When does a machine become human? When they can Lie…

Don’t worry I am not talking of androids or cyborgs. I am talking about the classical definition used to determine artificial intelligence. The most widely used benchmark or at least the most well know is

  1. The Turing test: is a test of a machine’s ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human.
  2.  

    The key take away here is the fact that artificial intelligence should be indistinguishable from that of a human. Now computers such as Watson from Big Blue etc. have long proven that we have the technological capability to create a system fast enough to browse the internet and return the correct or most logical answer 9 out of 10 times. However, we neglect a few other things that make us human. One of the most common examples is emotion.

    To tell you how many people died of cancer last year and to say it with genuine concern are two different things. The latter not mastered even by some human beings.

    This might be something we cannot achieve since I believe that is what really makes us Human. However, there is another aspect about human nature that I think we can program computers to do and that is ‘to lie’. Like any BI system, Watson can crunch the numbers and tell us what happened or did not happen. The game changer here is when we come close to self-awareness i.e. when computers can lie for their own self-interest.

    In order to do this all, we need is a decision tree that computers can use to determine when to lie. I am trying below to represent one such tree at a very high level.

    The idea here is that once a data warehouse contains a list of all possible “lies” and their weight ages in order of benefit , damage and acceptability ( how likely the target is to believe the lie) any system can be thought how to lie. Why am I making such a big deal about it? Being able to lie is one of the key elements of passive self-preservation. Active self-preservation will come when we build humanoid robots that know Karate :-). And intelligence (artificial or not) is knowing; when to give the right answer, the wrong answer and most importantly when to Shut Up.