The Double Ethical Bottleneck of A.I.
This is a small excerpt from my book The Journey Towards Compassionate A.I. The whole book describes the why’s, what’s and how’s concerning this.
Getting through the A.I. bi-bottleneck
On the road towards genuine super-A.I. – encompassing all domains of intelligence and in each being much more effective than humans – I see not one but two bottlenecks. This is no guarantee that after the bottlenecks, there will be heaven on earth, but let us suppose now that, from the day after, all will be well indeed.
That enables us to focus on the bottlenecks.
- The first one is human-made. Either rogue or naïve developers may take very wrong decisions. There is a sheer infinite amount of scenarios that one can think about. In the next section, I give a few examples. One may be enough not to need to fear any other anymore. It is the end, my friend.
- The second bottleneck is A.I.-made. On its way towards super-A.I. and beyond, there may be many stages and many changes. Even if the final stage is benevolent, and even if most stages are benevolent, it’s enough for only one stage to be less human-friendly, and we’re gone.
The bi-bottleneck may be quite long with many different stages.
Will humanity be able to control to a sufficient degree everything that can happen?
I think, and repeat, that reliance on control alone will not save us. I also don’t think we should relinquish all control and hope with crossed fingers that A.I. will eventually be Compassionate just like that, after the bottleneck as well as at any stage along. So, we should very well think about control, AND we should think about Compassion. BOTH are indispensable.
I hope that this book will be a wake-up call in this direction.
Related Posts
It’s RAG-Time!
Retrieval-Augmented Generation (RAG) is a component of an A.I.-system designed to synthesize knowledge effectively. It can also be viewed as a step toward making A.I. more akin to human intelligence. This blog is more philosophically descriptive than technical. RAG lends itself to both. Declarative vs. semantic knowledge Understanding the difference between these types of knowledge Read the full article…
Do It for the Children
We live in an age of immense possibility — and danger. Artificial Intelligence is not just a tool; it is a mirror. What we put into it will shape how it reflects us, to ourselves and our children. This is not just a game of tomorrow. It is a responsibility for today. See, for instance, 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…