Oracles, sages, genies

Oracles, sages, genies


Hi, I'm Tim Tyler, and this is a video about oracles, sages and genies.

To start with, some definitions. For the purposes of this video will define oracles as agents who can predict the future, sages as agents who can answer questions, and genies as agents who can grant wishes.

My "matching pennies" web site is mostly oriented around forecasting - so: it is concerned with oracles. This video illustrates the position of oracles in the traditional pantheon of powerful helper agents.

Here is a diagram. The least powerful creatures at the bottom and the most powerful creatures are at the top.

  • "Oracle" is the least powerful agent I will be considering. It can only do one thing - given a stream of symbols, it predicts the changes of the various possible symbols coming next.

  • "Sage" is the next most powerful agent. A sage is able to answer questions - based on the information that is available to it.

  • "Genie" is the term I will be using for a powerful, intelligent helper agent. It is the most powerful of the agents I will be considering. It has connected itself up to powerful actuators, and learned how to use them.

If you have a Oracle, it isn't trivial to make an Sage. You could attempt to create perceptions of an environment which appears to have a sage in it, and then try to print out its source code - or simply predict its answer to questions. However, this may not work too well. For one thing, if your forgery is not highly convincing, the Oracle might figure out your ruse.

If you have an Sage, making a Genie seems more practical. You can just ask the sage how to construct a genie. You might have to expose the sage to a lot of information before it can help you. However, you can ask it what information it thinks would be best to supply it with - which should help.

These agents are portrayed as general agents. However specialised "expert" versions of all of them are also possible. So:

  • An expert oracle might be able to tell you what was likely to happen in the future - provided what you were interested in was the weather.

  • An expert sage might be able to answer any question - provided it was about the second world war.

  • An expert genie might be able to grant any wish - provided what you wanted was some sort of culinary experience.

The interest in oracles and sages arises since they are cut-down versions of intelligent agents that look as though they may well be easier to construct:

  • Oracles can be trained using machine-generated test data, in a virtual environment.

  • Sages are harder to construct - but can be trained using databases of questions and answers - which can be crowd-sourced, with the answers being voted on. Some companies create these kinds of databases anyway - to act as question-answering resources for their users.

One common vision of how we will make general purpose intelligent machines is that we will start with a range of "narrow" expert machines - and then will gradually expand their scope, so that they become more generally intelligent.

Oracles and sages suggest a complementary vision - one of general purpose agents whose capabilities start of shallow and then gradually deepen over time.

Hopefully this helps to puts oracles into a more familiar context.


Tim Tyler | Contact |