“Is knowledge knowable? If not, how do we know this?”—Woody Allen

I think you just proved that it is.  Proof by contradiction.

We either know something or we don’t know it. Boolean 1 or 0: very limiting.

Degree of belief, regarded as probability, is more flexible as it can take any value between 0 and 1. Rather than “I know X”, I prefer “I believe X with probability P”.

This is my current approach to bot development: the bot’s “knowledge base” is actually a set of beliefs, each belief comprising a proposition and a probability value. (If I were truly Bayesian I’d use a beta probability distribution.) Since I’m training the bot using supervised learning, it believes whatever I tell it to believe. However, it needs to be able to link beliefs together in order to infer a belief value for a proposition it hasn’t actually been trained with. For example a donkey bot needs to be able to link the proposition “I am a donkey” with “I like carrots” to calculate a degree of belief in “Donkeys like carrots”. This is not as straightforward as it might seem.

David

Further to the above, I note that the movie Ondine is being promoted with the tagline “The truth is not what you know. It’s what you believe.”

David

I like your approach to this David. It is actually quite ambitious, as I’m sure you know. Most bots are given their beliefs in the form of some kind of rule or hardwired data in the program. So called “learning bots” accumulate trends, rather than actually “knowing” about the subjects they discuss. I know of a few programs that have been given a world to exist in and talk about, but only in fiction have I run across systems that acquire new beliefs of their own.

I am always tuned in to http://www.convo.co.uk to see your progress.

Robby.

Thanks Robby!

Progress is painfully slow and the Convo website hasn’t been updated for a long time. However, I do offline research and development work every day, so hopefully one day I’ll have something to show for it.

I’m not keen on large data sets, so I use small ones for my experiments. My theory is that if something doesn’t work properly with a small amount of data, adding more data is only going to add to the confusion. (I know others would disagree with this view, thinking bigger => better!)

David

the chat bots are always ofline for me apart from 3