Hacker News Reader

Comment by imranq | Hacker News Reader

Why can’t all these companies just make one self driving car? Seems like a lot of overlapping effort with all of these separate initiatives towards the same goal.

The technology is agnostic unless you redesign the car from the ground up. All the training data and model development should be open source with some monetary incentive for the organization that contributes the most

View on HN

Because there might be multiple different development approaches to solve the same problem. It makes sense to explore more than one of them.

Some cars don't use LIDAR (Tesla), some use end-to-end machine learning (geohot thing), some are more of an "assistant" systems and others aim at full autonomy. It's unclear which of these approaches will yield results long term.

Understood, but at the same time it's like multiple companies trying to fly to the Mars. They might never achieve such a huge goal without combining efforts. Instead, they're going to have their proprietary ways, patent them so no one else can use them and all solutions will be less than the best that's achievable.

I’m just saying there should be more collaboration instead of silos. Machine learning is advancing so much because mostly everything is shared openly via papers, talks, etc

In all seriousness: because mankind is depressing and the only system of incentives we have requires people to compete with each other rather than cooperate.

Competition is a discovery process; if we knew the right answer ahead of time we might not need it. Having ten companies try ten different strategies is likely to find the right answer - the product that best fits our current set of technical/financial/social constraints - faster than having ten companies get together and hold joint meetings trying to get cooperation on ONE strategy only to discover it doesn't work, so they hold more meetings to convince the group to try ANOTHER strategy, discover it doesn't work either...until decades have passed. Attacking this kind of challenge serially instead of in parallel, by the time they have something that works the problem domain is likely to have entirely changed.

And sure, independent companies competing might all fail - some problems just can't be solved - but the odds of success are better. It's like the difference between having a dozen archers all aim arrows at a target, versus having the dozen archers all try together to cooperatively fire one arrow at the target.

What's wrong with 10 companies working together to try 10 different things, then eliminate the poorer performing ones, increasing resources on the better ones, until there's 1 winner they've all agreed on?

Seems a lot more logical than 10 companies working on 10 things where 8 doing the exact same thing and wasting 8x the resources.

> What's wrong with 10 companies working together to try 10 different things, then eliminate the poorer performing ones, increasing resources on the better ones, until there's 1 winner they've all agreed on?

Isn't that exactly what the market does? The company that is doing a better job makes more money which increases that company's available resources to improve the product and sell more of it, while the companies that are doing a worse job tend to make less money and eventually give up and do something else.

Apple tends to enter a business when they have a clear competitive edge based on features other companies aren't providing. They will putter around refining a new technology for a while in the labs without selling it then eventually find there either is or isn't a clear window for them to sell something great that does something existing products don't do. If the window is there, they jump in with both feet. If not, they go back to the drawing board and putter some more.

Companies working separately will automatically try different things, because separate companies have different strengths and different understandings of the problem space. Absent the kind of cooperation you want there's just about zero chance that Microsoft, Apple and Google would all be doing or attempting "the exact same thing". They're attempting different things, with different odds of success.

I don't even understand where you got this straw man from of "everyone should work on one thing". You do realize there are more targets out there, right? The reality of the market right now is that unless you put sufficiently high of a reward on hitting one target that you get ten people aiming arrows at it, no one bothers to fire arrows at all... there are so many things for mankind to be accomplishing :/... we just flat out don't need to figure out how to hit that target so dead center that we need ten people aiming at it: we should have people aiming at all the other targets. The most charitable version of your argument is "people don't work that way", but that's what is so depressing: people are just flat out disappointing. If we weren't so depressing that we needed competition to motivate us, we could achieve so many things it would boggle the mind.

I used to think of competition as depressing too but the reality is that, competition is necessary for many of today's technological advances.

As long as humans continue to have disagreements around ideas, which stem from differences in perspective and understanding, competition will always remain inevitable.

Imagine if Larry Page had accepted the status quo in search, the idea behind PageRank which made Google possible would never have been conceived.

I didn't say competition is depressing: I said humans are depressing. If Larry Page and the rest of mankind were less depressing of a species, he would have a way to be incentivized to work on PageRank without it requiring him to redo the work that was already done spidering the Internet, and it would mean that the next time someone came up with an even better way of doing it it would be plausible to launch it (as now re-spidering the Internet is effectively impossible if you have to compete from whole cloth against Google).

Autonomous vehicle mono-culture is just as bad as any other mono-culture. You have to have multiple efforts, each doing it a different way, for the best way to emerge.

Reply on news.ycombinator.com