allowing for more diverse responses or by allowing users to generate their own (but then it’s just a text message conversation and we already have apps for that). I suspect, however, none of these messaging systems would be wholly desirable. One problem is that even if you removed the explicit recording and storage aspect, these apps would still create records of the conversation. This might encourage the menacing air I mentioned earlier and discourage conversation. A purely educational app, with no recording of responses and just provision of information, might fare better. This may be what ‘Good2Go’ ends up becoming. The technology in that case would serve as a way to package and distribute the information. But then this would need to supplemented by plenty of ‘offline’ education too.

In terms of the evidential problem, I’m not sure that there is any desirable technological solution. The obvious one would involve more complete video and audio recordings of sexual encounters. They would need to be ‘complete’ to allow for consent to be withdrawn, and they would need to be far more extensive than what can be provided by a single smartphone or wearable tech camera in order to avoid the problem of decontextualisation. But then, if the recordings need to be that complete and extensive, you have a significant invasion of sexual privacy. Dave Eggers imagines something like this happening in his dystopian satire The Circle, and although I am pretty convinced that privacy is dying out, I’m not sure that sexual privacy is something that should be given up in this manner. There is a trade-off that needs to be considered here in terms of positive and negative sexual autonomy. In any event, even a complete recording of a sexual encounter will require interpretation of the signals being sent back and forth between the participants. People may continue to misinterpret those signals in ways that harm victims of sexual assault. You would need to overcome the social biases and prejudices mentioned at the outset to make a dent in that problem.

In the end, I suspect that consent apps are indicative of something Evgeny Morozov calls ‘technological solutionism’. Morozov defines this as an ideology that sees complex social problems as “neatly defined problems with definite, computable solutions or as transparent and self-evident processes that can be easily optimized — if only the right algorithms are in place!” (Morozov 2013). Here, we see the complex social problem of sexual assault, and the associated issues with giving and receiving consent, being defined in a way that renders them amenable to a technological solution. If we simply use the right prompts, direct the conversation using the right menu of options, and then record the output, we will reduce sexual assault and rape.

I don’t think the problem can be solved in that way.

Friday, September 11, 2015

Philosophers' Carnival ♯179

August can be dry month in the philosophy blogosphere. It's traditionally vacation month in Europe and a time when many academics must correct repeat exams and prepare classes for the new semester. But that doesn't seem to have dampened enthusiasm among the internet's diligent coterie of philosophical bloggers. As this month's host of the Philosophers' Carnival, I have the pleasure of collating and introducing some of the best posts from the past 30 days or so.

In no particular order:

  • Terance Tomkow continues his impressive contributions to philosophy blogging with an excellent introduction to counterfactuals. Confused about possible worlds and the ways things could have turned out? Start here.

  • Wolfgang Schwarz investigates the updating of beliefs based on evidence with a clever thought experiment involving a broken duplication machine.

That's it. Be sure to send in your submissions for next month's iteration, and like the Carnival on facebook.