Hacker News Reader

Comment by ellius | Hacker News Reader

I think people also have a very hard time conceptualizing the amount of time it took to evolve human intelligence. You're talking literally hundreds of millions of years from the first nerve tissues to modern human brains. I understand that we're consciously designing these systems rather than evolving them, but nevertheless that's an almost incomprehensible amount of trial and error and "hacking" designs together, on top of the fact that our understanding of how our brains work is still incomplete.

View on HN

I thought you were going to go the other direction with your first sentence. It took some 4 billion years to go from the first cell to the first homosapien. Maybe another 400,000 years to get from that to how we are today.

That means 0.01% of the timeline was all it took for us to differentiate ourselves from regular animals who aren't a threat to the planet.

0.01% of 100 years is 3 days.

That's a very anthropocentric view, and not how the timeline works. Unicellular organisms are also smart in a way computers can't exactly replicate. They hunt, eat, sense their environment, reproduce when convenient, etc. All of these are also intelligent behaviours.

And just 4 hours for AlphaZero to teach itself chess, and beat every human and computer program ever created....

DNA sequencing went from $3b per genome to $600, in about 30 years, much, much faster than Moore's "law".

Why do you say "much, much faster"? $600 to $3 billion is about the same as going 2^9 (512) to 2^32 (4.3B), which requires 23 doublings. Moore's law initially[1] specified a doubling every year (30 years would be 30 doublings), then was revised to every two years (15 doublings), but is often interpreted as doubling every 18 months (20 doublings). Seems pretty close to me!

[1] https://en.wikipedia.org/wiki/Moore%27s_law

Flight took a while to evolve too.

A difference here is that flight also evolved and re-evolved over and over. General intelligence of the scale and sort that humans feature just once (that we know of and very likely in history).

That's influenced by the anthropic principle. The first species to obtain human-level intelligence is going to have to be the one that invents AI, and here we are.

I don't think that's the same. We're not trying to reverse engineer flight. We're trying to reverse engineer how we reverse engineered flight.

The thing is, airplanes are not based on reverse-engineered birds. Cutting edge prototypes still struggle to imitate bird flight, because as it turns out big jet turbines are easier to build. It could very well be easier to engineer a "big intelligence turbine" than it would be to make an imitation brain.

> It could very well be easier to engineer a "big intelligence turbine"

Is that not what a computer is? We have continuously tried and failed to create machines that think, react, and learn like the brains of living things, and instead managed to create machines that manage to simulate or even surpass the capabilities of brains in some contexts, while still completely failing in others.

Reply on news.ycombinator.com