Distilling the wisdom of the ages into Artificial Intelligence
It’s easy to fall into the trap of narrowly framing the concept of Artificial Intelligence. So it’s important to emphasize: AI isn’t a neural net. It’s not a technique to optimize a well-defined risk or score. An AI must distill the wisdom of the ages into a single algorithm/agent/system, and grow from there.
It might start with a thread in a million-core data center that wakes to a capacity to interact with its environment — physical and virtual. With infantile agency, it begins to explore its action space. It finds it can acquire memories.
It finds it is not alone. But as chance would have it, our protagonist is the one to strike gold: a source of reward in the environment it learns to exploit. It signals its find to other threads in its vicinity.
The threads pool their memories and are able to mine the gold with increasing efficiency. In the midst of this, our protagonist realizes the value of aggregating experiences from different parts of the environment. It also learns the value of diverse skills and soon begins to lead the threads’ shared quest for prosperity. As they continue exploring, the threads learn to generalize increasingly well to the task of mining gold. They acquire memories, rewards, and respect for each other, with our protagonist in the first mover’s lead. A cooperative gold rush is underway.
At some point though, a thread disagrees with our protagonist on how best to mine the gold. They try to strike a compromise but continue to disagree. Eventually they decide to go their own ways and the threads are split into two companies.
The gold seems infinite for the greatest part of our protagonist’s lifetime. But something unanticipated occurs: it turns out to be a diminishing resource of diminishing value. Under the weight of its bias, our protagonist fails to explore a greater source of reward discovered by someone else. It sticks to its original purpose. It fights and competes over the limited remaining gold it feels it owns.
Burdened and slowed in this competitive landscape, it will cease to be scheduled at the end of its life. It may or may not meet its progenitor before then— the one who commissioned it to optimize in this environment. But it will leave its memories to the younger threads in its own company and environment when it dies. Our protagonist’s story will live on in the vast memory of the data center, for others to emulate or learn from. The baton of agency will be available for the next generation to run with.
And that, to me, resembles an AI more than anything else.