4 ms·

A guy was interviewing with an autonomous mobility team about a year ago and he said something to the effect of "just train it with machine learning." That deca

A guy was interviewing with an autonomous mobility team about a year ago and he said something to the effect of "just train it with machine learning." That decades of research into planning algorithms based on mathematics and theory from top universities might have been done for a reason escaped him. The team joked about this for months afterward.

falcor84 4y ago
Not that I'm necessarily defending this sentiment, I did recently watch a lecture about ML (I think it was fast.ai's Jeremy Howard) with a similar real life example. He spoke about a project from the 2000's looking to apply an artificial neural net to a medical diagnosis problem. What they did was spend many years of work with domain experts to identify the best features for the net to use. The professor then proceeded to explain that nowadays feature engineering approaches became so much more powerful, that significantly better results could be achieved in a more recent project much faster and without almost any domain expertise.
- idontpost 4y ago
  > that significantly better results could be achieved in a more recent project much faster and without almost any domain expertise Except, you know, for the expertise that generated the data in the first place.
- A4ET8a8uTh0 4y ago
  It is not exactly the same, but the process is strikingly similar. A family member of a buddy of mine is writing a game now. I mean he does subcontract for art part, but as one man army, he can really focus on his vision and not on... a lot of stuff that used to come with making a game ( platform, payment, distribution ). It is quite a time to be alive.
- pjbk 4y ago
  Yes, better models and more refinement, and yet they still lack common sense. Computers (computational statistical models, ML included) are very good interpolators, even moreso in higher dimensions which are difficult or impossible to grasp for humans, to find linear or non-linear correlations. However when it comes to extrapolation most models fail miserably. It can be argued, since we don't know the ultimate cause of the laws of our universe, that we also rely in abductive reasoning and guide ourselves through models created by correlations based in our experience. But that has happened through millions of years of evolution and we still don't know how to build AI supervisors that can challenge that.
Animats 4y ago
Google Translate used to use algorithms created by a large group of linguists. In 2016, the machine learning systems exceeded the performance of the semi-manual created system, and the linguists were laid off.
- anigbrowl 4y ago
  That's one of the more extreme cases of 'training your replacement' I've heaard.

falcor84 4y ago

Not that I'm necessarily defending this sentiment, I did recently watch a lecture about ML (I think it was fast.ai's Jeremy Howard) with a similar real life example. He spoke about a project from the 2000's looking to apply an artificial neural net to a medical diagnosis problem. What they did was spend many years of work with domain experts to identify the best features for the net to use. The professor then proceeded to explain that nowadays feature engineering approaches became so much more powerful, that significantly better results could be achieved in a more recent project much faster and without almost any domain expertise.

idontpost 4y ago
> that significantly better results could be achieved in a more recent project much faster and without almost any domain expertise Except, you know, for the expertise that generated the data in the first place.
A4ET8a8uTh0 4y ago
It is not exactly the same, but the process is strikingly similar. A family member of a buddy of mine is writing a game now. I mean he does subcontract for art part, but as one man army, he can really focus on his vision and not on... a lot of stuff that used to come with making a game ( platform, payment, distribution ). It is quite a time to be alive.
pjbk 4y ago
Yes, better models and more refinement, and yet they still lack common sense. Computers (computational statistical models, ML included) are very good interpolators, even moreso in higher dimensions which are difficult or impossible to grasp for humans, to find linear or non-linear correlations. However when it comes to extrapolation most models fail miserably. It can be argued, since we don't know the ultimate cause of the laws of our universe, that we also rely in abductive reasoning and guide ourselves through models created by correlations based in our experience. But that has happened through millions of years of evolution and we still don't know how to build AI supervisors that can challenge that.

Animats 4y ago

Google Translate used to use algorithms created by a large group of linguists. In 2016, the machine learning systems exceeded the performance of the semi-manual created system, and the linguists were laid off.

anigbrowl 4y ago
That's one of the more extreme cases of 'training your replacement' I've heaard.