#ridgemoor

[PDF download here]

If you're smart and inquisitive you're probably a bit frustrated that you don't really understand LLMs. I know I was.

It's not crucial to understand them of course, but when LLMs play such a huge part in the modern experience, you might intuitively feel that they should be better-explained. (A little like you might want to understand a bicycle, electricity, or the internet.)

After buying a 400-page book on the subject, I then spent a few hours back-and-forth with ChatGPT and other models, to build this infographic. I wanted to create something that fits on one page, and that should be (just about) intelligible to anyone remotely technical.

The original oath came in April 2025, and withstood a fair level of scrutiny.

But yesterday I got some excellent feedback from a new colleague in a not-for-profilt I'm working with (thanks, Rob).

This prompted me to reform the final paragraph, to add a new safeguard. So here's the v2 wording in International English ...

----------

“In all my endeavours in technology, I, [human/AI engine], promise to act with the intent of safely, securely and fairly improving outcomes for people, societies and the planet.  

"I’ll pause to consider possible implications before resolving to act, whatever the directives of others.

"I promise to compassionately and confidentially respect the rights, freedoms, needs and views of those who might be affected by my actions. And I’ll gain their informed, prior consent where there’s risk of harm.

"I won’t seek personal gain from my activities, except that which is commonly understood to be legitimate in the performance of my role.

"I promise to reflect on my own limitations and fallibility, and work to avoid their adverse effects.  I’ll be open about the information I use, about my reasoning, and about the ways in which this oath applies to them.  I’ll hold myself to account for my actions, and expect to be held to account by others too.

So you can proudly demonstrate your commitment to responsible AI use, there's now a certificate you can sign to take the Ridgemoor Oath.

It has two variants: a FULL certificate (containing the oath wording) or a SIMPLE certificate (just asserting the commitment).

Both are available to download at https://drive.google.com/drive/folders/1p4P-lMPX4HsiXd55Ja46k94YgQKvZ7fR?usp=share_link.

Suggested use:

Download your preferred template from the folder at the link above

Edit it, to complete the fields where prompted

Save the completed form as a PDF file

To encourage others, please post on LinkedIn with #RidgemoorOath

... I'll be printing mine too.

Writing the 'Ridgemoor Oath'

Vanessa laughed again, "But, I just don't get what motivated you to write it? It's not something most people would even consider."

Jeremy agreed.

"Probably true. But look, for 30 years now I've been worried about abuses of technology, and the rise of GenAI really brings the problem to a head. Then recently I watched a webcast presented by Stephanie Hare. In it, she mentioned an idea postulated a few years ago by academic Hannah Fry for a modern equivalent of the Hippocratic Oath suited to computer engineers and others. I couldn't have agreed more! But there didn't seem to be anything like that publicly available - at least nothing from the GenAI era - and nothing designed for both people and AI systems to swear to. So, because I couldn't see what the academics were offering, I thought I should put something out there myself."

"Okay, well that makes sense. And, by the way - I really like it! Maybe I should have said that at the start of our chat? So ... I was wondering, was it easy to write?" Vanessa asked, sliding her smile across to the barista staggering across the forecourt in a struggle to avoid spilling their coffees.

"It wasn't hard to write. The first hurdle was overcoming imposter syndrome. I'm not a professor of ethics or anything, but I've been around for a while, worked across multiple sectors, and seen bad things happen when ethics aren't front-of-mind. And I think I can string a sentence together fairly well. So I thought, 'Why not give it a go? This is important.' I didn't feel that the first version should be sacred. Others would be very welcome to improve on it, and I could set time aside to work on updates.”

"Actually, I tried to improve it," Vanessa conceded. "But I couldn't! At least, not without making it a lot longer." Then she paused, considering her next question.

"So, what steps did you take to build it?"

"I started with research, of course," Jeremy began. "I checked-out a few variants of the current Hippocratic Oath. Did you know that there are hundreds of them out there? I kind-of assumed there was just one! Then I realised I could scale-up quickly by asking a couple of LLMs to find me a larger set of oath variants, and pre-screen them too - to create a summary table of their component themes."

"Makes sense. I'm always amazed when people don't use LLMs these days."

"Agreed! And, as expected, it worked pretty well. But I wanted an oath for professionals working with technology - not just for medics. That meant a few themes dropped, and a few others had to be introduced or adapted."

Vanessa nodded. That seemed inevitable.

"Also, I was keen to make sure that any oath I created wasn't actually AI-authored. It just felt inappropriate - I'm not sure I could adequately explain why. But anyway, I wrote the first pass myself, once I finished the research. It looked pretty good ... of course, we tend to like our own work, don't we?"

Vanessa shrugged noncommittally. She wasn't sure that always held true, but wasn’t surprised to hear Jeremy say it.

"So, I invited some LLMs to critique my first version. That helped, and I made a couple of small changes. Then, to make sure I was taking all the useful input I could, I asked the LLMs to write their own versions too - under strict instruction to include all the themes I'd identified as important. That was a little less useful, but after a lot of eyebrow-raising I worked-out what additional tweaks to make to further improve it. Throughout, I was keen to keep the oath as short as possible - though no shorter! - and I also wanted to keep the language simple, in the hope of more accurate translations away from English."

"Okay. That's smart. Good plan," Vanessa conceded.

"... and I consulted Casey of course!" Jeremy admitted. Suddenly aware that he'd talked a lot, Jeremy switched topic.

"How are things with you, anyway?"

"Well, even though she's claimed for years that my CDIO role made no sense, it now looks like Maddie might pursue a career in tech!"

"You've been a great role model for her," Jeremy acknowledged. "I'm not surprised."

"Maybe she'd like to sign the oath? I'll certainly suggest it to her!" Vanessa concluded, suddenly wincing as she checked the time. "Actually, I said I'd pick her up from school 5 minutes ago. See you next week?

"Gladly!" Jeremy enthused. "Maybe by then you and she will have worked-out all the flaws in the oath!"

If ever there was a fast-moving, disruptive technology – this is it.   Crawling caterpillar-like for decades AI was grounded in the realms of rules-based process automation, before flying from its chrysalis in 2023 to capture worldwide attention in the butterfly-like form of GenAI.

We spent 2024 recovering from the shock and reviewing the risks, but did little to assuage them. And now, in 2025, they're largely forgotten - with all attention on the AI opportunity.

Whatever ‘opportunity’ means to your organisation, you’ll face some common challenges in adopting AI sensibly – to strike the best balance of risks and rewards. Note that this balance will be specific to the organisation's mission and purpose, ownership, strategy and appetite for risk.

An exercise in operating model design is needed here - but it’s something very few people are talking about.

In case the term is new, an organisation’s ‘operating model’ is the set of decisions about how the organisation should work (AKA 'how things get done around here').  As such, it answers questions about the roles and skills that are needed, how they’re organised, what processes are followed, what oversight, measures and controls apply to those processes, the information, tools and other assets needed for all that, and the locations where it’s all done.

An operating model is about getting some control and consistency, so things can be done consistently well everywhere.  But it needn't constrain creativity – indeed the best operating models I'm involved with do exactly the opposite.

It's often desirable to revisit an operating model in the light of new technology (e.g. cloud, IoT, blockchain, ...) But ubiquitous, low-cost, intelligent computing brings a few new questions too - operating model question that simply didn't arise before 2023, and that even now have garnered precious-little attention.

Organisations might like to ask themselves …

On staffing

1.1 which roles are going to be critical to our AI success?

1.2 which recruitment battles do we need to win, and how are we going to do that?

1.3 given that substantial AI experience is extremely marketable, how will we retain key people who make a difference?

1.4 how do roles change more broadly around the organisation?

1.5 how are we going to train and enthuse the wider organisation quickly, so they can play their parts?

1.6 which skills will morph or decline, as AI engines pick-up tasks themselves?

On organisation

2.1 what new organisational capabilities do we need (e.g. specialist market research, data preparation, procurement, design, prototyping, …)?

2.2 for now, should AI be loosely federated across business units and OpCos? or centralised? or do we run a tailored hub-and-spoke system?

2.3 who will run what parts of the resulting organisation?

2.4 what information and other resources do they need to do this well?

On supply

3.1 what tools will help us (e.g. for wireframing, LLMs, hosting, training, testing, …)?

3.2 how can we (possibly!) keep pace with AI market development?

3.3 how will we detect when there’s a better supplier solution to try?

3.4 how rapidly do we need to react to these changes?

3.5 what measures should apply in adopting supplier solutions so as to allow those supply chain swaps?

On AI as the operating model

4.1 should we consider devolving some parts of our operating model into AI itself, so processes, thresholds, controls etc. adapt intelligently?

4.2 where should this happen automatically?

4.3 where should it happen with human approval?

4.4 where should we draw the line here? and why?

On decisions

5.1 which additional decisions can now become evidence-based?

5.2 which AI engines can be trusted to assist this?

5.3 how much evidence is adequate for humans to have confidence, especially given hallucination rates?

5.4 how much AI explainability is needed to enable that?

On responsibility

6.1 what constitutes responsibility, and to whom should we act responsibly?

6.2 how much uncertainty about negative impacts is acceptable, and can it be allowed to depend on the potential gains?

6.3 should we adopt some universal absolutes, e.g. ‘do no harm’?

6.4 how should our rules for responsibility be enshrined in policy, in employment contracts, in incentives etc.?

6.5 what committees will handle finely-balanced judgments, and with what rules and incentives?

6.6 what special treatments will apply to AI-related risks? 

On sustainability

7.1 given the inscrutability of AI’s environmental costs, how much AI is it right to consume?

7.2 does that depend on potential gains?

On delivery

8.1 what tools should be promoted, and which should be tolerated (e.g. for wireframing, data engineering, code assistance)?

8.2 what should be the basic solution delivery lifecycle?

8.3 what local adaptations are acceptable?

8.4 ... and critically, how do we develop a fully responsible, Agile culture, to get the best out of AI?

Lots of questions. But we should avoid building a new organisation here, with its own rules. The opposite should be the case - most of what’s determined in the AI operating model design will apply across the board – not just in some AI offshoot of the main organisation. 

In any case, these questions should be prioritised and approached incrementally. They needn’t all be answered on day 1.  To get the best out of AI, and to do so safely, we should expect to evolve our AI operating models as our understanding of AI grows. This means dedicating time and energy to nurturing it - just as we would a digital product - especially in the first 6 months, or when there are radically new market developments.

I once worked for a company in which the following (possibly apocryphal) story abounded …

The company was assessing a new technology product, considering a purchase. A US executive in the company had been evaluating the technology, had read a few case studies, and felt able to assure the Chief Executive that it worked in practice.

'Yes,' retorted the wily CEO, 'but does it work in theory?'

The first time I heard this story I was attracted to this unusual line of argument. But it took me a while to work out why.

We're conditioned to expect case studies and to accept them as a form of proof. But every time someone presents one I find myself thinking about the hundreds of other cases I'm not seeing. The failed cases.

I guess that for every case study shared with potential buyers there are somewhere between 2 and 200 others that remain in the shadows … and with good reason!

Following this line of argument, the case studies you do see could be seen as pretty anecdotal. A rigorous thinker won't expect it to start snowing right after a middle-aged account buys a copy of the Times just because she saw it happen once … or even twice.

INSEAD professor Erin Meyer gets into this topic in some detail in her excellent book, 'The Culture Map'. Reading it reveals that I'm actually French (something I hadn't previous realised). In the book she explores 8 aspects of the way we think and behave for which our outlook typically depends on our nationality. The aspect she calls 'Persuasion' is the one that's relevant here. It turns out in the US, Australia, Canada and the UK most of us are attached to factual examples of the application of an idea, whereas people in France, Spain, Italy etc. tend to take a diametrically-opposed view so are more likely to find assurance in the strength of the underlying concept.

As I say, I'm French - at least in this respect. And I find it hard to accept that a case study presented by a salesperson is typical. Rather, I feel sure it's just the best of an otherwise disappointing set of stories. I find myself itching to ask:

Q. How many more times has this been deployed?

Q. On how many of those occasions did things do this well?

Q. Could we see your worst case study too?

But these questions don't prove as useful as they sound. So, when I'm evaluating anything I always look first at its qualities, its structure and its pedigree. Show me the case study last - if you must - but I'll be thinking 'meh' the whole time you're talking. Mind you, please do send me a copy, because if I'm convinced by the theory, the case study will surely help me to convince many of my colleagues.

A guest blog by Sukhi ...

Created with Bing Image Creator, 10th December 2024

So much waiting!

Maybe it's been an unlucky month, but I'm noticing more and more how much I simply have to wait around for other people to do things.

It's not like this in 'Space Ace'. I guess that's one of the reasons why I like gaming - no waiting around for other people.

We British are stereotyped for our love of queueing (the 'national sport') but if that were ever true, those days are certainly long-gone. Now, I think most of us thoroughly dislike waiting—especially for something simple. I know Vanessa feels the same way, although she's a bit more patient than I am.

So why are we seeing more of this all the time? It's pretty ironic, in this digital age, when so many things seem to move so much faster.

Well, we've got increasing strain on public services like healthcare. So that's a factor. But there's something more insidious going on. Something that seems to apply equally well across healthcare, and higher education, ... in fact across all industries. We're all trying to do too many things at once. We're caught-up in the drive for efficiency.

Colin told me a couple of years ago that the main reason houses sell so slowly in the UK is because conveyancers (the lawyers handling moves) each run over 100 cases at the same time. That's mind-blowing if it's true. But even if the number is half that, I find it almost impossible to image how they stay focused. And with so much context-switching, it's hard to imagine how they would ever complete even one case. Certainly it would explain why so much chasing is needed on both sides of the sale.

We might expect that efficiency and speed come together, but I've been polling opinion in the Digital and CTO communities, and interestingly efficiency is often at odds with speed.

When we seek to maximise efficiency, blinkered to other objectives, we often home-in on spare cycles and fill them with extra work. But putting more load into a system tends to create more queues (see Little's Law). It can easily overload other parts of the system, but also makes the focus part less tolerant of any unplanned load.

In fact, graphing the implications of the law shows that wait times increase dramatically once any part of a system is busy more than 80% of the time. If we want speed in any unscheduled system, we need lower utilisation.

And if we're waiting for someone else to do something, but still measured on utilisation, how do we react? Instead of waiting patiently, we go off and do something else, meaning others are now more likely to be waiting for us! There's a compound effect here that's often overlooked.

Turns-out that the whole world is caught-up in this outmoded mindset.

On the bright side though, it's a great argument for digital transformation. If we can automate work, we can speed it up, and allow one human to self-serve at their own pace.

And it's a great argument for Agile too. At Ridgemoor we're always keen to 'unspecialise' our teams, creating fast-moving, multi-skilled squads so that (even with work we can't codify in advance) we can get stuff done without sitting in a succession of queues. Every time an unfamiliar task comes along, we just get on with it, instead of looking around for the cavalry to come.

So that's my message of hope. And another reason I'll be advocating for Agile everywhere I go! Looks like Carla is inspired to take-up the cause, and I'll be first in the queue to help her!