Machine Learning amplifies existing signals in the data and also does so when we ourselves are quite unaware of that bias. Identifying biases and engaging them is, or has largely become, an art of post facto, retrodictive engagement. The futures technology generate are so often unacknowledged as recursively self-propagating compression waves, signals as an encoded history or memory of the past. Is it a function of entry-level human narcissism that we (all, at least generally) fail to see that a moving, competitive and commercial frame of reference must always in some regards invoke opacity into the plausibly Machiavellian incentives that percolate through the methods of its own operation?
If the primary signals are of difference and relationships inflected by power and anything other than collaborative and equitable control, we seem to be bound to a spinning wheel of accelerating technical and technological metamorphosis that will not, perhaps can never, see the causal factors or engage them such that they will remain plausibly unintelligible from within the communications matrix of its own normative assumptions.
I don’t think any of us can deny the presence of bias and inequity or that simple existence in such commercial, competitive systems is in some measure to disavow ourselves of the dichotomies and discontinuities that inhabit their core. Language, technology and communication have a mischievous way of eliding causal mechanisms.
When the caterpillar in Lewis Carroll’s tale asked Alice to explain herself, she was quite nonplussed. There is a sense in which Machine Learning and the bundled panoply of dissonant biases it revivifies represent a very similar question, as posed by technology to us.
As one of the most clever people I have met in the last few years put it: “When something is broken, you know that there is a system.” In regards to bias in proxy neural technologies, I have a few intuitions but I think we have not yet identified an actual, underlying and causal systemic or communications system substrate. I think that we need to look much deeper into system dynamics to disentangle the threads and engage this issue of bias with anything like the gravity it requires.
Daedelus Kite 