Why the Turing Test won’t save us from murderous AIs, but I can

26 Jan

As has become a normal state of affairs with me, I’m worried.

This week, I’m worried about detecting artificial intelligence.

Sure, we have the Turing Test, but that only detects AIs too stupid to pass the test, and they’re hardly a threat.

Which is a major flaw. A test that only tells you when an AI isn’t intelligent isn’t terribly helpful if you’re trying to root out whether or not your loved ones are or have been replaced by a malevolent, manufactured intelligence hell-bent on wiping out the Human race, starting with you.

Some of you may feel compelled to point out that we have a better test for detecting AIs: the Voigt-Kampff machine.

Briefly, that test was effective, but as AIs evolved, detecting them with the Voigt-Kampff machine took so long that the AI in question ended up having plenty of time to administer its own test on you, the Living Or Dead Test. And the outcome of that test is always the same: you fail it, ending up, at a minimum, mostly dead.

So we need a better test. A Turing Mark II test, say, or a Voigt-Kampff Jr. machine. Something that can detect an AI and detect it quickly, before it detects (and deletes) you.

I believe I have come up with such a test.

Approach the person (or persons) you think may be a machine masquerading as a loved one, close friend, or colleague, and tell them they smell bad.

I know, I know, you’re thinking, “No, Ian, you want to ask them personal questions only the real person could answer. That’ll trip them up!”

Two problems with that approach:

  1. Asking those sorts of questions are a sure tip-off to the AI that you’re suspicious. Have you seen the beginning of Blade Runner!? Don’t alert the machine intelligence you’re on to it!
  2. You’re assuming the alleged person being questioned was ever actually human. What if the AI has been living here, alongside us, all along? It would have those memories you’re asking about, but still be a cold and calculating Human-killing machine biding its time until the right moment to strike.

So leave it to someone who has given this a lot of thought and do as I say.

Tell them they smell, that their hair looks especially unkempt and scraggly, and that they should avail themselves of your shower.

Free of charge.

The shower, incidentally, that you’ve rigged with hidden cameras so you can observe them.

For expressly this purpose. Yeah, sure, that’s the reason you installed all those cameras.

Those high-definition, internet-ready cameras.

By the way, don’t tell the test subject about the cameras. That will alter the outcome of the test.

In a decidedly fatal-to-you way.

If said test subject has demonstrated a certain moral flexibility / lack of inhibitions, and / or you don’t have the necessary cameras installed, you can offer to assist them in their showering in order to observe them during the test.

If you go this route, do not, under any circumstances, offer to soap them up. While this might lead to some short-term fun (or not), you run the risk that they will expect you to wash their hair, and if that happens, the test has been nullified.

Also, should you opt to join them, make sure your concealed weapon is very well concealed. To do otherwise leads to all sorts of awkward questions.

Once in the shower, observe the test subject as they shampoo their hair. This is the key moment, and hence, the reason I have dubbed this test the Shampoo Test.

A human will read the instructions, Lather, rinse, repeat and do so, once.

An AI will read the same instructions and get caught in an endless loop of lathering, rinsing, and repeating.

At this point, your work is done, as the AI will continue to lather, rinse, and repeat until its batteries go dead.

If this AI is nuclear powered, you might not be able to use your shower ever again. But having a man-destroying machine using up all your hot water for the next five thousand years is a small price to pay for your life.

Now AIs, as the name suggests, aren’t stupid. The whole Lather, rinse, repeat gag is not uncommon in Human comedy routines. So the more sophisticated AIs will have an extra bit of coding in their logic that will prevent this test from working:

Instead of performing personal hygiene [item]:
     if item is ('shampooing' or 'conditioning'):
          Stop the action;  // skip Repeat, causes execution issues   
     otherwise:             // non-hair washing action, do as normal
          perform personal hygiene (item).

This is very bad. It completely invalidates a negative outcome from the Shampoo Test, rendering its use pointless.

Or it would, if I humans weren’t such clever bastards.

The fix is simple:

It leaves your hair with a glorious, yet sticky shine that also attracts flies! Win win!

If the AI you plan to test has an aversion to ketchup, try catsup or some other, less offensive condiment.

Remove all shampoo (and conditioner) from your shower prior to administering the test.

Replace the shampoo with a bottle that has been modified, where the ‘shampoo’ label has been replaced with the word ‘Ketchup’ (or ‘mustard’ or ‘cottage cheese’ or ‘Miracle Whip’ or whatever non-shampoo goo you want to replace it with, though I recommend not using ‘Nair’).

Now the AI will wash its hair with the ‘ketchup’ and the sub-routine to avoid the Lather, rinse, repeat issue is completely bypassed, because the AI thinks it is using ketchup instead of shampoo.

Simple and yet brilliant.

So far all of my loved ones have passed this test, and I can sleep easier at night. I’m not quite sure how I’ll get my boss or co-workers to use my shower, though, and that makes going into work very stressful.

But at least in my own home, I can sleep safe at night.

Unless I’m the AI.

Oh crap.

And now, a word from our sponsor: me! My books are available!

The Santa Claus Gang:

The Santa Claus Gang: A Marlowe and the Spacewoman short story

Marlowe and the Spacewoman:

Marlowe and the Spacewoman

Kleencut (FREE, and another fine showcase for my artistic abilities!):

So bad it won a Voidy for the next THREE consecutive years (would have been FOUR, but 2012 was a leap year)

Leave a comment

Posted by on 26 January 2013 in Other Blogs


Tags: , , , , , , , , , , , , , , , , , ,

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: