Last week a good friend of mine was back in town for a few days, and we took the opportunity to catch up over coffee. As usual the conversation turned rapidly to technology and politics, current events, and so on.

This friend spent a couple of years working at Amazon, before moving on to a Machine-Learning startup. We got to talking about our shared relative disillusionment with the current state of Machine Learning, and supposed “AI”, but, you know, the ability to do almost-arbitrary curve-fitting is still pretty cool, right? That’s still useful even if it doesn’t lead to the creation of a Synthetic Subject.

Anyway, my friend observed that in large companies there is a drive to restructure the company and its systems to make data available to these curve-fitting algorithms. Having “big data” is nowhere near enough, the whole structure of the company changes to facilitate this technological shift. The data must be available, it must be comprehensible, it must be possible to act on the outputs from the algorithms.

This instantly reminded me of a passage in Stafford Beer’s “Brain of the Firm”, a classic in Management Cybernetics. Back in the 60′s, Beer lamented that firms were adopting computer technology and automating some processes, but the way they were going about it was entirely wrong-headed. Instead of transforming the firm to match the potentialities of the new technology, they were simply transplanting their existing paper-based workflows into the computers, often producing a resulting flow that was much more fragile than the old manual flow.

For Beer, the question should not be “how should we adopt computers?”, but instead: “Given computers, what is the firm?”. My friend and I had even seen first-hand (at a previous mutual employer) how this same problem repeats in the 21st century. It seems nothing has been learned in a half-century.

We’re faced with a similar question today, even with the limited form of Machine Learning that is actually available. How can we avoid simply transplanting existing structures and flows into the new context?

Given curve-fitting, what is the firm?

– S