Is AI the thing?
I've had a few conversations with people recently where they've been so deep into the hype-cycle of AI that they seem to have lost a connection to reality.
They genuinely believe that AI is the thing itself, that it's their solution to everything.
Now, don't get me wrong. I'm not an AI hater and I really do believe it is a game-changer for technology and how we work, but it's not the solution itself.
It's a tool.
It's a very powerful tool, and one we've not had before, but it's still a tool.
Many people just throw everything at AI and take its responses verbatim. AI is always very certain, very confident with its answers - even if it's wrong!
AI models feel very masculine in their tone and language.
Is it the new frontier of the patriarchy?
I'll leave that one hanging out there for people to think about!
The solution is the thing
Bring me your problem first
For me, the key is to ask "What are you trying to solve? What does it actually need to do?"
AI may be part of the solution, but if you don't know what you are trying to achieve, how can you articulate that to an AI tool?
So often you hear people say:
"Let's use AI for that!"
Really? Such certainty without understanding the problem.
I suppose it's the trend of just running head first at AI that gets me. Have you really thought through all the requirements, the constraints, the edge cases?
It's very easy to just build something with AI and it often just works – well, the happy path does.
Do you really know how it works, though? Would you trust the solution in production?
One of the challenges I often give to the FoundryOS team is to ensure that they really understand what the solution is, in order to be able to maintain it, to support it if it goes wrong.
That applies even if AI is involved.
And then there is regulation
I've spent much of my career in financial services, and regulation always comes to mind for me.
AI today is all about Generative AI (Gen-AI). Obviously AI is much broader than that, but Gen-AI is all about Large Language Models, or LLMs.
LLMs are probabilistic models, i.e. models that weigh up evidence and data points and output the most likely outcome to a given question. Sometimes that's correct, sometimes it isn't, depending on what the model knows and was trained with, and that includes the context you can give it.
Straight away, that brings a challenge for financial services. In a regulated world such as financial services, a system must respond to a question with the same result based on the same information, every single time.
You can't guarantee that with an LLM, unless you put some very strong guardrails in place. Those guardrails need to include the whole regulatory context: the local jurisdictional regulations, the policies of the company and even the policies of the product involved.
In financial services, in order to make a decision, we need to ensure that the system that makes the decision is deterministic, not probabilistic, otherwise it's very difficult to answer typical regulatory questions like, "Why was that decision made and why?"
LLMs are brilliant but alone they make this challenging because they are probabilistic by design.
This is really a whole blog post on its own; I'll do a follow up on how FoundryOS's policy engine manages this to ensure all outcomes are governed by regulatory controls and policies.
AI is the tool
So I go back to the start of this.
In the dotcom boom, people thought having a website made their company fly - but we know how that ended!
AI is part of the toolkit that lets us build and run modern solutions, and it often lets us deliver those solutions more quickly.
But, it can't replace understanding of what the problem is, what the solution actually needs to do.
AI is not the thing!
I love what AI brings and the power it has unleashed on us, both personally and in business, but to quote Aunt May, with great power comes great responsibility!
We can unleash the power of AI but let's not forget what we are trying to actually do.
Always question and challenge AI like we would each other.