The robot regency

How long is it safe to have our learning left unsupervised?

February 28, 2024 — February 28, 2024

bounded compute
collective knowledge
edge computing
extended self
faster pussycat
incentive mechanisms
machine learning
neural nets
Figure 1: Look, they are totally doing it for themselves.

Perhaps we will as humans always have a niche in the ecology of thinking organisms. Are machines always simply augmentations? Let us consider a short mental exercise about that.

Suppose that you are a parent. Suppose also that you are fabulously wealthy. Suppose further, that you have a beloved child. Let’s call the child Robin. This is the story of that child, and what they need to thrive, and what they need to do to contribute.

Robin is a toddler and you’ve just started to realise that something is not right with her. She is just not as quick on the uptake as children her age. You take it to the paediatrician, and you learn that Robin has tragically developmental disorder. She won’t be non-verbal, not quite. She’ll learn enough words to get by. She’s always going to find learning difficult and she’s never going to be quite as sharp as the other kids, even when she is an adult. She will need to be cared for, for the rest of her life, and sheltered from those who would take advantage of her.

The paediatrician tells you that, in the face of this devastating situation you are not completely powerless. As a prosperous parent, you can spend money on the resources to help Robin learn and develop. You can’t change the fact that she has a developmental disorder, but you can help her to learn and grow to the limits of her potential.

So you start homeschooling her. You hire a tutor to help her with reading and writing she’s never going to be amazing at it, but she can learn some stuff. You can put her smart devices into accessibility mode. You can use bright fonts and simple lettering to keep her attention span engaged. That won’t get her through high-school with university-grade scores, but it will get her reading and writing enough to get around the streets.

You would like her not just to be comfortable but to have a meaningful autonomous lifestyle like the rest of us can, even though her employment opportunities are limited with her skills and capabilities. She still deserves that dignity, right?

The good news is that you can extend her capabilities further in the time honoured fashion, by hiring personal assistants. Okay, so if she doesn’t have the ability to reply, emails or calendarise, that’s no big deal. Many executives don’t do that. No shame in hiring a personal assistant to handle the diaries, and the logistics, and the complicated vocabulary. And so why not hire a few more personal assistants for job specific skills?

What does she want to do, you ask her? “trains! I like trains” she says. That sounds like an interest in engineering. She’s never gonna be great at engineering mathematics. But that you can hire, too.

Eventually, with her small high-functioning team and healthy bankroll, she is able to start up a small engineering consultancy, a successful one, since you have hand-picked her staff well. She is not great at assessing new hires, though, so you also hire her some HR professionals who can help with that.

They start getting large-scale contracts, so some PR and marketing professionals are hired, and business development officers, and contract lawyers. Rob is not really going to rock it out as the face of the company, so she’ll need spokespeople and negotiators as well.

As the firm grows, hires further, and goes into talks with another company that is interested in a merger… you admire all she has achieved. You visit Robin in her office in the city centre. She is in a light, airy corner room with floor-to-ceiling glass windows, bean-bags with fun fluffy things to play with and some board games, and an ipad, upon which she chats to some friends she met in crèche, back in the day.

She doesn’t have to do anything, really. No one asks her to do anything, and she doesn’t ask anyone to do anything. Effort is, in fact, directed to make sure she does not ever exercise her notional authority, because this is a pretty good business, and she is less qualified to run it than literally anyone else here. Robin doesn’t even understand what this business is about, in any but the vaguest sense. Almost every part of it is beyond her comprehension, let alone its relationship to the broader economy. It an intricate part of a system more intricate than she can ever comprehend. As far as she is concerned, it is best understood a great place to hang out on bean bags and order takeaway.


  1. Has this project bought Robin autonomy?
  2. Has it bought her dignity?
  3. Has Robin brought anything to the project?

These are the same questions I have about humans sitting in the notional centre of the web of machine-learning systems, as the machines grow ever more relatively sophisticated, autonomous and incomprehensible.