1
$\begingroup$

It was noted today that automated text generation is advancing at a rapid pace, potentially accelerating.

As bots become more and more capable of passing turing tests, especially in single iterations, such as social media posts or news blurbs, I have to ask:

  • Does it matter where a text originates, if the content is strong?

Strength here is used in the sense of meaning. To elucidate my argument I'll present an example. (It helps to know the Library of Babel, an infinite memory array where every possible combination of characters exists.)

An algorithm is set up to produce aphorisms. The overwhelming majority of the output is gibberish, but among the junk is an incredibly profound observation emerges that changes the way people think about a subject or issue.

Where the bot just spams social media, the aphorism in question is identified because it recieves a high number of reposts by humans, who, in this scenario, provide the mechanism for finding the needle (the profound aphorism) in the haystack (the junk output).

Does the value of the insight depend on the cognitive quality of the generator, in the sense of having to understand the statement?

A real world example would be Game 2, Move 37 in the AlphaGo vs. Lee Sedol match.

$\endgroup$
  • 1
    $\begingroup$ I would say it depends on the amount of content. If an algorithm is stochastic, patterns may be discernible over a large enough data set which could pose a problem. It also depends if the bot is actively learning. Say you got a human to make 500000 GO moves, as humans are always actively learning, it's unpredictable as their previous moves will have an affect on future moves, where as alphaGO is already trained, and so should make the same decisions every time (unless some random element was thrown in, in which case it should be stochastic) $\endgroup$ – Recessive Oct 4 at 2:15
1
$\begingroup$

sometimes voice pitch also matters(and system designers forget about that) - if system is badly designed then you will feel that it is a bot, if design is great than you don't feel it's a bot at all

$\endgroup$
1
$\begingroup$

There are several aspects to this.

Firstly, content. I guess a further comparison would be to the monkeys on typewriters coming up with the complete works of Shakespeare eventually. You will probably have a huge mass of tedious text, with the odd nugget in it. One would hope that the signal-to-noise ratio would be better with human authors, though looking at Twitter that might not actually be the case (unless there are already more bots on there than humans!) Content evaluation/filtering is still an issue, especially with longer texts. How many computer-generated novels are there? Not as many as weather reports or news summaries.

Second, legal issues. Who 'owns' the text and is responsible for it? If a computer generates a text that incites violence, who is to blame for the consequences? The developer? Some texts require 'ownership', as they might have consequences attached to them. A wrong weather forecast that causes people to make wrong decisions, leading to loss of life. Do you sue the computer for that? So in this respect, clear responsibility is important.

So even if the content is good (point 1), there might also be responsibility (point 2) where it matters whether a text has been generated by an algorithm.

$\endgroup$
  • $\begingroup$ Assume it's not copyrighted formally, and the aphorism in question was discerned by a massive number of twitter retweets by humans. (There has been a recent initiative to list an AI as an inventor on some patent applications, so we'll have to see how that plays out re: elements of personhood for artifacts that create!) PS-Intentionality is what I was pondering after I asked this question--a big issue for AI ethicists is the idea that algorithms can't be meaningfully punished and so shouldn't be making impactful decisions. Thanks for your insightful answer. $\endgroup$ – DukeZhou Oct 4 at 22:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.