The relation between machines and judgment in the legal sense is already somewhat actualized in terms of “actuarial” judgment: in parole boards, for example, the “judges” are given a computer-produced risk-probability based on preexisting statistics and the convict’s behavioral pattern. The “judgment” they render is thus really not ought to be considered on a par with legal judgment in the non-actuarial, more common form. Antoinette Rouvroy, the French philosopher of law who has coined the phrase “algorithmic governmentality,” has given an insightful talk on the subject, which I believe is available in written form, entitled “Governmentality in an Age of Autonomic Computing: Technology, Virtuality, and Utopia” (there is also another talk, “Algorithmic Governmentality and the End(s) of Critique,” available on the web for free).
She meticulously analyzes the forms of governmentality that are increasingly dependent upon predictions made from large data-sets by autonomic machines. Examples include risk-assessment of individuals to identify likely terrorists or offenders early on and preempt crime, as well as the already mentioned actuarial decision-making in certain legal settings. The most important issue here as well as in every discussion regarding the data-technologies is the notion of protocol and the “standardization” of the data produced by different profiling resources (these include Facebook as well as state-related polls and profiling projects): not only is the human being to be reduced to a (huge) number of data-fields (name, age, …) processable by “intelligent” and autonomic machines (the latter are defined by their autonomous decision-making; the more common examples are AI enemy players in games and shopping-bots, not to mention the Google PageRank and Facebook’s now (in)famous NewsFeed), but the data thus produced are to be standardized according to protocols and pooled together to form data-mines as “complete” as possible, making for more “accurate” predictions, whether about potential criminality in the “risk society” or the personalization of ads and services.
The decisions and predictions made by these autonomous agents is the result of turning the human being (and the world) into a black-box, whose internal life, intentionality, and states of mind are simply made to do not exist, at least where it matters. They are thus not in any way comparable to human judgment, although their end-result can be made to approach human judgment to determinable degrees. The most concise way to describe the difference between the two is to say that machinic decision-making does not know anything of the “excluded middle” and (perhaps) syllogism in general: it is absolutely singular and does not bypass the universal-individual continuum characteristic of judgment.
There is a most subtle Occasionalism at work in the cluster-concept of AI. It is well-known that Descartes first comes up with the idea of Occasionalism (for anyone reading this and not familiar with the term, roughly put it is: a cosmo-theological doctrine that attributes all causation to an omnipotent God who constantly intervenes to ensure the workings of the world) is humans’ lack of knowledge of their bodies’ movement: how does one move one’s arm without knowing how, without even knowing what physiological processes are at work? For a philosopher with the highest regards for the cogitative capacities of humans, acting without knowledge is a scandal; the idea of an omniscient God intervening between my will (to move my hand) and its fulfillment (my hand moving), acting as the cause sets things right again, for Descartes’ God knows the what and the how of every act.
Now let us fast-forward to the present. In the preface to Ethem Alpaydın’s Introduction to Machine Learning, 2nd Ed. we read that one of the two principal reasons for using machine learning (as a sub-field of AI, it can be implemented in different fashions, from simple transfer functions to neural networks to genetic algorithms) is the inability of humans to “explain their expertise.” Language recognition is one among such cases. In a sense, the creation of intelligence can be seen as the final piece of what Freud termed the “Prosthetic God” humans are forever building. Some philosophers object to the use of the term “intelligence” for merely algorithmic behavior; but what about instances of artificial behavior for which no human-devised algorithm exist? I would also like to ask what forms of intentionality are valid in cases where we don’t know what we know, meaning of course the tasks we perform without knowing how to perform them.
There is also another point that I’d like to discuss, and that is the issue of the smart fire-detector (Jürgen Lawrenz had asked whether a “smart” fire-detector that calls your cellphone in case of a fire can be deemed as really intelligent or capable of judgment). Fully agreeing that such a gadget can in no way be deemed intelligent, I still feel it is important to remember that machines and technical objects were unable to respond to changes in their milieu before Wiener’s feedback-based cybernetic devices. The very possibility of “learning” even learning as bare memory accumulation (but there is also change in behavior) is of recent origin. All this aside, I too believe that the most important feat that distinguishes the human consciousness from the (perhaps) intelligent machine is judgment, something the latter is incapable of, but also something that it renders more or less obsolete.
Defined in terms of subsuming the particular to the universal, that is defined in loosely Kantian terms, judgment is a human process, and a costly one. Plato was somewhat right in disregarding the Sophists for their attempts at creating shortcuts (“short-circuits”, in Stiegler’s terminology), claiming that universal knowledge was essential to understanding and judgment. Such knowledge could only be amassed by years of study and learning and as such a very costly thing. This is one of the reasons why the earlier AI models based on human judgment failed to yield any efficiency. We must not forget that machines are built to be efficient, and the human process of judgment cannot be a successful model; Instead, we get the less-than-judgment machines; the cybernetic machines operating at a level below representation and knowledge.