A Survey Question

Here is a simple one question survey that I’d like to get a hundred or so folks to answer. It is a surprisingly interesting question, and I have a bet with Bryan Caplan on it, but I won’t say more now, so as not to bias your answer.

Added 9p: The survey(s) are now closed.  I did three of them:  a few of my facebook friends, 100 folks via Survey Monkey, and 1000 folks via Quick Survey . The question was “In a popular book with a modestly technical readership, what word should be used to refer to the set that includes both humans and artificial/robot intelligences?” Here are answer counts:

FB SM QS Total
Sentients 0 20 282 302
Intelligent Agents 3 29 238 270
Agents 26 35 181 242
Intelligences 17 23 176 216
Entitites 1 8 98 107
Sapients 0 8 90 98
Beings 1 8 86 95
Turings 2 1 82 85
Actors 1 8 60 69
Persons 5 6 52 63
People 1 6 38 45
Players 1 1 22 24
Folks 3 1 16 20
Creatures 3 4 10 17
Souls 0 1 14 15
Total  64  159  1445  1668


My bet with Bryan was on if a larger survey would confirm the initial small fb survey. He was right that results changed lots – the final winning item got no votes initially!  The initial survey put the temporarily leading items at the top of the list, while the other surveys randomized the order each time. Also in the initial survey, people could see three comments saying sentients is an incorrect term relative to sapients.

Right now its a hard to imagine filling a book with phrases like “a population of roughly a quadrillion sentients.” But perhaps I’d warm to it. And it would flow smoother than the “intelligent agents” phrase. A term like “sentients” probably wouldn’t make Bryan happy though; he wants a term agnostic on if robots are really conscious.

I’m struck by just how varied are people’s intuitions on how to talk about and compare humans and robots.

Added 9Sept: Bryan is ok with using “sentients”, to avoid reader confusion. That helps me warm to it.

GD Star Rating
Trackback URL:
  • DW

    *Should* suggests that maybe there is only one answer, yet we can choose many. That obscures the question a bit for me.

  • Pingback: Robin Hanson’s Survey | feed on my links()

  • Dremora

    I’d also add that the set of possible humans is much smaller than the set of possible robots/artificial intelligences. Deciding on some of the terms includes unspoken assumptions about the types of AI, e.g. whether it has social emotions etc.

    • AnotherScaryRobot

      This was exactly my thought. I can conceive of AIs that it would feel natural to count as “people” or “folks”, but I can also conceive of AIs to which I would never apply those terms.

  • Stephen Diamond

    Are we allowed to discuss, or would that bias the survey?

  • Peter

    I had to write some off as I have a  bias against them based on c: Not human or robot intelligence.  For example “beings” would include non-carbon based life or something like (to steal from a book series I don’t remember) plasma based life or post-physical (from Peter Hamilton).

    PS: Speaking of that book series I don’t remember, google is failing me.  Was a series about how the universe is pretty much owned by artificial intelligence who are chasing the last human as he has a killswitch / virus embedded in his DNA.  The plasma intelligence who usually don’t care were involved as the AI’s were making galatic scale objects big enough to manipulate space and kill them.  Large parts of the series take place in a black hole event horizon.  Was written sometime between ’85 and ’95.  Thoughts?

  • Silas Barta

     Where did my first reply go?

    • mako

      note that they’re ordered by number of votes?

  • JonLoldrup

    …..aaaaand the survey died. (I clicked ‘next’ after answering the first question, and then,, nothing)

  • JonLoldrup

    also, the term should be “intelligent entities”.

    ‘agent’ implies a thing that can act.

    • Stephen Diamond

       Exactly, which is why I chose agent. I thought an intelligent robot could act. Is the problem that we apply the term “robot” differently, not the term “agent.”

  • Robert Koslover

    Sorry, but it seems to me that the phrasing of your question itself biased the answer.  After all, you referred to “intelligences.”  And then you included that option in the list.

  • Army1987

    No write-in answer? I’d use “self-aware beings”.

    • Stephen Diamond

       Where is it written that intelligences are necessarily self-aware?

  • Can you have the choices show up in random order each time someone views the page? I voted for “agents.” I can’t entirely ascribe my decision to its placement at the top of the list; I genuinely believe I like it most. But I can’t be sure.

    • Stephen Diamond

       My initial choice was “Intelligent Agents.” But when I reflected, I decided that’s redundant.

      • manwhoisthursday

        True, but “intelligent agent” flows better.

  • Valid

    “Souls”! Let’s co-opt the term.

  • Paul Tiffany

    As a proponent of panpsychism, I like the idea of robotic (or other temetic) agency.

  • I couldn’t take the survey…

  • Stephen Diamond

     I can’t exclude hindsight bias, but I think I would have gone along with Caplan. Don’t we know these little things have large consequences?

  • Gregory

    The term “folk” could be useful in the future. It could be used to promote unity without infringing on human-conservative notions of personhood. I would describe folk as anyone who participates in folk culture. It wouldn’t be an appropriate term to describe all robot intelligences. 

    One of the things I like about the term “folk” is that it fits nicely with popular narratives of good vs bad sentients. E.T. is alien folk, because he participates in folk culture through telling jokes and enjoying folk consumables. The aliens in War of the Worlds are non-folk. Dr Who is folk, the Daleks are non-folk. The T-800 is probably non-folk in the first Terminator movie, but is folk in Terminator 2. Vampires are bad folk but are often relatable because of their participation in human culture.

    I think the term sentients will be used because it minimises the similarity between humans and robots.

    • Dremora

      If a program like Siri can tell canned jokes, is it folk even if it has no social emotions? I do think we should have a term for all systems who can feel emotions. Some of them are not very self-aware or intelligent, which complicates the whole thing.

  • I voted for Intelligences, Intelligent Agents, Entities, and Turings. Turings is great because it describes what they *do* rather than attempting to define their essence. It is a bit of a harsh neologism at first blush, but hey, popular books on behavioral psychology (Nudge, Thinking Fast and Slow) talk about “econs” so why not? Intelligences is similarly good, but somewhat more stiff.

    Sentients is not good. It indicates “feeling” beings and is thus question-begging. Might “machine intelligences” have a sense of internal feeling? Maybe, maybe not. The label is a distraction. Souls is even more of a distraction than “sentients”.

    Agents is unhelpful because it refers to people in a certain capacity. Humans are “agents” with respect to action, “sentients” with respect to feeling. When we refer to humans we aren’t always referring to them in their capacity as agents. Likewise with machines.

  • Tleahey1

    The cognitive scientist Zenon Pylyshyn coined the term informavore to include both groups, several decades ago.

  • manwhoisthursday

    The problem with sentient is that it can also refer to your dog, or really any animal with a significant amount of consciousness.  I voted for intelligent agent.

    • manwhoisthursday

      You will note that the intelligent agent and agent votes together dwarf the others.  Intelligent agent flows better than you think (people think shorter is always better, but that isn’t the case) and you don’t need to use the adjective all the time.

  • Michael Bishop

    Robin, because its easy, and you’ll learn something, I suggest trying this again using: 
    http://www.allourideas.org/ for the survey.  The two key ideas are as follows: 1. People can write in their own responses and 2. People do pair-wise comparisons which makes voting a game.  The theory is written up by an innovative sociologist here: 

  • Silas Barta

    Now that the results are up, my top 4 were, in decreasing order, Sapients, Intelligences, Sentients, and Intelligent Agents.

    I wanted ones that are understandable at a glance, are not over-general or over-specific, and do not collide with well-defined terms in other contexts.

    Entities and beings are too general (they can apply to groups and lower life forms), Turings is too obscure, Persons and People would be ambiguous in this context, Souls has too much of a religious connotation, Creatures is already used for animals, Players and Actors have strong connotations toward other meanings (though Acters would work), and Folks is too informal (among other problems).

    Even within the ones I prefer, Intelligences and Intelligent Agents are accurate but a mouthful.  Sentients is good, but I prefer Sapients since the former is etymologically closer to “can sense things” than “is intelligent”.

  • Bo Johnson

    I voted for “intelligences”. I’ve seen it used before and thought it was very effective. I could also see “turings” as a type of future slang term, along the lines we use “folks” today. 

    This is great timing, I just finished “The Life Cycle of Software Objects” and have been thinking a lot about AI in fiction.  

  • John Dougan

    There is already a word for this: sophont

  • Kahn

    I would have chosen Sentients, as I’ve always thought it to mean “self-aware and having cognitive abilities.”