Provisional version, not for quotation. Do only refer to the final published version: M. Hildebrandt, Profiles and Correlatable Humans, in: Christoff Henning, Nico Stehr and Bernd Weiler (eds.), Knowledge and the Law. Can Knowledge be Made Just?, New Jersey: Transaction Books 2006

Are profiles justiciable?
1 Introduction
Participating in the European NoE FIDIS (Future of Identity in Information Society: www.fidis.net) I was triggered by the use of the term knowledge society. Is there a crucial difference between an information society and a knowledge society, in the sense that they describe the world from a different perspective? If so, (how) are the two related? Could it be that our knowledge depends on information, while at the same time the question what counts as information depends on our knowledge? Could it be interesting to describe non-humans and their relationships in terms of knowledge and information (does a gene know how to produce a specific type of protein, depending on the input of specific information (epigenetics); does a bird know how to whistle to seduce a possible mate, depending on the input of specific information, like for instance a change in temperature or sunlight) (Van Brakel 1999): 7/15)?

For centuries we have been used to think of both information and knowledge as objects of reflection or at least consciousness. As if we know only what we know to know. As if information only counts as information if we are aware of its presence (criticised by (Mead 1959/1934; Merleau-Ponty 1945; Polanyi 1966; Ryle 1949)). If we allow ourselves to recognize knowledge and information in the world of nonhumans, this might change our perception of an information or knowledge society and it might also shed some light on the advanced identification technologies that will soon impact our sense of self and our understanding of the validity, relevance and legitimation of knowledge. I will focus on profiling as the most advanced and comprehensive technology (and practice) of identification. As profiling is the automated generation of a multiplicity of evershifting profiles that are the precondition for Ambient Intelligence and advanced risk-assessment, it constitutes a very interesting nexus of information, knowledge and risk-assessment.

In this paper I want to discuss some implications of profiling practices and attempt to answer (or at least raise) the question if and how the information and/or knowledge generated by these technologies can be made justiciable.

In the second paragraph I will explain what is meant by the term profiling practices. After that, in paragraph 3, I will discuss purpose and effects of the type of knowledge they produce. Par. 4 will explore the way the law tries to deal with this type of knowledge, raising some questions about the effectiveness of the law as it tries to fit this kind of knowledge into the present legal framework. In par. 5 I will explore the possibilities for the law to make profiling practices justiciable. I will claim that for data protection legislation to take effect specific technological design and an effective fair trial are preconditional for the justiciability of profiles. The fair trial in fact presumes and creates a position from where humans as linkable data subjects with a sense of self can make knowledge-constructs like profiles justiciable.

2 Profiling practices
One could say that profiling practices are a way to generate knowledge from data (Custers 2004): 17-20, KDD-model: knowledge discovery in data bases). This knowledge consists of patterns or correlations between data(sets). To give a crude example: if we have a data base with data on the colour of your eyes and a series of transactions at your local grocery, a correlation may be found between the colour of your eyes and your preference for certain products (eye shadow would be predictable, but other, more unexpected correlations might turn up). In marketing this way of generating correlations (exploratory research) has been used since the ’70-ties (regression analysis). The tools for discovering such correlations have improved exponentially, bringing down the transaction costs for searches in huge databases enormously.

The whole process of profiling is often described in 4 steps: data collection, data preparation, data mining and interpretation (KDD-model: knowledge discovery in data bases) (Custers 2004): 17-20). These are not obvious steps, as they all require a certain infrastructure and restructuring of information in a way that fits a database. As Lyotard remarked in 1984 ’Along with the hegemony of computers comes a certain logic, and therefore a certain set of prescriptions determining which statements are accepted as ’knowledge’ statements’ (quoted by (Van Brakel 1999):3/15). Data collection requires that certain types of behaviour (buying stuff, visiting certain places, surfing on specific internet-sites, speaking with colleagues, friends, clients, suppliers) are not only observed but also recorded and stored. This is a major difference with previous observation of behaviour in public surroundings as this was not often recorded and stored (Lessig 1999): 143; 150-151). [Example: if you buy bread around the corner and pay with cash, this will be observed by others, but your ’transaction’ will not be recorded, stored and linked with other data that were recorded, stored and aggregated; money is still anonymous today, but think of plans to trace banknotes by adding rfid tags]. Second, these data have to be aggregated into databases. To profile clients of a bank, all transaction data that come into different departments of the bank, have to be stored as transactions of the same client; to profile customers as high or low spenders it is necessary to locate as many different transactions possible as transactions of this one person. This is why integration of different databases becomes interesting (and profitable), leading to a market for data and databases. Third, data mining - the process of generating correlations or checking on the outliers of established correlations - requires creativity (to select data to be correlated and the algorithms that may produce correlations) and practical wisdom to be aware of spurious correlations that do not have any independent explanatory power.

The fourth step is the central issue of profiling: the emergence and interpretation of correlations. The point is that a correlation does not mean anything until it is interpreted: it does not necessarily indicate reasons nor causes. Of course it can be used as a hypothesis, claiming a causal link. The interesting thing about profiling is, however, that it does not start from a hypothesis that is than rigorously tested/falsified/verified, but that it generates correlations without necessarily even being interested in causality or human reason. Example. It is possible to observe online behaviour like the relative speed with which certain keys are touched on the keyboard. It is also possible to profile this behaviour to such an extent that a person can be recognized as the same person on the basis of her typing behaviour. This identification technology is not interested in the causes of your typing behaviour, nor in your reason to wait between the a and the i. It is interested in identifying you as the same person over a period of time (personalizing), and perhaps, linking the information with that of others, being able to predict certain traits or behaviour as correlating with your typing behaviour (group profiling).

Profiling thus does not necessarily build on the traditional methodologies of for instance the social or natural sciences (Custers 2004): 56-58), looking for explanations in terms of causes or understanding in terms of meaning. Interpreting the correlations is not usually done to construct theoretical knowledge about society or persons (sociology or psychology), but to decide on the next step to be taken. Data mining technologies are instrumental for decision-making processes (marketing: which type of persons will be identified as potential customers, and to which approach will they yield; crime control: which type of persons will be targeted as potential suspects to further investigate; anti-terrorism, immigration: which type of persons will be identified as potential terrorists or illegal immigrants). Actually correlations seem to create new meaning. Insignificant personal data may turn out as highly significant data, and/or for instance correlate with sensitive data that are protected by privacy legislation (using such seemingly insignificant data is called masking, (Custers 2004): 57). Traditional protections may not work here, and even fighting traditional conceptions of science may be an irrelevant exercise. This is about generating knowledge, in the sense of patterns in data that are interesting and certain enough for a user, not about science. And perhaps the impact of these kind of technologies and practices is more fuzzy, more precise, more implicit and more worrying that the impact of scientific knowledge.

3 Purpose and effects of profiling practices
Western societies are thought of as information and knowledge societies, but also as risk societies. Not because other types of societies are not prone to risk, but because while manipulating some of the risks of ’nature’, we seem to fabricate others (Beck 1992; Shklar 1990). Looking into the relationship between risk, information and knowledge at the level of non-humans again, it should be clear that information can indicate certain risks if we have the knowledge to read it like that. At the same time it is obvious that the use of explicit knowledge and information can itself be a risk, as our modernist risk society demonstrates. Profiling technologies, which are focused on processing information to produce knowledge, also create risk in a broad sense. To localize this we will investigate the purposes and effects of profiling practices (Custers 2004): 74-78).

The purpose of profiling is selection, implying the wish to include certain objects or persons and to exclude others. This in itself is everyday business, life depends on it. Organisms select certain traits to fit in their environment and sometimes to fit their environment to their needs. While neo-Darwinists understand the survival of the fittest as justification for power play, one could say that Darwin demonstrated that evolution in the end rewards those organisms that fabricate the best fit with their environment (Stengers 2003). In that sense evolution is a constant process of profiling: taking in and processing information to decide the next step (profiling as risk-assessment and screening for opportunies).
Moving to the production of explicit knowledge we can look at genetic profiling in molecular biology and epidemiology, which aims to select genes that correlate with disease, hoping to find therapies to prevent or cure such disease and/or to be able to target certain medication or surgery to the type of patient that will most likely respond. But, of course, in human society, selection can be legitimate or illegitimate; it can for instance exclude people from equal opportunities on grounds that we find unjust. As Lawrence Lessig describes, social hierarchies require information to discriminate between different social ranks. With increased mobility the costs of acquiring such information rose, because of the difficulties in tracing people. Thus many old hierarchies broke down. Profiling changes this, empowering profile users to re-establish inequalities (Lessig 1999):155). Selection, like technology, is neither good nor bad, but it is never neutral. It impacts the lives of those that are selected and thus calls for justification.

Another way to look at the purpose of profiling is to describe it as prototyping: enabling a user (business enterprise, police, immigration policy makers, doctor) to make decisions on the basis of a prototype, a knowledge-construct that filters our perceptions and expectations. Prototyping can be described as a psychological process (Canhoto 2004), but also, at an epistemological level, as a Vorurteil in the sense of Gadamer: without some form of prototyping we would be flooded by meaningless information. The problem is - of course - that prototyping is close to stigmatisation (Goffman 1963);(Hudson and Bramhall 2005). In more concrete terms, it may be the case that profiling of persons as possible terrorists leads to unacceptable legal consequence, violating the presumption of innocence and the right to defend oneself against such legal consequence.

A third way of looking at the purpose of profiling is warning persons of the risk they run and thus confronting them with knowledge about themselves they had no access to (genetically determined disease, for instance). This confrontation will enable them to take measures, but it will also impact their sense of self in an existential way. Profiling may offer them choices they would have lacked otherwise, but profiling may also reveal secrets of the self that force a person to reconstruct her identity (Hudson 2005a).

A fourth function profiling may take on is customisation. As a result of targeted advertising a person may be confronted only with those advertisements that should interest him, considering his past behaviour. If we move on to the advance of Ambient Intelligence this perspective becomes pervasive. Ambient Intelligence, the combination of pervasive and ubiquitous computing with intelligent devices or learning technologies, can respond to your wishes before you become aware of them and restructure your environment in tune with the anticipations you carry under the skin. This sounds like heaven - and like hell. It can reduce the feedback you get from your environment, as you begin to live in a world of your own making. The diversity of unexpected and unwanted confrontations with others, whether human or nonhuman, could be diverted by your intelligent agent that ’knows’ - on the bases of your past behaviour - what you would probably want and expect.

We can add here that the line between customisation and manipulation is a thin one. To quote Lawrence Lessig ’When the system seems to know what you want better and earlier than you do, how can you know where these desires really come from. (...) profiles will begin to normalize the population from which the norm is drawn. The observing will affect the observed. This brings us to the last paradoxical effect of profiling. While profiling seems to individualise (customise) your environment, it may in fact de-individualise your way of life. Both group profiling and personalisation judge your needs, expectations and desires on the basis of past behaviour, building a well-fitted and unusually comfortable cage from which escape will be nearly impossible, precisely because profiling becomes ever more ubiquitous and intelligent.

4 Data Protection and informational privacy
If the knowledge produced by profiling practices entails exclusion, stigmatisation, confrontation, customisation and even de-individualisation, the question is how to constrain these practices in order to make the knowledge they produce just. The traditional means to constrain application of new technologies is to provide good practice guidelines for industry and/or legislate on the matter. In the last decades alternative instruments have been elaborated, of which technological design is perhaps the most interesting. In the field of data mining so-called PET’s (privacy enhancing technologies) have been developed, that combine a measure of linkability (a necessary precondition for profiling) with anonymity and/or pseudonymity. To fine-tune one’s level of privacy and security (in terms of linkability) per contact would be unthinkable, so to make these PET’s work one needs a digital agent (or identity management device, IMD) that is programmed to choose the desired level for you (Agre and Rotenberg 2001; Clarke 1994; Lessig 1999).

As to the legal constraints, from the ’70-ties onward attempts have been made to regulate the collection, storage, exchange and use of personal data (Bennett 2001). The purpose of this regulation is of course not the protection of data in itself, but the protection of the persons that can be harmed by the use of those data. Data protection legislation is a tool to protect the informational privacy of persons or groups. This legislation is generally based on a set of principles, first developed in the 1974 American Privacy Act, later expressed in the (non-binding) guidelines of the OECD and numerous national statutes on data protection (see also the EU Directive 95/46/EC). The principles can be summarised as (1) the collection limitation principle, stating that collection of personal data should not be unlimited; (2) the data quality principle, stating that personal data should be correct, complete and up-to-date; (3) the purpose specification principle, stating that the purpose for which personal data are collected must be specified, and that they may only be used for that purpose; (4) the use limitation principles, stating that disclosure or use for other purposes is only allowed in case of consent of the data subject or on the basis of the authority of the law; (5) the transparency principle, stating that the data subject should be able to know about the collection and storage of personal data, its purpose and the identity of the data controller; (6) the individual participation principle, stating that a data subject has the right to erase, rectify, complete or amend her data; and finally (7) the accountability principle, stating that the data controller should be accountable for complying with these principles.

This all sounds very just. However, as we have seen, the essence of profiling is the ubiquitous process of collection, storage, aggregation and processing of data in databases. The extent to which such ubiquitous processes take place and develop into forms of ambient intelligence that interact with our personal agents on a real time basis, seems to render data protection with its dependence on traditional legal tools totally inadequate. The amount of decisions taken by intelligent devices, software programs and personal agents seems to invalidate traditional legal concepts like liability for individual actions, transparency of and access to personal information, limitation of the use of data for specific purposes and, especially, consent as the basis for collection, storage and processing of personal data. The actions of electronic devices that (will) impact our lives cannot easily be reduced to actions of a specific human agent; the amount of data being collected and the low transaction costs for the data controller make transparency and access virtually impossible; as the essence of profiling is linking data and discovering for what purpose the emerging profiles could be used, not much can be expected from attempts to prohibit linking and using data for other purposes; especially as so many daily transactions require consent of a kind that can hardly be taken serious, considering the consequences of refusing consent and the impossibility to go over all the different conditions on which consent is given.

So, unless these principles of informational privacy can be built into the personal digital agents (PDA) that manage the exchange of data, and unless these privacy-enhanced PDA’s are widely used by data subjects (us, the humans), profiling will simply disable data protection legislation. The logic of profiling (ubiquitous linkability, unobtrusive correlatability) is at odds with the logic of data protection (providing citizens with the means to refuse and/or direct their linkability). The one builds on invisibility, the other on transparency.

This brings in a new problem. The risks of illegitimate (unjust?) profiling could be managed by developing architecture for these technologies that prevents ubiquitous transparency of citizens, while promoting transparency of data controllers and data users. The same objective is promoted by legislation that confirms the principles of fair information practices. These two instruments to both enable and constrain the production of profiles could work, if they became part of the socio-technical framework that rules our actions, if they became habits, social norms, rules, sound expectations. If not, the enforcement problem that arises is beyond control. So the problem is that legislation by itself or privacy enhancing technological design in itself cannot make the knowledge that will be produced just, as long as the use of these technologies and the practice of these principles is not part of our shared sets of habits/common sense/socio-technical infrastructure.

5 Can Profiles be Made Just; Are Profiles Justiciable?
What is new about profiling, as a type of knowledge production: (1) the scope of the data that can be recorded, aggregated and researched at a reasonably low cost is enormous; the low transaction costs have as a consequence that profiling practices often do not involve extrapolation of samples to populations, but on searching an entire field (Custers 2004); (2) profiling is typically ubiquitous and unobtrusive, which implies that the invasive character of profiling seems absent (Lessig 1999): ; (3) the low cost of searching an entire data base leads to exploratory research, generating correlations instead of starting with a theory and then deducting hypotheses that can be falsified/verified (Custers 2004; Scott Armstrong 1970); (3) profiles can impact our lives in a number of ways without us ever being aware of the fact that we were included in or excluded from certain opportunities or risks on the basis of a profile; (4) profiles ’know’ things about ourselves we don’t know (Hudson 2005a; Rose 2003):86-87.

The simple fact that profiles can, do and will affect our lives in positive but also in negative ways, raises many questions. On the one hand these questions concern the impact of false positives and false negatives: how to organise the possibility of resistance against knowledge-claims that base selection of an individual on the profile of the group she is supposed to be a part of (group either in the sense of a community, or in the sense of a category)? If certain attributes of the profile are non-distributive, it cannot be concluded that all members of the group share the profile. On the other hand the questions raised concern the impact of knowledge about a person that stigmitises this person or an entire group of persons; knowledge that confronts a person with information about herself she may not want to know; knowledge that is used for targeted servicing leading to customisation and de-individualisation. Can the law make these knowledge-constructs justiciable in relation to both types of questions (questions about fitting people into wrong categories and questions about fitting people into categories)?

What could it mean to make profiles, as knowlegde-constructs, justiciable? And is that the same as making them just? In my information on this conference I was confronted with two titles: ’is knowledge justiciable?’ and ’can knowledge be made just?’. As a lawyer I feel rather attracted to the idea of making knowledge justiciable, because it is a more modest claim than making knowledge just. Justice is something to be strived for, certainly by the law, but to claim justice is - perhaps paradoxically - easily
a bridge too far (Derrida 1994). The difference between law and morality is that the first can settle disputes, while the last probably causes them. However, morality is also a part of the law: the disputes it causes at the heart of the law nourish the vitality, complexity and responsiveness of the law. The morality of the law can even be said to be embodied to a large extent in the fact that law makes things disputable (justiciable): the difference between rule of man and rule of law may be that rule of man thinks of law as a set of commands of a sovereign; while rule of law implies the relative autonomy of the law as a shared perception of the normativity of our world (Geertz 1983). So to make knowledge just would be claiming the archimedic foothold from where to dictate true knowledge, while to make knowledge justiciable is rather the opposite. It would mean that we bring knowledge claims within the jurisdiction of a fair trial that allows opposing knowledge claims to plead their case with regard to a specific case.

So, if we want to make the knowledge claims of profiling practices justiciable, what does it take?

From the perspective of democracy and rule of law, my answer is three-fold: legislation that constrains technological design and application to fit the demands of a democratic constitutional state; a fair trial that makes the knowledge claims of profiles justiciable as they have legal consequence and the concept of the legal person that can question the construction and application of profiles as they impact humans as correlatable data subjects and as persons with a sense of self.

As discussed in par. 4 data protection legislation cannot stand alone. To work, it needs to be implemented at the level of technological design, while also constraining both technological design and application. But for this to happen we must move beyond the confines of administrative legislation. The integration of ’fair information principles’ into both technology design and its use in everyday practices, cannot take place by means of top-down instruction (administrative law) only. The establishment of data protection commissioners can supply some surveillance on dataveillance, but to make profiles justiciable the extent to which this type of knowledge impacts individuals and groups must become transparent at the level of those individuals and groups. Settings like the fair trial could provide a framework for making the knowledge claims involved justiciable. As discussed in other work the fair trial offers an interesting ’ideal type’ or ’good practice’ for the testing of knowledge claims (Hildebrandt and Gutwirth 2004). The combination of the interrelated principles of an independent and impartial judge, a public hearing, equality of arms, presumption of innocence, contradictory proceedings and the principle of immediacy provides a setting that allows lay persons to have the last word on competing expert knowledge claims (Hildebrandt 2004). In fact, some social research claims that jury trials contain an interesting setting for rethinking both the construction of knowledge on matters of concern and representation (participative instead of aggregative: (Wakeford 2002).

In this paper, however, I will end by focusing on the concept of the legal person, that is preconditional for the democratic constitutional state and for the fair trial that embodies the constitutive principles of the rule of law. The concept of the legal person affects two things. First it provides a position from which actions can be made justiciable if claimed to constitute harm (and knowledge construction is an action). This effect of the construct of the legal person is closely intertwined with the ’fair trial’.

Second the fact that the law thinks in terms of the legal person provides an artificial position that shields the human as correlatable data subject and as a person with a sense of self from complete determination. The concept of the legal person refers to the Greek ’persona’, which was the mask behind which actors hid themselves when they performed their drama’s. The mask indicated the role they played. When a subject takes the stand in a court of law, the construct of the legal person prevents confusion between the role the subject takes on the one hand and the embodied indeterminate subject on the other hand. One could articulate this in another way by saying that the concept or construct of the legal person allows the human person as a correlatable data subject with a sense of self to resist the profile (the correlated human) and/or the way it affects her life. The importance of the legal person is that it protects the essential underdetermined nature of the human person against the desire of the state and the market to categorize its subjects in such a way that they fit the logic of state bureaucracy and/or market imperatives. By insisting on the correlatability of humans against knowledge claims concerning correlations, the legal person confirms and protects what Deleuze would perhaps have called the virtual character of the correlatable human. For Deleuze the virtual is not opposed to the real but to the actual. He relates the possible to the real, in the sense that the difference between them is merely that the one exists while the other does not. When the possible becomes real is does not change, only becomes existent (it is already determined). The virtual however is already real, it already exists. Whether, how and when the virtual actualises is, however, not entirely determined. The virtual - present in the actual - is underdetermined. Thus being correlatable implies being virtual. It means that the correlations that will be proliferate cannot all be determined in advance. As Pierre Lévy writes about the actual seed that contains the virtual tree: ’starting from the constraints that constitute the seed as this particular seed, it has to invent the tree, to coproduce it together with the circumstances it will meet’ [my translation] ((Lévy 1997).
In that line realisation must be understood as the occurrence of a predefined possibility (or probability (Stengers 1997a): 27, footnote 10), while actualisation is the invention of a solution required by a complex problematic. Pierre Lévy goes on to ask the question how we should understand virtualisation (the inverse of actualisation). He answers that question by saying that ’virtualisation of an entity consists in discovering a general question to which this entity relates; forcing the entity to move in the direction of this interrogation and pushing it to redefine the actuality of its starting point as an answer to a particular question’. Actualisation goes from problem to solution; virtualisation passes from a given (actualised) solution to the problem, ’thus making the instituted distinctions fluid, augmenting degrees of liberty, creating a productive vacuum’. In fact, Lévy writes, ’virtualisation is one of the main vectors of the creation of reality. The law, by instituting the legal person, creates a distance between the correlatable human and the correlated data subject, thus creating a specific type of freedom: challenging the actual, virtualising the profile back to the questions it presumes, thus creating the possibility to claim the irrelevance of the actual profile.

Two examples to clarify:
§ imagine a group profile has been constructed that attributes certain properties to members of the group. Since some or all of the properties are non-distributive in this group, categorising members by means of the profile will produce false negatives and false positives. If a person takes action in a court of law claiming that the profile is not applicable in her case, she in fact challenges the move from the probable to the real.
§ imagine a person is profiled as a psychopath, because she has the properties that define a psychopath. Though a true positive, she may want to question the knowledge construct that lies at the basis of her profile and/or she may want to question the relevance of categorising people on the basis of Hare’s checklist. She may claim that the profile Hare constructed answers the wrong question, looking only for the probabilities that define an ensuing reality, instead of virtualising actual traits and thus opening the way for new actualities (Stengers 1997b): 147).

6. Concluding remarks
Profiling practices entail a specific form of knowledge construction, used to provide assessment of risks and opportunities. On the basis of these knowledge constructs decisions are made that impact the life of the data subjects and their sense of self. This in itself is not a new fact, biology and information science meet on this issue. However scope and scale of the collection, aggregation and processing of data that is made possible by pervasive, ubiquitous and intelligent computing impact power relations between those that are profiled and those that control or use profiles.
For democracy and rule of law it is important to understand the human person as a correlatable data subject with a sense of self, that cannot be defined entirely in terms of probabilities. This is precisely why it is important to make profiles justiciable and also indicates the importance of the concept of the legal person. The legal person empowers the actual person (the correlated human) to claim her virtual identity (the correlatable human), to resist determination by probabilities. In the end it will be the access to the fair trial that actually opens up the possibility to challenge profiles and their implications.

Putting it that way the concept of the legal person makes knowledge claims of profiling practices justiciable: (1) by claiming a mistake in categorising the possible, thus reaching the wrong conclusion about reality; (2) by claiming a right to virtualisation (back into the future?), resisting the idea that a profile holds the definitive determination of the becoming of the actual. In the first case one may claim to be selected on false grounds; in the second case one may claim to be selected, prototyped, confronted, targeted on irrelevant or illegitimate grounds.

Agre, Philip E. and Marc Rotenberg. 2001. Technology and Privacy: The New Landscape. Cambridge, Massachusetts: MIT.
Beck, Ulrich. 1992. Risk Society: Towards a New Modernity. London: Sage.
Bennett, Colin J. 2001. "Convergence Revisited: Toward a Global Policy for the Protection of Personal Data?" Pp. 99-125 in Technology and Privacy: The New Landscape, edited by P. E. Agre and G. Bramhall. Cambridge, Massachusetts: MIT.
Canhoto, Ana; Backhouse, James. 2004. "Constructing categories, Construing signs - analysing differences in Suspicious Transaction Reporting practice." IS-CORE 2004 - 4th Annual SIG IS Cognitive Research Workshop.
Clarke, Roger. 1994. "Human Identification in Information Systems: Management Challenges and Public Policy Issues." Information Technology & People 7:6-37.
Custers, Bart. 2004. The Power of Knowledge. Ethical, Legal, and Technological Aspects of Data Mining and Group Profiling in Epidemiology. Nijmegen: Wolf Legal Publishers.
Derrida, Jacques. 1994. Force de loi. Paris: Galilée.
Geertz, Clifford. 1983. "Local Knowledge: Fact and Law in Comparative Perspective." in Local Knowlegde. Further essays in interpretive anthropology, edited by C. Geertz. New York: Basic Books.
Goffman, Erving. 1963. Stigma. Notes on the Management of Spoiled Identity. Englewood Cliffs, NJ: Prentice-Hall.
Hildebrandt, M. 2004. "The Trial of the Expert: épreuve and preuve." in Paper for the Conference ’Putting expertise to the test’ Brussels 21 October 2004.
Hildebrandt, M. and S. Gutwirth. 2004. "The proof of the pudding is in the eating: testing expertise in citizen panels and courts." in Paper for the 4S and EASST meeting 28th August Paris. Brussels.
Hudson, Barbara. 2005a. "Secrets of Self: Punishment and the Right to Privacy." in Privacy and the Criminal Law, edited by E. Claes and A. Duff. Antwerp Oxford: Intersentia.
Hudson, Barbara and Gaynor Bramhall. 2005. "Assessing the ’Other’." Br J Criminol:azi002.
Lessig, Lawrence. 1999. Code and other laws of cyberspace. New York: Basic Books.
Lévy, Pierre. 1997. "Sur les chemins du virtuel." vol. 2005.
Mead, George H., edited, with introduction by Charles W. Morris. 1959/1934. Mind, Self & Society. From the standpoint of a social behaviorist. Chicago - Illinois: The University of Chicago Press.
Merleau-Ponty, Maurice. 1945. Phénoménologie de la perception. Paris: Gallimard.
Polanyi, Michael. 1966. The Tacit Dimension. Garden City, New York: Anchor Books.
Rose, Hilary. 2003. "The Commodification of Virtual Reality." in Genetic Nature/Culture. Anthropology and Science beyond the two-culture divide, edited by A. H. Goodman, D. Heath, and M. S. Undee: Universtity of California Press.
Ryle, Gilbert. 1949. The Concept of Mind. New York: Barnes & Noble.
Scott Armstrong, J. 1970. "How to Avoid Exploratory Research." Journal of Advertising Research:27-30.
Shklar, Judith. 1990. The Faces of Injustice. New Haven: Yale University Press.
Stengers, Isabelle. 1997a. Cosmopolitiques. Tome 1. La guerre des sciences. Paris: La Découverte / Les Empêcheurs de penser en rond.

 . 1997b. Cosmopolitiques. Tome 7. Pour en finir avec la tolérance. Paris: La Découverte / Les Empêcheurs de penser en rond.

 . 2003. "How to understand the kind of power associated with genes by contemporary biology?" published on the internal portal of Imbroglio.be.
Van Brakel, J. 1999. "Telematic Life Forms." Techné: Journal of the Society for Philosophy and Technology 4:http://scholar.lib.vt.edu/ejournals/SPT/v4_n3html/VANBRAKE.html.
Wakeford, Tom. 2002. "Citizens Juries: a radical alternative for social research." Social Research Updat 37.