I recently posed some common cannabis-impaired driving questions to ChatGPT. For example, "does cannabis affect my driving" "can police detect if I'm high" and "is driving high safer than drinking and driving". I've been asked these questions countless times by young drivers. I was astonished by the timeliness and accuracy of the responses and thought... can we use this technology in behavioural interventions? Or has anyone done this already?
I'm looking for thoughts/ resources/ experiences/ recommendations. Thanks!
I've tried Chat GPT a few times, and have been impressed by the coherence of the answers, almost always accurate. The problem for uninformed enquirers is that they can't recognise an inaccuracy. True, of course, when dealing with human experts too, but one sometimes has sufficient background knowledge about the expert's status to exercise due caution, which isn't the case with ChatGPT since there's no person or agent there!
But I don't want to be unduly pessimistic. The system is open to correction. When it oversimplified its answer to a question I posed about Repertory Grid technique, it acknowledged my correction and amended its answer accordingly, with a little difficulty over the syntax of its amendment.
What was particularly interesting was when I asked a question in Polish, a language in which one addresses relative strangers in the 3rd person singular; to do so in the 2nd person singular can be seen as impolite or over-familiar (though for certain situations, e.g. giving a series of instructions, this has changed in recent years). When I pointed this out, the system politely acknowledged the error and tried to correct, but got into difficulties.
Which highlights the issue; as I understand it, part of the machinery depends on a history involved the parsing of thousands of speech utterances– and either the system hadn't encountered sufficient examples in Polish or, (possibly more likely), didn't have sufficiently explicit rules mapping meaning onto Polish syntax and grammar in this instance.
I must try it with a bit of French 'tutoyer'!
Thanks for the response Devi Jankowicz. Corrections are helpful, but they can also be abused. I've seen threads where people purposely argue the wrong answer to see how the AI responds. I guess it's something to keep in mind.
- Badges
- Science topic
- Similar topics
- Psychology