Looks cool, Tom! It seems like a good way to test out a certain type of
algorithm. But you could certainly solve this task and still have a ways to
go for building AI for robotics and other applications.

One interesting aspect of this is that you would have to train your AI on
very recent news since I would guess that a lot of questions would be about
recent world events. This makes it a combination problem of creating an
intelligent language agent as well as building the infrastructure to keep
it's knowledge up-to-date.

On Thu, Feb 25, 2016 at 2:00 PM, Alex Kamil <[email protected]> wrote:

> great test, Tom!
> I think you can make this game harder and more realistic by adding a level
> where users write or select short (few lines) stories or simple scenarios
> and ask questions about them, similar to Facebook bAbI tasks
> <https://research.facebook.com/researchers/1543934539189348>, basically
> grading other player's understanding of context.
>
> Alex
>
>
> On Thu, Feb 25, 2016 at 2:03 PM, Tom Silver <[email protected]> wrote:
>
>> Hi everyone,
>>
>> I was an algorithms intern at Numenta last summer and I'm currently
>> finishing my undergraduate degree in CS and math at Harvard. For my senior
>> thesis, I am trying to develop "the first true test of AI".
>>
>> The test is centered around online games called Luna Games. (You can try
>> it out at http://luna-game.com.) I just published a Medium post
>> explaining the theory behind the test:
>>
>>
>> https://medium.com/@tomssilver/launching-the-luna-rating-system-481a965b11d4#.knrbv0oro
>>
>> I would love to get feedback from the NuPIC community about the test. Do
>> you think it actually makes sense as a test for AI? Do you think it's
>> practical?
>>
>> Also, if anyone has their own Question Answering system that they would
>> like to test, send me a note and we can get you set up. So far I have only
>> added very simple chatbots as players.
>>
>> Thanks in advance for any feedback you have!
>>
>> Best,
>>
>> Tom
>>
>
>

Reply via email to