Setting up your own Advisory Board created with Artificial Intelligence - Part I
Jesper Andersen
Artificial Intelligence | Communication Measurement & Evaluation | Thought Leadership | Strategic Communication | Public Relations
Having an Advisory Board gives you access to expertise and help with decision-making, opens doors and facilitates introductions to interesting people, and stimulates professional growth.
But how much of that can be synthesized using Artificial Intelligence?
Welcome to another edition of The A.I. Exploratorium - the newsletter where I test out use cases and application of A.I. in professional communication.
I started working on my Advisory Board idea almost five weeks ago (alongside other projects) and due to a very busy work schedule I am only just now ready to press the launch button. It will be another two weeks (next issue) before I have anything to tell about whether the experiment was a success or not. ????
In this issue of The A.I. Exploratorium we dive into:
I hope you enjoy it! ????
The purpose and role of an Advisory Board
Unlike a board of directors (who represent the interests of the owners or shareholders), an Advisory Board is an assembly of people who volunteer their time and expertise to help a manager run the company better and their advice carries no legal responsibility.
Typically, having an Advisory Board gives access to, enables or promotes:
Notable differences between a human and an A.I. Advisory Board
An A.I.-powered virtual Advisory Board may be able to synthesize parts of what a normal, human Advisory Board does. In some areas - like data analysis - it may even surpass the capabilities of humans, creating new and unique possibilities. But there are also limitations.
Humans draw from lived experiences and tacit knowledge and can adapt to unexpected situations using intuition and creativity, while A.I. board members rely on the training of the A.I. model used plus any provided training data.
This also means that any bias inherent in the model and training data will likely repeat itself in the advice of A.I. advisors.
A human Advisory Board will be able to recollect past discussions and take those into consideration going forward, while an A.I.-powered Advisory Board will face much greater difficulty 'evolving' from its starting point - unless its training data is constantly updated or modified to reflect what happens in the company.
The major advantage of an A.I.-powered Advisory Board, of course, is the 24/7 availability. Some managers will also appreciate that they don't have to actually know people with particular skill sets who are willing to serve on their Advisory Board.
Recruiting or replacing members of an A.I.-powered Advisory Board is as simple as adding or removing a new persona (at least in theory - I will know more once I test it during the next two weeks ??). That means you can supplement your Advisory Board with new areas or types of expertise at almost a moment's notice, giving you extreme flexibility because onboarding new members is practically instantaneous.
Why do I want to create an Advisory Board with A.I.?
For a few months now I have been using A.I. to give me feedback on my thinking and how I have been feeling. Initially, I was pleasantly surprised by the depth and perspective, but as the conversations dragged on, I found that the A.I. gradually lost focus and started repeating the same things.
I am guessing this is because of the limited context window - at some point the A.I. tends to forget 'where the conversation started'.
Memory introduced by ChatGPT in September alleviates part of this problem, but I still found myself wanting an A.I. that knew more about me, my background and challenges I am facing.
In the past I have built Custom GPTs with a single purpose or persona / advisor role, and I guess that I could build an Advisory Board as individual Custom GPTs and 'draw them into the conversation' using the '@' functionality in the chat. However, that would not enable multiple personas to present their points of view all in one coherent conversation.
So, I decided to attempt to build a Custom GPT with a range of personas - all with individual backgrounds, expertise, quirks, personalities and opinions.
My work process
Throughout the process, I have been working with two of OpenAI's models: The regular ChatGPT-4o model and the o1 Preview model with reasoning capabilities.
You will notice that unlike my normal newsletters, this edition of The A.I. Exploratorium is pretty much devoid of screenshots of conversations with the A.I. That is because at this point the conversations take up the equivalent of 300+ pages of text in Word. ????
Presenting that in a meaningful way with screenshots was simply not possible, which is why I am summarising my actions instead.
Getting started
I started out just having conversations with ChatGPT about what I wanted to create and getting its feedback and input. It suggested a plan along these lines:
ChatGPT also gave me an idea of what I would probably have to provide to successfully build an A.I.-powered Advisory Board:
When I asked about the training data I should upload to the Custom GPT and how to best structure it (to make it easier for the A.I. to navigate), I was given a helpful list of how ChatGPT would suggest making room for dynamic business data that would change over time, while keeping other documents intact (more about this further down this article).
Setting up the roles and personas of the Advisory Board
Populating the Advisory Board with relevant personas was the first major challenge. I started out by having a conversation with ChatGPT in which I described my company, how I would like it to grow and evolve, and the major challenges I face (as I see it).
Based on this information, ChatGPT provided me with a basic outline of 5 different profiles / roles along these lines:
The first five personas were all directly related to the goals and challenges, I had stated, so I asked ChatGPT if there was something obvious that I was missing? If it had any suggestions for additional profiles / roles that I might benefit hearing from?
That resulted in seven additional suggestions, most of which were not relevant or practical for one reason or another.
For instance, it suggested that I should include Networking and Partnerships Connector, which would be a great idea if it were a human Advisory Board member. But the A.I. cannot synthesize 'a well-connected insider' who can 'facilitate connections and open doors'. (I wish it could! ??)
In the end, after several rounds of revisions and adjustments, I ended up with six personas that I am going to use in my initial Advisory Board:
In a normal Advisory Board you would likely have a lot of overlap between the expertise of the members with additional individual areas of deep knowledge.
In this case, I have created six very individual personas - both in terms of experience, personality, approach to their role and task on the Advisory Board. I figure that one representative can argue a point just as well as six, so there was no need to create 'dominating opinions' in the group (but I could be wrong about this of course? ??).
Training Data file overview
By far the most time-consuming task in setting up the A.I.-powered Advisory Board was producing the necessary training data files to upload to the Custom GPT.
I was surprised how much I could include - if I cared to dictate or type it all. My company has operated for 11 years now and during that time I have experienced many ups and downs, have pivoted to new areas, learned new skills and explored new markets.
Documenting all of that seemed a nearly impossible task to me, but once again it helped simply having conversations (written or audio) with ChatGPT about what I was trying to accomplish and how I could reach my goal of launching the Custom GPT in the most efficient and quickest way possible.
The file list
At the moment, my Custom GPT's file list of training data looks like this:
This way of keeping the data in separate files makes it easier to do minor changes or replacement of data without having to interfere with all the data each time.
I only have two emply 'slots' left though, so at some point I may have to combine two or more of the documents to free up more space for additional training data.
Creating data from nothing but thoughts
The reason why this project has been 5+ weeks underway and the conversations take up the equivalent of 300+ pages in Word is because I had to create a lot of the training data from scratch.
In the 11 years Quantum Public Relations has existed, I have contemplated things like Mission, Vision and Core Values many times. I have even tried writing them down several times. But I always ended up tearing up the paper and throwing it in the bin because it sounded corny or like a cliché.
领英推荐
This time around, I used ChatGPT o1 Preview and had a lengthy discussion about how I found it difficult to articulate what I wanted to accomplish. I set aside all feelings of insecurity, vanity, and doubt and just spoke about what I dreamed of doing and why I felt so strongly about it.
And it turns out that o1 Preview is an absolutely fantastic interviewer and was able to distil my thoughts and ambitions into clear language for me. For the first time in a decade, I actually feel like I now have a clear articulation of why I do what I do! ???
Following my successful creation of my Mission, Vision and Core Values-document, I used the same approach to 'Business Goals and Objectives', 'Operations Data', 'Market Research' and 'Client Feedback and Case Studies'.
The later two were really difficult to write because I didn't have any reliable data. In a way, part of the reason why I wanted an Advisory Board in the first place was because I wanted help doing market research and collecting client feedback in a structured way.
Thus, those two documents can best be described as a work-in-progress, but that gives my Advisory Board members an indication of what I struggle with and where their input can create value.
Coding the Instructions for how the Custom GPT interacts
The final piece of the puzzle, completed this weekend, was setting up the Instructions for the Custom GPT.
I know from experience that it is very easy to make instructions that do not work as intended, so again I asked o1 Preview for help and examples of structures and how to word the instructions according to the output I desired.
I want:
This is where I expect to have to make the most testing and fine-tuning of the Custom GPT in the next two weeks. And it is quite possible that it won't work - that ChatGPT will mix up profiles and have them answer in each other's place or role. ????
Time will tell.
Next steps
Over the course of the next two weeks, I will have to 'field test' my Advisory Board Custom GPT and probably make adjustments to the Instructions and possibly also the training data documents.
Furthermore, if everything seems like it is working, I will have to complete the financial data about yearly revenue and clients and services. That is going to be quite a load of work, so I simply didn't want to do all of it now, before I have tested the core concept of the Advisory Board with the existing training data as it is right now.
I have named four of the six advisors so far, but I still need to come up with names for the last two. I also want to generate character illustrations for them to help me immerse myself in the experience when I interact with the personas.
Reflecting on my Advisory Board experiment so far
It may seem premature to reflect on the success of the experiment before I have even really tested the output and outcome, but here are nine thoughts that I have been thinking about.
1. I probably should have started with multi-persona testing
Only after building all the documents for the training data did it occur to me that the critical part of this experiment is getting the multi-persona aspect to function properly. And to test that I really did not need all the background information. I could have run a much simpler test to see how a CustomGPT with multiple personas would interact with me and based on that decided whether proceeding with the experiment was feasible. Oh well. ??
2. Providing training data can be a chicken-and-egg situation
To provide advice, the members of the A.I.-powered Advisory Board need data. But in my case coming up with that data was exactly why I needed an Advisory Board in the first place (e.g. because I cannot figure out how to do proper Market Research on my own).
If you are stuck in a similar situation, the best advice I can give is to explain it to ChatGPT and let it help you come up with a stop-gap measure.
3. How is this going to work without ChatGPT Memory?
So far, OpenAI has been silent on how they expect to integrate Memory with Custom GPTs. All we have heard is that it will eventually be a feature in Custom GPTs as well. But without Memory my Advisory Board will face difficulty 'evolving' as we have more and more conversations together.
I considered putting something into the Instructions about having the members suggest to me when someting is important enough to justify an update of the training data - but I fear that this could become an arbitrary step that the A.I. would have difficulty setting up good, clear parameters for. And therefore it would constantly ask me to update the training data files after each chat session. ??
4. I am not going to mess with DEI in this particular experiment
I gave a lot of thought to diversity in my 6-person Advisory Board. Right now I have three women and three men, ages 32-56. But that diversity is mostly for my own immersion with the characters. There is nothing in their personas about acting a certain age or gender.
I considered building even more diverse personas, but ultimately decided not to risk derailing the project on that account. My reasoning being that A.I. models have been trained in a way that reflects the bias of their creators (this has been established in several papers) and any representation of certain cultural values or norms would reflect this.
5. There were roles that I excluded on purpose
I chose to focus on the types of roles for my Advisory Board that were (fairly) easy to synthesize with training data. What is missing from my advisory board's members is:
This is important when thinking about what types of advice I can ask them for. And it means that this Advisory Board is probably less well-rounded than an optimized human Advisory Board would be.
6. Will they be able to disagree with each other - and with me?
A.I. is notoriously willing to serve and eager to please. This sometimes leads to hallucinations and even when it does not, you will rarely if ever hear an A.I. tell you that you are wrong. Or that you should not do something, you are intent on doing.
But if I cannot get the advisors to (sometimes) disagree with each other or with me that drastically reduces the usefulness of the concept of an A.I.-powered Advisory Board. We shall have to wait and see.
7. Using different A.I. models yields different advantages and disadvantages
It has been a long 5+ weeks working on the different aspects of this experiment. Along the way I have used both ChatGPT-4o and o1 Preview. And the experience is like night and day. The more I use o1 Preview, the more ChatGPT-4o feels a little simple-minded. The advantage of ChatGPT-4o is that it understands brevity!
Each time I prompted o1 Preview - even when I told it to not go off on a tangent - it came back with answers, comments and loads of suggestions taking up several pages. I enjoyed the reasoning behind its thinking, but at the same time got extremely tired by the long-winded answers. ??
Particularly in voice chat on ChatGPT this can be exhausting, which was a problem because I actually enjoyed dictating a lot of my inputs because it was faster than typing them.
In the end though, I found o1 Preview to be the superior model for working out a lot of the training data documents because it was able to work with me when I had trouble articulating something and could often distil my thoughts much better than I or ChatGPT-4o was able to.
8. Maybe there is a better system for building an Advisory Board than a Custom GPT?
At present, I am aware of Gemini Gems, Claude Projects and Copilot Studio - all of which allegedly works in a manner similar to Custom GPTs from OpenAI. I haven't gotten around to playing with any of them, though. So, for now at least I have designed my experiment around a CustomGPT. But I would love to hear from anyone who has done anything similar with one of the above tools - or perhaps something I have not even heard of yet?
9. Could we see a future in which A.I. and human Advisory Board members work together?
I recently saw an article about how an A.I. would be deployed as a jury member at a US marketing award. The A.I. would be trained on a decade's worth of data, encompasing more than 150,000 companies and 180 million different marketing campaigns.
It made me think that it is only a matter of time before we will see A.I.-augmented decision-making - in board rooms, at C-suite level and all the way down through the layers of organisations. Why wouldn't we - when the data is there and just need an A.I. to structure it and see the patterns that we as humans cannot?
That's all folks - join me again in two weeks for another edition of The A.I. Exploratorium where I will be introducing you to my six Advisory Board members and telling you all about how their first weeks on the job fared!
Wish me luck! ??????
Do you have any questions or ideas for the newsletter?
If so, I’d love to hear from you, either directly or in the comments section. An exploratorium is always more fun when the experience is shared with others.
Thank you for your continued interest in reading the A.I. Exploratorium!
My name is Jesper Andersen, and I am an independent communication advisor. My core areas of expertise are communication measurement and evaluation, communication strategies, Thought Leadership, and the use of Artificial Intelligence and how it is transforming the entire field of communication.
I teach the MBA and Master’s programmes at Quadriga University in Berlin, speak regularly at conferences, and conduct workshops and training sessions.
If you’re facing a challenge where you think my experience and skills could be helpful, or if you’d simply like to exchange ideas and perspectives on modern communication, I would be more than happy to start with a non-committal chat and we can take it from there.
AI Passionista | Master Data | Supply Chain | Project | Service | Digital | IT | Graphic Design | Humans | Management
2 个月Very interesting
R?dgiver. Skaber konkret v?rdi + Business developer + Foredragsholder + AI + Innovation + C-level + Dr.med + Professor + Executive MBA + H-index 75
2 个月Det er et rigtig sp?ndende setup som jeg vil pr?ve at forf?lge - vil det v?re muligt at overtage en klon af dit pretr?nede AI advisory board og s? Customise det til min virksomhed?
Product Manager | Empowering Product Growth with Tailored Strategies | Professional Certified Coach
2 个月https://personalboard.ai
AI Advisor I Strategist I Educator I Speaker I ex- NBC, CNN, BBC & Reuters “She demystifies AI for companies, C-suites, classrooms and communities.”
3 个月I have been experimenting as well! On pause for a couple of weeks due to lack of bandwidth. Looking forward to hearing more Jesper Andersen.
International PR for impact tech @?Kemp & Kj?r
3 个月?? Godt t?nkt! ??