Hacker News Reader

Comment by rebootthesystem | Hacker News Reader

Non-engineers see this as a failure. Engineers and scientists see this as an awe-inspiring success from which to learn, iterate and try again. I don't fault reporters for exagerated headlined focusing on failure. They don't know any better. That is not what their lives are about and most might be challenged to understand that nothing in that video constitutes failure in the context of evolving a design. In fact, failure is absolutely necessary in order to better understand the problem you are trying to solve.

View on HN

For an example of how this works in practice, just go back to the original rocket programs of the 50's and early 60's. Those guys dealt with failures much worse than this, and ended up with some of the most awesome achievements in human history.

This.

This video is one of the most beautiful things I have ever seen. This is not a failure. This is an incredibly rich mine of data on how to do better next time.

This is how we improve the world, and it's beautiful.

We are witnessing 0-1 technology of space exploration and its amazing. Its going to be an inflection point in the history of our species.

I don't agree that failure is absolutely necessary. I think success is generally more informative, because it gives you data about something that works end to end.

If your rocket blows up on the launch pad, you just learned that you shouldn't overpressure your fuel tank (or w/e). But if your rocket lifts off, goes to space, and lands, you learn thousands of things that work that you can use again.

I strongly disagree.

I have seen a company be destroyed because they got things almost right, from the beginning. This "success" informed their every decision and created an enormous resistance to trying anything new and in the end it turned out almost simply wasn't good enough. The initial success destroyed them.

That destroyed two Space Shuttles and killed their crews.

Both the failing field joints ("O-rings") and detaching external tank insulation weren't quite bad enough to kill a Shuttle, were all but ignored, until of course they couldn't be ignored.

I agree, this video makes it look like there was a strong chance that it could have been a success if the random variables on the way down aligned better. If it had of been they may not have learned and iterated based on this possibility and may have had a failure down the track.

Consider this: If you're conducting tests and measuring your points of failure over numerous iterations, you eventually find the "threshold" of success. Exactly how much howsit and whatsit you need to produce a desirable outcome. Then you can pile on a margin of safety as needed. Continuous improvement of this sort eventually results in an efficient system or structure along with robust documentation of failure modes which you can then anticipate and accomodate.

Also, there are different ways things can fail, E.G. catastrophically vs gracefully. For example, when your steel bike fails it will generally start creaking and slowly bend, allowing you to come to a stop or even ride home. Your carbon bike bike on the other hand is more prone to snapping or shattering, a catastrophic failure mode with potentially grave results.

I've +1'd because while I disagree, yours is a reasonable opinion that's provoking a useful discussion, and I think the downvotes you're getting are not appropriate.

The problem with aiming for success from the start is that it promotes a highly conservative, over-engineered solution. If it does work, it also acts as a brake on iterative improvement because your system already works, why risk failure by trying to make it more optimal? Also if you have a proven system, people start making bigger and bigger bets on it's success (more expensive payloads, for example), so the cost of any failure goes up.

If you aim for optimal design form the start, you're almost bound to experience failures. That's because say there are 100 parameters you need to tone and you aim for the estimated optimum for all of them. Statistically, half of those estimates are goign to turn out to be sub-optimal. As you iterate on the design, you fix those untill you end up with a truly optimal design. At each iteration, you have a positive incentive to make the necessary changes.

Depending on the cost of failure, this can be more expensive than aiming for a conservative, over-engineered design to start with, but not necessarily. Over-engineered systems, for a high chance of initial success, are by definition more costly than optimised designs and there's still a non-zero chance of initial failures anyway.

Reply on news.ycombinator.com