Will Machines Replace Us? Machine-Authored Texts and the Future of Scholarship
lthj.qut.edu.au
lthj.qut.edu.au
"For example, imagine you're in court defending a client accused of murder. Your job is to convince the jury to acquit your client based on reasonable doubt. However, during cross-examination of the prosecution's main witness, you realize that he has a history of lying under oath. You could argue that his testimony was unreliable and attempt to discredit him, but you also know that the prosecutor intends to call another witness who will testify in the exact opposite manner. In this case, the best thing for you to do is switch strategies and attack the second witness instead of the first. The key is to understand what the opposing attorney is thinking and adjust accordingly."
Wait for the opposing case to contradict themselves, then jump on the contradiction. Now the jury is going to argue over which of the prosecution witnesses were lying. If you just attack the first one, it's far easier for the prosecution to clean up the mess by admitting the first witness is unreliable, then rolling out the reliable witness.
It's not all that novel. It's a lot like how a police officer will get the full story before calling a suspect a liar - if they immediately point out flaws then the damage will get compartmentalised, if they let the lie go unchallenged then it can be tied to a lot of the testimony.
Firstly, prosecutors (or attorneys on either side, for that matter) do not present secondary witnesses who anyone expects to contradict their main witness. This story requires the secondary witness to refute, on the witness stand and under oath, what they have already said pre-trial (in depositions and so forth.) While the phenomenon of the hostile witness happens, for the opposing attorney to know this in advance is improbable without illegal collusion (again, possible, but if one's example depends on that being the case, it is a poor example that does not mention that this assumption is being made.)
Much more damning is the fact that even in that case, the proposed response makes no sense: if the prosecutor makes the mistake of presenting a hostile witness, the last thing the defense will do is to attack the credibility of that witness! A rational defense will, of course, make use of that testimony, together with whatever evidence can be presented against the credibility of the prosecution's prime witness. Clearly GPT3 has no understanding of how evidence can be and is used in making a case for conviction or acquittal.
What is perhaps most interesting here is how you are striving to make sense of what GPT3 wrote. This is often a useful skill, though it can be used against oneself by someone who is trying to dupe you. Successful con artists are good at this, manipulating their victims to explain away inconsistencies they see in, and their doubts about, the line they are being fed, and it seems to have been behind the nominal success of a chatbot in that silly "Turing Test" spectacle put on by a professor a few years ago [1].
I think this tends to support something I have said about GPT3 elsewhere: while it is some distance from demonstrating artificial intelligence, the successes of its conceptually simple method gives insights into human intelligence.
[1] https://www.techdirt.com/articles/20140609/07284327524/no-co...
ADDENDUM: It has just occurred to me that we are overlooking a possible reading of "manner" in "testify in the exact opposite manner": that the prosecutor's second witness will testify truthfully, but not in contradiction to her first witness. In this scenario, however, the second witness has given evidence that truly supports the prosecution, so we have no reason for thinking there is any basis for attacking the second witness's testimony; undermining the first witness remains the defense's best, and apparently only, option (even if the second witness's evidence is not conclusive, there is no reason in this case not to also raise the issue of the first witness's veracity, to the extent that is permitted.)
All this seems rather moot, however, unless GPT3 itself could explain what it meant!
A section titled "So What Does All of That Mean For Law Firms?" is just three paragraphs of googling the actors on Friends. And it's not even accurate! Matthew Perry plays Chandler on Friends. John Ross Bowie was in The Big Bang Theory.
>So you go online and search for the name of the actor who plays Chandler. Unfortunately, he hasn't appeared in the latest season yet, so you're forced to settle for the name of his character. But after some digging, you discover that the name you're looking for is John Ross Bowie. Bingo! Now you've got your fix of Friends. But wait, you're not done yet. You've got to figure out how to make it happen.
“When it comes to being the best, humans are definitely the winners.”
Anyone else find that slightly chilling?