Comment by anupshinde | Hacker News Reader

Comment by anupshinde | Hacker News Reader

Statements like these suggest that another AI Winter is coming (sooner than 3 years I guess)

"""He thinks even programming is vulnerable to being automated and reducing the number of available programming jobs."""

I believed something similar could happen within 1-2 years of learning/writing AI programs (more than 12 years back). I believed it so much that it consumed most of my weekends as I took on the Genetic Programming approach. Yes! computers can write programs - BUT trying reading those. Eventually after spending hours or days, you will be able to read those programs and you might find a simple "hello-world" program represented by a complex mathematical equation. Good luck trying to get such program fixed by humans. Imagine an experience decoding deep-learning-neural-nets. However, that is black-box from a programmer perspective.

From a business/management personnel perspective - the code is a black box anyways. When they get NNs that can generate required software, they will replace the people-manager with a NN-manager (who is a programmer btw!)

Are you referring to something like this?

Using Artificial Intelligence to Write Self-Modifying/Improving Programs http://www.primaryobjects.com/2013/01/27/using-artificial-in...

No, that looks like GAs with string/array representations. Similar thing worked for me when I tried randomly referencing nodes within a chromosome (say from index A to index B) - generating a graph like structure

The outputs are like this for some not-so-easy targets:

Op nodes ['ADD', 'ADD', 'MUL'] EXPR: ['ADD[ni_99](ADD[ni_49](I__7[ni_43](), ADD[ni_19](I__8[ni_66](), ADD[ni_79](GET_CONST_3[ni_25](), I__9[ni_71]()))), I__1[ni_61]())', 'ADD[ni_13](I__6[ni_17](), ADD[ni_49](I__7[ni_43](), ADD[ni_19](I__8[ni_66](), ADD[ni_79](GET_CONST_3[ni_25](), I__9[ni_71]()))))', 'MUL[ni_68](GET_CONST_8[ni_73](), FLOAT[ni_42](I__1[ni_91]()))']

With genetic programming (using an AST), it can solve complex equations:

However, this simple equation i.e. correct answer "(a + b + c - d) / e" could be evolved and will result into either of these (depends on my luck maybe)

Case1: ((int)((b+((c-d)+a))/e)&(int)((b+((c-d)+a))/e))

Case2:

((((int)a&(int)(((((((mod(e,a)c)/e)(((((int)(d-a)&(int)b)+e)/a)/e))/((((int)e&(int)c)+e)+(c+(e/a))))c)/e)e))/((((int)e&(int)c)+e)+((b/e)+(e/e))))+(((((((((mod(e,a)c)/e)(((((int)(d-e)&(int)b)+e)/a)/e))/((((int)e&(int)c)+e)+((b/e)+(e/e))))c)/e) <..........10383 characters here.........> ))))))/e))/((((int)e&(int)c)+e)+(((d-e)/e)+((d/b)/e))))+(((mod(e,a)/e)+(b/e))+(((c/b)+((d+b)/e))/e)))))))))

The GP output (case 1 and 2) above was generated with a tweaked version of https://github.com/rogeralsing/go-genetic-math

> Statements like these suggest that another AI Winter is coming (sooner than 3 years I guess)

I wouldn't bet against it. It's very risky to extrapolate progress in AI.

A DARPA project named CHI [1] succeeded in automatically producing programs from a high-level knowledge description language. And when they fed it its own description, CHI rewrote itself, with a better implementation. Do you fear for your programming job yet? Well, that was 36 years ago.

[1] www.dtic.mil/get-tr-doc/pdf?AD=ADA105661

Reply on news.ycombinator.com