While reading your article, the part I thought about most was whether artificial intelligence could be deceived in its evaluation and decision-making in any given situation. I agree with your general idea on this point. However, the question in my mind remains: can artificial intelligence be easily manipulated by humans, another artificial intelligence, or both simultaneously? As you mentioned in the article, studies are being conducted to observe the effects of artificial intelligence after theoretical research. I hope we can get answers to many critical questions as early as possible. Thank you for your intriguing article.
I have read this as I focus a lot of my current work on judgement, whether that be ai or human. Your point on it not getting the outliers is important. I think in standard assessment it performs well, within the written assessment of a vocational subject, models are struggle to pick up the tacit knowledge to which I learner would write about.
I do the sampling in a similar way but this is the end of process. I feed in the lessons as I teach them. I discuss with the AI what the particular rubric looks like. The students know the generic version. Then I get the AI to rank the essays first before we 'calibrate'. Not my word. That's the sampling. My concern has been with feedback to students. I am not reading 80% of student work, but the 80% are getting much better (quicker, accurate, personalised etc) feedback from the AI than was possible in the past. How should we be managing this? And perhaps more interestingly is the essay still the most reliable and valid test of what we think we are testing? Using AI the students can now generate the dozens of potential exam question answers the exam board may ask. All they need to do then is memorize them.
While reading your article, the part I thought about most was whether artificial intelligence could be deceived in its evaluation and decision-making in any given situation. I agree with your general idea on this point. However, the question in my mind remains: can artificial intelligence be easily manipulated by humans, another artificial intelligence, or both simultaneously? As you mentioned in the article, studies are being conducted to observe the effects of artificial intelligence after theoretical research. I hope we can get answers to many critical questions as early as possible. Thank you for your intriguing article.
I have read this as I focus a lot of my current work on judgement, whether that be ai or human. Your point on it not getting the outliers is important. I think in standard assessment it performs well, within the written assessment of a vocational subject, models are struggle to pick up the tacit knowledge to which I learner would write about.
I do the sampling in a similar way but this is the end of process. I feed in the lessons as I teach them. I discuss with the AI what the particular rubric looks like. The students know the generic version. Then I get the AI to rank the essays first before we 'calibrate'. Not my word. That's the sampling. My concern has been with feedback to students. I am not reading 80% of student work, but the 80% are getting much better (quicker, accurate, personalised etc) feedback from the AI than was possible in the past. How should we be managing this? And perhaps more interestingly is the essay still the most reliable and valid test of what we think we are testing? Using AI the students can now generate the dozens of potential exam question answers the exam board may ask. All they need to do then is memorize them.