The Robots Are Taking Over! (Perhaps Not Quite Yet)

Do you know the feeling? You open the latest issue of a publication which you have come to trust, since over so many years it taught you so much in so many fields other than yours; you note that for once it talks  about something you know about; and you realize that the author has no clue what she is talking about?

A few weeks ago the New York Review of Books published an article [1] by Sue Halpern with the sensationalist title "How Robots & Algorithms Are Taking Over." The author notes that similar predictions have been made before but that hers are for real. I am not sure whether the robots are taking over, but I noticed this paragraph:

There is about a 50 percent chance that programming, too, will be outsourced to machines within the next two decades.

In fact, this is already happening, in part because programmers increasingly rely on "self-correcting" code—that is, code that debugs and rewrites itself.

Wow! Pretty soon no jobs for testers, and shortly thereafter programmers too will be out of work!

It turns out that automatic correction is an area I know, having been working on it for several years. The name of our system is clear enough: AutoFix. In [2] you can find an overview of AutoFix and a list of recent publications in the expected software engineering and verification venues (ISSTA, TSE, ASE, FASE, and the latest ICSE). Several other groups are working in this field; see the bibliographic references in our papers. I like to think of our results as pretty advanced, so Halpern's statement was of great interest to me.

Current work on automatic bug correction is great,  but if you read the papers you will see that the risk of massive tester and programmer unemployment, because robots have taken over their jobs, is not exactly around the corner. I would love to be able to state that tools such as AutoFix solve the program correction problem; but even though my social conscience would not even hurt (these are tools to help programmers, not replace them), I am afraid that we are still far from large-scale applicability of automatic bug detection and correction. Yes, we have made many strides in recent years, and you can already download and use AutoFix, but this is still leading-edge stuff. It assumes that programmers use Eiffel and include contracts in their code, which the entire world is not doing yet. (Some other work in the field makes different technology assumptions.)

I had read Halpern's article in print; checking the online version, I noticed that she has added a footnote at the end of the paragraph above, to support her assertions: it is a quote from Vivek Haldar, a  "veteran Google developer," who stated in an interview with Nicholas Carr, author of the recent book The Glass Cage: Automation and Us, that

The behavior all these tools [IDEs] encourage is not ‘think deeply about your code and write it carefully,’ but ‘just write a crappy first draft of your code, and then the tools will tell you not just what’s wrong with it, but also how to make it better.'

At least half of the statement has a good deal of truth: programmers do in fact commonly  write a "crappy first draft of their code" (although not all do). Haldar is also right that modern IDEs provide some guidance to improve code. But those facilities are primitive. They come nowhere close to the kind of AI-style magic that Halpern's readers would assume to be in common use.

Just to make sure — who knows, I might have missed a major development in my own field —  I showed the article to a few colleagues working in automatic program correction, and they  laughed. I also looked up Haldar, a prolific blogger; unsurprisingly, I do not share all his views, but he is clearly an experienced software engineer and I do not believe he would himself make the sensationalist statements about self-correcting programs that Halpern infers from his comments  to Carr — and which a simple attempt to look up recent software engineering literature, or talk to researchers in the field, would have immediately shown as unfounded.

The journalist's tendency to exaggerate is not limited to tabloids. Highbrow publications such as the New York Review generally have higher standards; but there is something about writing on science and technology that leads authors and editors to throw away all caution.

 

References

[1] Sue Halpern,  How Robots and Algorithms are Taking Over (review of Nicholas Carr's The Glass Cage: Automation and Us), The New York Review of Books, issue of 2 April 2015, available online here.

[2] AutoFix page at ETH Zurich, available at se.inf.ethz.ch/research/autofix/, with links to articles by Andreas Leitner, Carlo Furia, Yu Pei, Yi Wei, Martin Nordio, Andreas Zeller (Univ. of Saarbrücken), Bertrand Meyer et al.