DomRSI: Domain-Specific Recursive Self-Improvement
This is where we will see real gains in businesses today.
The promise of recursive self-improvement is that machines make themselves better. The machine designs the machine, and the human steps back and feeds it the energy, water, and whatever else it needs to produce the next version of itself. That next version is slightly better at everything than the one before. Once it’s ready, it’s used to generate the next machine, which is also better, and so on and so on.
With many AI companies implementing versions of RSI today, it can almost feel like there’s no hope for humans in the future. I’d like to propose there may be a glimmer of hope. It’s something I’ve termed domain-specific recursive self-improvement, or DomRSI.
Better at thinking is not better at the right things
You see, as machines teach themselves to get better, they don’t necessarily get better at the right things. They do increase their general intelligence: their ability to think about problems in new ways, to use mathematics, and to apply new techniques in the most advanced ways we’d see any researcher in a laboratory apply them.
What RSI does not buy you is a guarantee that this “improvement” will drive innovation in the right direction. The direction humans would perceive as improvement.
We know improvement when we see it. In every domain and every field you see the best.
The best architects, the best engineers, the best builders, generally the largest companies of their kind, have developed processes, systems, and approaches within their specific domain of expertise to be great. To excel.
The crown jewels of the company
As Satya Nadella has said, and as I’ve posted before, these systems are in most cases somewhat secret. They’re the crown jewels of your company. Your proprietary knowledge and information. Your proprietary methods. What makes Apple special? How come everybody wants to work at Tesla? These are clear domain leaders.
It’s unclear today whether recursive self-improvement on a generalized model will supersede domain-specific businesses. After all, how do domain-specific businesses measure success?
The market is the arbiter
Well, the market does. The market measures the success of a business relative to others. It’s the ultimate arbiter. Whether in market share, producing the best products, services, and goods at the lowest possible prices, or creating a technology so unbelievable that people will pay almost anything for it, the market determines whether a company is successful.
So organizations today already have ways to measure how to become better. We do this every day. But it’s rather slow.
We have quarterly reports that tell us how we’re doing this quarter compared to last. We have annual reports. Last year was okay, how are we doing this year, what’s our projection for next year?
From years down to seconds
Our recursive loops take years. These loops need to take hours, minutes, and seconds. They also need to take into account every little process and activity conducted by your business, through your employees and your systems of record, that achieves outcomes for your customers.
You need to create a learning system. A system that understands your business. A system that understands your employees and what they do. Team that up with AI to help you see where the efficiencies lie, where time is wasted, where a product, process, or service can be improved, and the paths to make that happen quickly and seamlessly across tens, hundreds, and thousands of people.
This is domain-specific recursive self-improvement. And this is how we become competitive tomorrow.
If you haven’t started thinking about how to make your business get better every second with AI, you’re already falling behind.
That learning system is what we’re building at Memion: your company’s own knowledge, captured from the people who do the work, kept in-house, and put back into motion so the next decision is better than the last one. See how it works.