[Me & X] How do you go about working on what to invest in?
X: How do you go about working on what to invest in?
Me: Do you mean societal versus market benefit or are you talking about an organisation?
X: Organisation.
Me: In which case, I'd start with a topic such "Operational capability in Space". Then I would gather a group of people together with experience of that topic (that's the hard bit). Then I ask them to write down what matters on post-it notes. Then we group those post-it notes into themes. Then we vote on a few of the themes to map (anywhere between 3 to 6) and then we divide into teams and map those themes, for example "In orbit operations". We then use those maps to explore what is changing and what are the potential investment areas. Then we collect all the investment areas and prioritise in terms of value. Then we simply do a bog standard cost benefit comparison.
X: That's a lot.
Me: Not really. It's a simple process of explore, categorise, map, prioritise and then cost/benefit. There's nothing special about it.
X: How long does that take?
Me: About 12 hours or possibly much faster depending upon the degree of experience. I normally run these fairly gently as 6 x 2hour sessions over a period of three weeks. The time consuming part is it can take several months to gather the people.
X: Can you use an AI to do this?
Me: You can use a LLM/GPT as a participant. I have a judgement system (I built ages ago) which has multiple different agents using different models to explore topics in the form of a Hegelian dialectic with a jury process. The interesting thing about that is the answers are nowhere near as important as the questions that appear during the debate. In essence, this is what is happening here. The maps are simply the vehicle by which we get to the real value which is in the questions generated through the discussion between the people. The process essentially identifies and distils those questions into the form of a cost / benefit analysis.
X: Can't you get an AI to find the interesting questions?
Me: This is why I've been running and experimenting with my judgement system for so long. Alas, I've not found a successful way of getting the LLM/GPT to identify something as an interesting question. That doesn't mean it can't rank questions based upon some criteria that I define or highlight contradictions in questions or even generate questions of which some are interesting BUT ... identifying what is interesting is still very much a human skill that depends upon lived experience and context. This is also why you need to gather people with experience of the topic.
X: Is this how investment is done in most organisations?
Me: No. Most organisations run on stories and investment is often by executive whim with oodles of work done to justify a decision that has already been made.
X: Could AI help here?
Me: You'll probably get better results by simply asking the LLM/GPT to choose. At least it'll be based upon training data.
Originally published on LinkedIn.