4 ms·

How does the team plan to address the issues faced by Microsoft's twitter chatbot Tay [0], which had racist inputs and in turn gave similar responses? While I d

How does the team plan to address the issues faced by Microsoft's twitter chatbot Tay [0], which had racist inputs and in turn gave similar responses? While I don't know how recent the corpus is, the majority of reddit speaks like and holds the views of college-aged white males, and many of the things said on reddit have been deplorable. It'd be a shame if OpenAI pooled all that computing power into training on a bad data set, resulting in an AI that regurgitates memes and random references in response to anything. [0]: http://www.theverge.com/2016/3/24/11297050/tay-microsoft-chatbot-racist http://www.theverge.com/2016/3/24/11297050/tay-microsoft-cha...

clydethefrog 10y ago
I feel this is a problem of studies that are interdisciplinary, especially when it's within "hard science" and "soft science". I am currently doing a double degree in communication studies and information science. They are both interdisciplinary. Communication sciences integrates aspects of both social sciences and the humanities (both "soft"), and so far when doing research both of these fields were taken into account and no students have problems with combining these fields. Information sciences integrates aspects of formal sciences ("hard") and social sciences ("soft"). When the course is about analysing communication data, the methodology of social sciences is also important - for instance questioning the validity of your data. That's the thing what you're mentioning: the majority of reddit speaks like and holds the views of college-aged white males, so the data does not represent everyone, and is not valid if you truly want to develop an AI for everyone. Whenever the "soft" science comes around, like writing an assignment analysing the validity of data, many of my fellow students struggle with the concept of data not being neutral. This is where the two fields collide, and usually it just ends up with students scoffing at that "illegitimate" scientific field. Many teachers also don't spend much time discussing that field during the lectures. I admit, I have written some lazy essays which probably had been given a negative grade if they were written for communication studies, but easily passed in information science. Of course information science is not AI, but they're both sciences that have parts of formal sciences and social sciences (I know AI has many more fields). I am afraid many talents within AI research miss essential knowledge about social sciences or deliberately ignore it, because it's not "hard" science. Case in point: your comment is now at the bottom of this thread. And then you get nasty surprises, like Google Photos categorising pictures of black people as monkeys.
- cooper12 10y ago
  Thanks for the response. You raise an excellent point on the interdisciplinary nature of a lot of modern projects, and how ethical issues can often be ignored. While I don't doubt the team is using some kind of heuristic to ignore spam and the like, it still pays off to examine the methodology used because for example upvotes could still capture unwanted data. I guess we'll just have to see and hope that the resources went to good use, rather than to create something only for entertaining a specific portion of the population in a limited way.

clydethefrog 10y ago

I feel this is a problem of studies that are interdisciplinary, especially when it's within "hard science" and "soft science". I am currently doing a double degree in communication studies and information science. They are both interdisciplinary. Communication sciences integrates aspects of both social sciences and the humanities (both "soft"), and so far when doing research both of these fields were taken into account and no students have problems with combining these fields. Information sciences integrates aspects of formal sciences ("hard") and social sciences ("soft"). When the course is about analysing communication data, the methodology of social sciences is also important - for instance questioning the validity of your data. That's the thing what you're mentioning: the majority of reddit speaks like and holds the views of college-aged white males, so the data does not represent everyone, and is not valid if you truly want to develop an AI for everyone. Whenever the "soft" science comes around, like writing an assignment analysing the validity of data, many of my fellow students struggle with the concept of data not being neutral. This is where the two fields collide, and usually it just ends up with students scoffing at that "illegitimate" scientific field. Many teachers also don't spend much time discussing that field during the lectures. I admit, I have written some lazy essays which probably had been given a negative grade if they were written for communication studies, but easily passed in information science. Of course information science is not AI, but they're both sciences that have parts of formal sciences and social sciences (I know AI has many more fields). I am afraid many talents within AI research miss essential knowledge about social sciences or deliberately ignore it, because it's not "hard" science. Case in point: your comment is now at the bottom of this thread. And then you get nasty surprises, like Google Photos categorising pictures of black people as monkeys.

cooper12 10y ago
Thanks for the response. You raise an excellent point on the interdisciplinary nature of a lot of modern projects, and how ethical issues can often be ignored. While I don't doubt the team is using some kind of heuristic to ignore spam and the like, it still pays off to examine the methodology used because for example upvotes could still capture unwanted data. I guess we'll just have to see and hope that the resources went to good use, rather than to create something only for entertaining a specific portion of the population in a limited way.