I'm sitting with coffee in the lobby at UCLA's Institute of Pure and Applied Mathematics. At my table is a smiling Swiss professor of mathematics. He tells me he is in his last professional years before reaching mandatory retirement age. And me? Where am I from? From here. Los Angeles. No institution. Why am I at there, at IPAM? General interest. What sort of general interest? I ask if he happens to work on the problem machine learning has of not leaving a record of how the machine is learning. In fact, he does. He uses machine learning to complete, based on probabilities, partial body scan images. And the problem, I add, is that sometimes the probabilities assumed in completing the image gaps meet with a case of improbability, and produce false results, and because we don't know the rules used by the machine to calculate probabilities that have been constantly adjusted with experience, we don't know what sort of errors to look for. Is that an accurate statement of the case?
- You're well informed for the general public. Are you some kind of journalist?
- I like to think about things. Lately I've been thinking about the incredible, literally cannot be believed, lying of our American president and his political party. I think his kind of lying has a lot in common with what happens in machine learning. Imagine every lie is a new image used to teach an artificial intelligence, making in the case of your work the completing image supplement more accurate by comparing its results with the actual object. What if the president, and members of his party, are repeatedly making claims about the world - secret enemies are destroying us! - and those claims are answered ritualistically by the president speaking to his followers: we are fighting back, recovering our strength now that we have woken up to our danger, and we see more and more the real weakness of our enemy. Our waking up is our correction, our education confirming the probabilities that define our danger and our capacity to face up to it. This cycle of education, scripted in advance, with no relation to truth in the world is on display in how Facebook selects what is allowed to go on users' timelines, singling out posts that are predicted to make them angry and afraid, teaching them to respond with increasing certainty and intensely to the posts they are sent. Russian false accounts in the 2016 presidential elections similarly arranged for the world to confirm the probabilities of passionate response, increasing hysteria of posts confirming increasing fears. Like you mathematicians in your machine learning world, the politicians in the midst of these educational cycles have no record, no access to the constantly adapted calculations of probability done in their training of each other, telling stories to each other of their perilous situation. Lies produce an artificial learning environment rather than function as a description of the real world; the real world doesn't enter into the calculation at all. But it can: the identified enemy can be put in a concentration camp, in conditions in which their physical state begins to conform to the previously false taught probability of unnaturalness, sight of which leads to further adjustment of probabilities, finally justifying torture and murder of the unnatural enemies within. Or again, when a political party educate themselves in the ideology of small government / no regulation, the result is the flourishing of groups willing to use power in the form of money or violence to take control of government institutions; with this accomplished they more efficiently indoctrinate the people in small government / no regulation ideology allowing them to take further control. Employment agencies find employing older people to be of improbable benefit to employers, causing older applicants stay unemployed for long periods, a further sign of their improbable benefit making it more likely they remain unemployed, and remaining unemployed they become poorer and have to move to areas with zip codes residence in which is linked to additional improbable benefit to employers.
- Vicious cycles.
- Yes. So you see we have two levels or stages here. First, the construction of group ideology of 'us and them', learned without knowing how. And then the application of this ideology to the world, changing the world, adjusting the probabilities in the direction of confirmation of the ideology. At first the enemy doesn't look so different, but there is no certainty in life, all is probability, and anyway in no time at all reality is adjusted to fantasy. You know Chomsky's theory of language acquisition? That speaking is not learned by working out the probability of particular words referring to particular things, but through listening for inborn models in our minds of word relations or logic. If the American president's people look at him and see only a statement of probability,* cues to adjustment of social relations in imagination, and also potentially in practice, a normally functioning human being looks at another with love in which there is no doubt about the beauty seen. Beauty is knowledge of something good. Error seems to enter into love because knowledge is incomplete. Maybe you look on your loved one with consciousness of her beauty and she at that moment is terribly upset and you don't see it, immersed as you are in your experience of love, only her anger at your apparent inattention wakes you up to your mistake. So what do you do? You look back on your past with her, seeking other things you know about her aside from her general lovableness. You look for models, that is, habits of perception: you did this and she did that. You look for a model that fits her present dissatisfaction. In machine learning, there is no mechanism that brings together such a history of decisions and adjustments with present decision to be made, but with us there is: the body, with its habits of perception, habits gained in the past leading us to our present perception. What you see now includes your history, is what you have learned to see developing habits of perception. There is no certainty in love, in individual moments of contemplation, but there is certainty in memory of travelling the path to love, in the history of your repeated attempts to apply models to the world. Its certainty is like the certainty of science: that models are made and improved is real, their truthful relation to the world does not end but is supplemented, improved. You could be mistaken in the timing of what you see in the one you love, but it is a fact that your love, errors and all, is teaching you to love more and more, is helping you stay close to the beauty of the world. 'All things that are, are with more spirit chased than enjoy'd.' I'm getting off the track here. To summarize: the lies of the president and his followers are social acts, first constructed to secure power relations in imagination between the president's group and all outsiders; then the lies are brought to the world, training the world to exist in accord with their self-trained selves, the at first dissonance of trained view of the world to the world as it really is being allowed to sink into the silence and uncertainty of education by probabilities rather than models. Well, you listened to me to the end. What do you think?
- Very enjoyable.
Further Reading:
Selfless
Prostitution & Torture
Technology & Magic
________________________
* See: The Love Of The People
P.S. 'To know and not to know, to be conscious of complete truthfulness while telling carefully constructed lies, to hold simultaneously two opinions which cancelled out, knowing them to be contradictory and believing in both of them, to use logic against logic, to repudiate morality while laying claim to it, to believe that democracy was impossible and that the Party was the guardian of democracy, to forget whatever it was necessary to forget, then to draw it back into memory again at the moment when it was needed, and then promptly to forget it again: and above all, to apply the same process to the process itself. That was the ultimate subtlety: consciously to induce unconsciousness, and then, once again, to become unconscious of the act of hypnosis you had just performed. Even to understand the word 'doublethink' involved the use of doublethink.' - George Orwell, 1984
