Oct 18, 2017

Trying out DrWoeBot: lessons to learn

https://www.messenger.com/t/drwoebot


This Woebot is from Stanford, and the conversation above - I will summarize as "need for improvement".   It seemed only to be just regurgitating textbooks knowledge.

Let's identify all its weaknesses:

a.   Never seem to be listening to my message.

b.   It have a fixed path in conversation - predetermined, and seemed never to have any alternative available.   I suspect the same message is spewed out to millions of people identically without variation.

c.   I repeat after him verbatim, and he never seemed to recognize that as well.   Ie, legitimate instructions are not followed, and illogical statement are not detected.   I don't think there is any processing of logical statement.

d.   To assume that the conversing party will feel better after any conversation is not right.   Sometimes chatting will result in more anger, and there is no feedback on monitoring this - the emotional feedback of other party, or any feedback in general - whether understood or not, whether to cut short of details output or not etc.

e.   In general a conversation start off at the high level, and based on feedback, will try to probe deeper in the right directions according to the receiving party.   This one try to overwhelmed with details right upfront, quite annoying.   This feedback-based conversation will be able to capture the other party attention, and retune his own basic assumptions on what is the appropriate contents to be communicated.

References:

1.   https://www.wired.com/2016/05/googles-new-virtual-assistant-chattier-heres/

2.   www.cleverbot.com:   this seemed a lot better that all previous bot I have chat.

3.   https://www.newyorker.com/tech/elements/the-chatbot-will-see-you-now

4.   https://www.wired.com/story/a-sons-race-to-give-his-dying-father-artificial-immortality/

5.   https://www.wired.com/2015/06/google-made-chatbot-debates-meaning-life/

No comments: