The open world fantasies which AI has created has to be seen to be believed and people are lapping it as gospel truth. A magic bullet would automate everything and that too with no effort and professional life would transform from the present drudgery and stress to a bed of roses. In the process we are forgetting the engineering journey we have taken so far, and its methods too and the nature of finality which is mandatory to achieve. We have been talking of generative AI since ChatGPT was unleashed to the world on the 30th of Nov. 2022. Generative AI has now given way to Agentic AI.
The whole AI revolution’s main goal is to achieve the AGI threshold, though what it means for the gigantic level remains undefined and for sure not granularly parameterised. Currently, people are talking of AI agents like they are prompts away from replacing entire departments. The fairy tale autonomous systems would handle anything thrown at them, with no guardrails. Just hand over your AWS credentials and it’s done. But, unfortunately, the real world does not work in that manner, especially not in the enterprise, where reliability isn’t optional. In a dream situation of an agent being 99% accurate might not always be good enough.
The tech industry loves a moonshot and most of us do. The current moonshot in the AI age is open-world AI. What would actually get delivered given the present AI capabilities and also help generate trust and loyalty towards this new technology. What will happen if we keep chasing open-world problems with half-baked technology; you will burn time, money and trust. How then do we make the AI work for us as it stands today? We need to focus on the problems right in front of us, the ones with clear ROI and clear guidelines, we can make AI work today. This is what we call closed world problems.
These problems are well-defined, with complete knowledge, tasks repetitive and knowledge well-bounded. The rules and the work flows are crystal clear. We can sort out real world problems of businesses; invoice matching, contract validation, fraud detection, claim processing, inventory forecasting etc. No headlines, but this is what businesses care about solving. At the enterprise level prompts are not the best way of going about things. What does the enterprise look like (the real AI agents)? They’re event driven, triggered by changes in the system and not user prompts, and are autonomous meaning they act without human intervention. These agents work continuously, disappearing at the end of a single task and they are mostly, asynchronous; working in the background. This is the only way to create a full proof demonstrated capability in AI.
SOLVING A MILLION CLOSED WORLD PROBLEMS WILL PUT AI AT THE TOP OF EVERY ENTERPRISE.
Sanjay Sahay
Have a nice evening.