You can believe what you want about AI

28th September 2024

Some understandably take an attitude towards AI of "I'll believe it when I see it".

Hence Steve Wozniak said AI is when a robot walks into his house and makes him a coffee.

This seems reasonable at the surface level but I do take slight issue with it, because it's using human capabilities as a benchmark for AI.

The problem with this is that human intelligence is not general intelligence, it's very specialised animal intelligence.

Specialised intelligence

So this is how you might see AI progress if you believe intelligence is just a quantitative value.

However humans do not actually have general intelligence, we have a specialised animal intelligence.

So once you realise that there are different types of intelligence you realise it is really a venn diagram:

You can believe what you want about AI

Now at present the diagram looks like this:

There are some realms where the AI outperforms humans and some where humans outperform AI.

In this case you can either convince yourself that AI is stupid by focusing on it's weaker areas, or that it's a genius by focusing on it's strengths.

Now consider once it develops to this:

At this point the AI is over-all smarter than us, but it still may appear to be stupider than a basic human in certain areas because it's intelligence isn't specialised in those areas. Can it make you a cup of coffee at this point or not?

Finally it achieves "AGI" i.e. it performs as-well as a human in all domains:

Conclusion

The primitive AI of today already has some overlap with human abilities, and some abilities humans lack.

But it won't match humans in all domains until it's over-all intelligence becomes so overwhelming that it makes us look like monkey.

If you want to monitor AI progress, you have to pay attention to both the frontiers of where AI is performing best and also those areas where it performs badly. Only then can you get an accurate picture of it's progress.