(I was going to post this example for Harvey1 on my other thread, but I think it might have some general utility).
The Pitch
Let's pretend that you're a CEO of a medium-sized company. One day, you wake up, read the business section of your newspaper, and discover something very interesting.
Tessier-Ashpool (TA), a privately-held French/German startup, has issued a press release. They claim to have developed an Artificial Intelligence (AI) that is capable of mimicking human consciousness. You can talk to the AI on instant messenger, or on the phone, and it is capable of conversing with you at least as well as an average human being can. However, unlike a human being, the AI is scalable, enterprise-class, robust, and tireless; if you want more performance, all you have to do is buy more computers to run the AI on.
TA is licensing their AI as an ASP platform. That is, for a certain monthly fee, you get access to your own copy of the AI that runs on TAs servers. TA maintains and administers these servers, and the AI will handle whatever IM conversations or phone calls you feed to it. The AI can be easily trained in a variety of common tasks.
As a CEO, you see immediate benefits to this ASP solution. You are already outsourcing your technical support and customer service phone centers to Bangladesh; however, TA's AI can perform the same tasks 300% cheaper, and it can perform them 24/7. In addition, the AI could (for an extra fee, of course) be employed to translate business documents for you (it can currently only translate between English, French, and German, but TA is promising additional languages soon). It can filter spam, take surveys, generate product documentation, and do a lot of other work that your human employees are currently doing. TA claims that their AIs are indistinguishable from humans over the course of a normal conversation -- that is, in fact, their main selling point.
TA is offering a free demo period for its AI; you can call a 1-800 number, and talk to it -- so you do that. The AI identifies itself as "Bob"; it talks in a pleasant baritone. You chat with it for about half an hour, and it seems completely human. "Bob" doesn't talk much about his personal life, but he does reveal that he prefers cats to dogs ("cats are smarter !"), enjoys watching baseball, and likes jazz. In fact, you get into a heated impromptu discussion with "Bob" over music. While you don't manage to completely persuade him that Elvis is the true King, you can't help it but admit that "Bob" comes off as a fairly intelligent and friendly "individual" -- but then, your 60-minute demo period comes to and end, and "Bob" excuses himself and hangs up. An automated message from TA informs you that "Bob" is but a single instance of their massive AI cluster; it and other AI "personalities" can be configured and purchased directly from the TA website.
The Decision
You're pretty much ready to call up TA and order yourself a subscription to about 1000 "Bob" instances... but there's a problem.
How do you know that "Bob" is really a computer AI, and not some human phone operator that TA has trained specifically for their demo ? If TA is lying, and their powerful AI technology amounts to nothing more than an outsourced call center in Bangladesh, then they will not be able to handle any kind of a real workload. Real human beings do not scale easily; they are not available 24/7, and they inevitably end up costing more than machines. These problems are what you're trying to avoid in the first place !
The problem is, TA is very protective of their intellectual property. They won't let you download an AI to your local machine -- they will only lease you access to one, as an ASP solution. They won't let you tour their facility -- for fear of industrial espionage -- and their patents haven't cleared yet, so you can't look at them. TA's service agreements are wrapped in layers of labyrinthine legalese that even you can't fully comprehend, but you know that they mean business.
The Challenge
You'd like to know for sure (or, at least, with a reasonable degree of certainty) whether "Bob" and his "colleagues" are really AIs, or just humans pretending to be AIs. You need to make this decision fairly soon, because if TA is the real deal, and your competitors beat you to it, you'll be left in the dust. Unfortunately, the only way you have of verifying whether or not the AI is a real, is through talking to it on the phone, or conversing with it on instant messenger.
So... how would you determine whether or not "Bob" is an Artificial Intelligence ?
Caveat Emptor
Moderator: Moderators
Post #2
OK, good idea! But wouldn't the obvious test be directed towards the claims to scalability? If true AI was on offer then this would be the acid test - could, say, 1,000,000 simultaneous calls be serviced? These could be AI generated and would bog-down humans if they were all that was on the receiving end. I'm not sure if your scenario allows for this though as, realistically speaking, you would have to pay up-front for the capacity you require.
Post #3
Exactly. If you had that kind of money to waste on a DDoS attack, you'd probably just lease the TA AI on the off chance that it's real, and eat the loss if it turns out to be fake. But, you don't have that kind of money, so you have to come up with a better test :-)QED wrote:But wouldn't the obvious test be directed towards the claims to scalability? If true AI was on offer then this would be the acid test - could, say, 1,000,000 simultaneous calls be serviced? These could be AI generated and would bog-down humans if they were all that was on the receiving end. I'm not sure if your scenario allows for this though as, realistically speaking, you would have to pay up-front for the capacity you require.
Post #4
I just took another look at ALICE - the best example of AI that I could find on the internet (anyone know of a better one?). The same site also offers a rather interesting sounding igod but it doesn't seem to run on my PC 
(perhaps there's some deep significance to this fact 
). Anyhow, I'm stumped for any "Blade-runner" type test ("I'll tell you about my Mother!"). I look forward to other suggestions.
(perhaps there's some deep significance to this fact ). Anyhow, I'm stumped for any "Blade-runner" type test ("I'll tell you about my Mother!"). I look forward to other suggestions.
