Checking Citations from David Thorstad

Table of Contents

I checked three cites from Against the singularity hypothesis by David Thorstad. I specifically looked for quotations and checked the first three I found. I wanted to test my theory that Thorstad’s cites should be assumed unreliable due to some other errors he made. I also wanted to test my theory that misquotes are common.

Before reading this post, you may want to practice your skills by checking the cites yourself. If so, stop reading and go here to find the cites without my analysis.

Quote 1: Horvitz and Selman

That panel lamented “a tendency . . . to dwell on radical long-term outcomes of AI research, while overlooking the broad spectrum of opportunities and challenges” raised by machine intelligence (Horvitz and Selman 2012, p. 302).

Actually, the panel noted rather than “lamented” it. So the quote is misleading. The paraphrase of part of the quote, leading up to the quote, is wrong. Source

Panelists reviewed and assessed popular expectations and concerns. The focus group noted a tendency for the general public, science-fiction writers, and futurists to dwell on radical long-term outcomes of AI research, while overlooking the broad spectrum of opportunities and challenges with developing and fielding applications that leverage different aspects of machine intelligence.

Quote 2: I.J. Good

This is perhaps the best-known form of the singularity hypothesis, introduced to philosophers by I.J. Good:

Let an ultraintelligent machine be defined as a machine that can far surpass all the intellectual activities of any man however clever. Since the design of machines is one of these intellectual activities, an ultraintelligent machine could design even better machines; there would then unquestionably be an ‘intelligence explosion,’ and the intelligence of man would be left far behind (Good 1966, p. 33).

Good wrote this in 1965, not 1966, so the cite is incorrect.

This quote is in the Chalmers paper discussed in the next section (except the end is cut shorter here). Maybe Thorstad got it via Chalmers rather than reading old papers himself, but then he wanted to cite the original, not cite it via Chalmers, to make him look better-read than he is. I think repeating old quotes from other papers is common because people don’t read enough but want to quote old stuff so that it seems like they know the field from the start up through today.

The original and the Chalmers version (on page 1) use double quotes around “intelligence explosion”. For some reason Thorstad changed it to single quotes.

Chalmers cites the correct year unlike Thorstad.

Quote 3: David Chalmers

But here is the entirety of Chalmers’ argument against the possibility of diminishing growth:

If anything, 10% increases in intelligence-related capacities are likely to lead to all sorts of intellectual breakthroughs, leading to next-generation increases in intelligence that are significantly greater than 10%. Even among humans, relatively small differences in design capacities (say, the difference between Turing and an average human) seem to lead to large differences in the systems that are designed (say, the difference between a computer and nothing of importance). (Chalmers 2010, p. 27).

Here’s the source article (from Chalmer’s own website). I also double checked with a second source (from a book published later).

Both my sources give:

If anything, 10% increases in intelligence-related capacities are likely to lead all sorts of intellectual breakthroughs

Thorstad has “lead to all sorts” whereas Chalmer’s wrote “lead all sorts”. Thorstad added the word “to”.

Could it be “lead to all sorts” in the original journal article version, and then changed on Chalmer’s website and in the book compilation? Doubtful. Adding the “to” seems like a grammar improvement, and fixes are generally made in later versions rather than removed from later versions. It’d be surprising if the later versions are worse than the original.

Since I couldn’t find a scanned, electronic copy of the original, which Thorstad cites, it’s plausible that Thorstad got a paper copy of it, typed in the quote inaccurately, and doesn’t triple-check all the quotes he types in from paper. For academic research, I think triple checking typed quotes is the minimum reasonable policy. People should strongly consider additional steps such as reading both versions out loud. It helps to take a photo of the quote on paper then use optical character recognition (OCR) software on it and compare that to what you typed in using a software Find feature that finds exact matches (if it’s not an exact match, go through it in small chunks until you find every difference). It’s also good to keep a photo of any quoted paper pages in your notes for the article so you can easily check it again later or show it to anyone who questions your cite’s accuracy. OCR software isn’t expensive or difficult to use (I used it myself when writing this article). Another useful way to catch errors is having text-to-speech software read text to you – many typos will sound totally wrong and stand out when software tries to speak them.

Chalmer’s version has re-numbered the pages compared to the original journal publication that started on page 7 not page 1. Thorstad cites with original page numbers. I have not found a scan of the original to check Thorstad’s page number claim but it’s plausibly correct (he’s off by 7, and Chalmer’s has renumbered by 6, but maybe the journal article started near the bottom of a page, offsetting some stuff by a 7th page, or maybe it fit a different amount of text per page).

Conclusions and Why This Matters

All three quotations with citations have an error. This is another example which poses difficulties for people who disagree with my claim that errors with cites and (mis)quotes are a widespread problem. The expert scholars and public intellectuals, who try to serve as thought leaders trusted by many others, make lots of mistakes and shouldn’t be trusted so much.

Mistakes are bad. Truth and accuracy matter.

Small mistakes make it difficult to deal with subtle issues. When an issue is subtle, it means small differences matter, so small mistakes matter too.

Small mistakes make it difficult to deal with complex issues. Complex issues involve many parts or aspects. There’s a lot of stuff going on. You need to get many different things right for it all to fit together correctly.

Think of it this way: If you make an error 10% of the time, and you deal with something with only one part, you have a good (90%) chance to get it right. But if you deal with a complex thing with 5,000 parts, you’re going to make hundreds of errors.

Mistakes related to quotations mistreat the person you’re quoting because you’re putting inaccurate words in their mouth without their consent.

Making lots of small mistakes allows bias. If we allow some mistakes and view them as acceptable, then it gives people the flexibility to make mistakes which favor their preferred conclusions. When someone has any choice about what to say or think, they can make a biased choice. If the truth is X but mistakes Y and Z are also allowed, then an author has three choices and can pick the one that helps his side (often by subconscious bias rather than conscious intent). If only the truth is allowed, then the author has to say X even if his biases prefer Y (or else people will say he’s wrong instead of listening to Y).

Is Thorstad just one bad thinker, while most intellectuals do better? Should we blame Thorstad personally? I don’t think so. Based on doing this kind of thing many times, I think Thorstad’s mistakes are pretty normal. There’s a widespread problem related to intellectual culture and norms. The attitudes of many people need to change, not the actions of a few. My scholarship blog category has many more examples of errors with quotes, citations and facts. My Bad Scholarship forum topic has more, too.