Hacker News new | ask | show | jobs
by lars 2319 days ago
If I show you a picture and ask you if its a dog or a cat, you can answer the question, but you can't tell me why. You can make up a story to justify why it's a dog or a cat, but if someone tries to implement the procedure you give, it won't work. If it did, we could do computer vision by implementing your justification.

You are a black box AI. I can nevertheless trust you to classify dogs vs cats.
1 comments
opless 2319 days ago
No. I'm natural general intelligence with no safety features. One that has a hallucinationary view of the world an a neural architecture tracing back a couple of hundred of millions of years that continuously learns and has redundant features, and has the ability of self reflection.

We're talking about algorithms that are based on a simplified neural architecture, no redundancy, no self reflection and are still quite immature.

Nevertheless we're being asked to trust a black box AI, that you cannot interrogate?

Yes, of course, what's the worst that can happen?

link
lars 2319 days ago
It may well be that the human visual system is better for most tasks, but then that is what matters, which system is better for the task. The presence of an explanation doesn't matter. Neither system comes with an explanation.

On the other hand, if you place Magnus Carlsen against AlphaZero in a game of chess, I will bet on AlphaZero. If however you reduce the complexity of AlphaZero down to a level that it can produce an explanation I can understand, I would instead bet on Magnus Carlsen.

Of course we should care about the quality of AI systems, but chasing a human understandable explanation is just the wrong way to go about it, since it in many cases necessarily limits quality of the decisions.

link
naresh_xai 2318 days ago
Again the non-sequitur argument of an explainable model must be worse than a deep learning system and there has to be a tradeoff.

You don’t need to reduce complexity to induce explainability. You just need to decompose the function into smaller parts which you can understand.

Contrastive LRP for example is a Function decomposition technique for explaining deep neural networks with high fidelity.

link
opless 2319 days ago
I respectably disagree.

Here's a paper that you may not have read.

https://arxiv.org/pdf/1806.00069.pdf

link