You have spent the past 15 minutes in the back room of Café 66 on Fort York Boulevard in Toronto in utter fascination, staring at jar after jar of the green substance ever so beautifully packaged and presented on the shop shelves. You are now informed by the hip, young man behind the counter that “the seventh gram is on the house.” Indeed, this shop, at least on the surface, does not appear categorically different from any other shop wherein goods are exchanged for money. And that is how you might treat the upscale marijuana dispensary south of Toronto—like any other establishment. If you pay for your weed using your credit card, however, you have committed the most dreadful carelessness of the age of Machine Learning: you have provided the machine with a “link.”
A very good friend of mine took me on a tour of the “café”, and as she spoke at length with the shopkeepers about the properties of the high caused by different strains of weed, I tried to grasp the true nature of the place I was in. The roof was covered with closed-circuit cameras. Above the only door to the room with the product was a monitor showing the video feed of the camera installed just outside the room. Soon I also noticed the weak sound of a buzzer and realized why we had to ring in and wait for the mechanical click of the door lock.
My friend received a pitch–black, childproof bag, inside which individual pitch–black, childproof bags contained the different strains that she had purchased. She then reached for her purse and paid in cash. I had seen her use her credit card in shadier places. As we left, I could not stop myself from asking her if she was really concerned, for example, that news of her purchase might reach her insurance company. “There are strict privacy laws in this land, you know!” I said. “I would like to believe that you are right,” she replied. “However, that is only the most obvious way that this purchase can cost me dearly.” She then continued, under her breath, “and probably the most benign.”
Imagine a list of a few hundred million people, and imagine that linkage has been made between the credit card purchases of everyone on the list and the “unfortunate events” that have afflicted those individuals. An “unfortunate event,” in this context, can refer to anything from being involved in a car accident, to declaring bankruptcy, to getting a divorce. Now, imagine that privacy measures have been taken into account and that purchases are anonymized. In other words, given any individual of interest, one can only know that this individual spends her money on products and services offered by businesses A, B, and C; one does not know what line of business these establishments are in. For example, A might correspond to Café 66, B might correspond to Istanbul Café, and C might correspond to the gas station at the corner of Dundas and Church. So, any customer of Café 66 who has used her credit card in that premises would be linked with business A; however, no one knows what business A actually represents. Can this situation be considered “hazardous”?
Let’s assume that marijuana usage is correlated with risk-taking. If that is, in fact, the case, it is possible to imagine that the rate of occurrence of “unfortunate events” is significantly higher within the customer base of Café 66. This is where the link between Jane Doe and A becomes valuable to the machine for deriving an inference: because “unfortunate events” are assumed to be more likely between individuals linked with A, and although every other piece of information indicates that Jane Doe is a good driver, a careful spender, and in a happy relationship, for example, Jane’s link with A points to a heightened probability of future trouble. Therefore, Jane Doe is to be handled cautiously. When she applies for a mortgage, she is considered a higher-risk individual, and her insurance premium may rise ever so slightly.
The scenario depicted above is not the worst case, however. The situation becomes more concerning when the more cautious of the risk-takers start taking notice of the activities of the silent silicon surveyors and change their payment method in Café 66 and similar establishments. Such an imaginable and, frankly, optimal strategy will then strengthen the significance of a link with A. In other words, those linked with A are the ultimate risk-takers; they are the ones who take more brazen risks. And so increases the penalty of the mistake of paying at Café 66 with your credit card.
During his stay in Zion, Neo went on a midnight stroll with Councillor Hamann. While observing the marvelous machinery of the Engineering Level, Hamann queried Neo on his understanding of the concept of “control.” Hamann had trouble accepting the fact that life in Zion was only possible because machines tended to the needs of the occupants of the last human city on planet Earth. To him, it was only ironic that other machines were digging in in order to destroy the underground covenant. In response, Neo opined that one controls an entity only as long as one can turn that entity off.
The time for turning Machine Learning off has long passed, and the justification for doing so is even more distant. The situation is, in fact, even more ironic than Matrix Reloaded. One may argue that, today, one needs to start thinking the way the machine does in order to survive. And this, ironically, completes the circle. Humankind aspired to replicate its own cognitive abilities in order to delegate menial tasks to its creation. It now appears, however, that man is forced to adopt the machine’s way of “thinking” in order to survive the reign of its own creation.
Acknowledgment: This text has been proofread by A. I wish to thank her.