The Incubator

This took an age to write, started on Saturday and finished on Sunday. I’m wrapped up in blankets on the sofa with a weird illness. It came on Friday afternoon and kept me up in the night, so I’m low energy. Can’t work out if it’s a bug or an autoimmune response, though probably the latter. But I don’t want to let it take the shine off a good week.

As a twist on usual proceedings, I’ll start with personal notes first and move on to work later. If all you’re interested in is work bits, scroll down to Work.

Life stuff

Electric Dreams and sake

Last Saturday we saw the Electric Dreams exhibition at Tate Modern. It was pretty engaging and drew a line-in-the-sand in art history, but there was way less focus on the Internet and cybernetics than the introductory texts made out.

Despite that we had a fun time and there were several playful pieces. One piece moved iron filings around a board twice an hour, which meant a group of strangers were stood together for several minutes, waiting for the thing to turn on. A couple of people got tired of waiting and were repelled away.

There was also a segment in a book about a computer that had been programmed to write haiku. I’ve been reading about haiku, waka and other Japanese travel literature recently, so this piqued my interest. It could produce perfect haiku in form, but it completely missed the function: spontaneous composition in response to a moment. A subtle irony, possibly the point.

After that we went to Kanpai to drink sake brewed in London. It was pretty tasty though not as clean and mineral as the sake we had in Azumino.

Lazy Sundays

The last couple Sundays have been surprisingly delightful: no TV, no radio, just sitting in the lounge reading the paper. Maybe I’m turning into my grandfathers.

Dr. Strangelove

On Monday we saw Armando Iannucci’s and Sean Foley’s adaptation of Dr. Strangelove featuring Steve Coogan. As you might expect, it was very good. To be honest, I think Coogan fluffed one of the main jokes but it didn’t detract from the enjoyment at all. It was nice to see Giles Terera again too.

A few nods to Musk and Trump at points, and the overwhelming sense that this might all come back around some time soon. Lots of laughs, anyway.

Before the show we popped into Le Garrick which isn’t as good as I remember. Having said that, two courses for £17.75 can’t be sniffed at, that’s rare in central London.

Other watchings

This latest series of Wolf Hall is good, especially the tightrope Cromwell must walk between serving his king and enabling a despot. It’s good on class too, showing how a man from a ‘common’ background will still be othered by those from better-off backgrounds.

’Tis the season for Muppets Christmas Carol, which no doubt I’ll watch this week. Although it’s probably better suited to Beigemas – an annual festive gathering of our friends, accompanied by a range of frozen treats from Aldi, Lidl and Iceland.

Perfect Days was recommended to me in the week, though, and that will a good midweek mood-lifter.

Running

I’ve not been running since early November when I did a parkun in Tokyo, and I hadn’t been running for a while before that. Strava’s yearly round-up says I ran ~40% less this year compared to last year. That’s all down to the hip injury which sometimes feels like it’s still present.

Hopefully the long rest will have done my joints some good, and I can get back to running in January – albeit a little less fit than I was. I’d like to do another trail marathon and I miss my Sunday long runs, but I’ll have to find a physiotherapist who can give me some guidance on the malady.

Work

Q4 planning

I spent a few hours pulling together the objectives and initiatives teams had proposed to me, then compared those to the objectives on our programme roadmap. All moving in the right direction, which is wonderful. There’s a few tweaks I suggested back and a couple points where psd and I asked for more clarity, but it was much less effort than in previous quarters.

This is good, it shows that teams have a good grasp on their problem domain. There’s some operational work that needs better defining, but I think that can be helped along with a little bit of training.

Incubator for AI

In the summer we looked at creating data from existing documents but it took a lot of time and repetitive effort. It’s the sort of process that might lend itself well to automated extraction, tested and refined using evals. So our director dropped the Incubator for AI a line to see if they could help.

On Wednesday we went to see their early proofs-of-concept. They came up with a range of other ideas, for other use cases (and I liked their prioritisation method), but this one has a lot of legs. They’ve only spent a few days pulling something together, so there’s lots of iteration to happen. And then we have to bear in mind that it’s only a prototype that has come out of a skunkworks. There will be necessary product and design work to understand who it’s valuable for, how to make it usable, how we build it to be maintainable, and whether it’s viable from a copyright perspective.

I’m not going into all the details here as it’ll no doubt become one or two official blog posts. They made nice comments about my positions on generative AI and the work fits with those. Doubly nice.

Outcomes not outputs

I’m going to mangle some good words Mike said in a chat we had, but mostly in the hope it’ll encourage him to write the clear, prescient thing he said out loud.

When we talk about the platform or the problem we’re working on, we can end up talking about the data and why the data is important too much. But if we spent more time talking about what the data does or could enable, and then showing that, we’d get more buy-in.

He’s right, of course. Outcomes over outputs. Benefits instead of features.

GitHub Projects

Oh yeah, I forgot to mention last week that I had a chat with the product manager for GitHub Projects. I wanted to ask what had happened to task lists as we found it handy on GOV.‌UK Design System last year.

It surfaced issues in their Markdown pipeline, so they’ve started again and changed how to create sub-issues, types and search using filters. It’s pretty cool! I’ve got a job to update our roadmap and make the related work more visible, so I’ll be using those features to do it.

Better doesn’t just happen

After reading the ‘test and learn’ thing from Pat McFadden, Rachel Coldicutt’s rightful criticism of the harmful semantic drift in the speech, Martha Lane Fox’s debate on increasing digital skills in civil service, plus this article on necessary and unnecessary organisational friction, I got thinking about the change that really needs to occur. Better doesn’t just happen, it arrives by changing things.

These are pretty scrappy notes, not a perfect plan. Just thinking out loud and inviting comment.

The first thing that needs to be changed is that there is no workforce strategy for digital skills in government. CDDO made a new brand last year but there’s no evidence this has lead to an increase in the right types of skills in the right places. I doubt a new brand is enough.

The second thing that needs changing, that would likely fall out of the workforce strategy, is a plan for training and developing skills in line with what we’d expect people to know 10 years ago. GDS Academy died, few courses have appeared since, and the Service Manual is a scant playbook for actually developing digital services. So let’s help people learn ‘the basics’ and execute it well.

The third thing that needs changing is to create a proper environment for ethical innovation in public services and policy design. Despite a few atomic examples dotted around, the structures, funding and governance of public services don’t bring together transformative capabilities with the appropriate guardrails to work through hard societal problems in new yet careful ways.

The fourth thing that needs changing is the removal of all the old cruft. There’s a lot of bureaucracy and accountability reporting that needs to be redesigned. It can be incredibly useful, it just needs to fit into the new ways of working.

I don’t think we’ll get proper reform and better public services without updating the operating system. GDS 1.0 brought about better public services, for sure, but I don’t think we’ll improve on those (or make the rest better) without deeper changes. I’m looking forward to seeing what the Digital Centre come up with.

Bookmarks

Here’s the best 3 bookmarks from the week.

  • The GenAI App Step You’re Skimping On: Evaluations, 10 mins. Basically describes adding automated tests to check whether generative AI is giving the outputs you expect. Simple to implement in some cases, like when you’ve got existing data as a control group. More complex in other scenarios when you might chain together LLMs to create inputs and assess outputs – when presumably you need evals for your evals?

  • Churchill’s call for brevity, 2 mins. Winston Churchill emphasised the importance of brevity in government communications during the 1940s. He urged officials to avoid jargon and keep reports concise to save time. Nice blog post from the National Archives in 2013 that points to good principles for working in the open.

  • One for all and all for none, 2 mins. A good example of how slapping an app on the front of broken services won’t work. In this case it’s a very real problem with the NHS App. Points to Rachel Coldicutt’s lines on consequential risks in ‘Words Matter’ (referenced earlier).

Other bookmarks

· Weeknotes

0 replies, 0 reposts, 3 likes