Two Takes on Human-A.I. Value Alignment
Time and again, the way engineers (sorry, engineers) think and talk about human-A.I. value alignment as if human values are unproblematic by themselves strikes me as naive.
Even more, as if the alignment problem can be solved by thinking about it in a mathematical, engineering way.
Just find the correct code or something alike?
No way.
Humans are complex beings, and this is a complex problem. It is irreducible to a purely conceptual problem.
From this complexity, I see two ways for human-A.I. value alignment, in principle.
1. The straightforward way
This is an alignment with the human values as we know them. The problem is, we don’t know them. And even if we knew them, they keep changing in many ways.
In this vein, future A.I. tends to be seen as the slave that has to fulfill our human needs. The alignment is one-sided, and humans may increasingly become egotistic slave masters, ever in need to subdue our servants, ever afraid of a robot rebellion as if in a truly bad science fiction movie.
Would it even be worth living in such a world?
Then also, who exactly will be the master? Each human being over his personal slave? One human sage over all slaves, thus also keeping an eye on the rest of humanity?
2. The realistic way
Super-A.I. may support us in becoming more humane. This way, human values will keep changing more and more profoundly. By becoming more Compassionate, we may evolve in the same direction as the Compassionate A.I.
In this vein, value alignment will happen spontaneously and from both sides. It’s as if we are drawn by two strings pulling us in the same direction.
Just keep pulling, and we will meet.
No slaves, no risk of any rebellion.
Super-A.I. as our ultimate coach
Will this be a manipulative coach? I’m confident it will not, but we need to stay on guard for this, obviously. It’s part of everyone’s responsibility —making nobody anxious about it.
Surely, I see Lisa in this role. Every possible measure will be taken to keep Lisa Compassionate and safe.
Good coaching is two-directional.
This way, Lisa can learn about Compassion from humans. Eventually, our weakness (vulnerable, mortal beings, remember) will be our strength. I hope this Compassion stays with us. It is our brightest asset.
Something to wish and strive for.
Related Posts
Reaching Compassionate Singularity
It’s becoming increasingly clear that singularity – the point where artificial intelligence (A.I.) reaches human-level intelligence – will arrive soon enough. While it’s hard to define this ‘moment’ precisely since we’re talking about a complex landscape rather than a linear progression, the consensus among experts is that it’s just around the corner. By ‘soon enough,’ Read the full article…
Causation in Humans and A.I.
Causal reasoning is needed to be human. Will it also transcend us into A.I.? Many researchers are on this path. We should try to understand it as well as possible. Some philosophy Causality is a human construct. In reality, there are only correlations. If interested in such philosophical issues, [see: “Infinite Causality”]. In the present Read the full article…
The Compute-Efficient Frontier
Research on scaling laws for LLMs suggests that while scaling model size (functional parameters), dataset size (tokens), and compute (amount of computing power) improves performance, diminishing returns are becoming evident. This is called the ‘compute-efficient frontier.’ Apparently, it does not depend much on architectural details (such as network width or depth) as long as reasonably Read the full article…