Computers fail to fool humans
The undefeated Turing test
SIX ARTIFICIAL conversational entities (ACEs) have failed to fool interrogators into thinking they were human during a real-time chat.
Although all the ACEs passed themselves off as human to at least one individual, the programs did not manage to convince 30 per cent of the participants, thereby failing the famous 1950 threshold set by Alan Turing.
Indeed, the winning machine, known as Elbot, only achieved a 25 percent success rate.
According to Kevin Warwick from the University of Reading's School of Systems Engineering, the test was designed to "raise the bar in Artificial Intelligence ".
Warwick explained that while the machines weren't yet good enough to fool all the people all of the time, they were certainly capable of "fooling some of the people some of the time".
Warwick also noted that the conversational abilities of each machine was scored at an impressive 80 and 90 percent.
"This demonstrates how close machines are getting to reaching the milestone of communicating with us in a way in which we are comfortable", said Warwick.
Alan Turing, often considered to be the father of modern computer science, worked at Bletchley Park during World War II. He developed a number of techniques for cracking German ciphers, including the method of the Bombe, an electromechanical machine capable of locating settings for the Enigma machine.
In addition, Turing assisted US-based Bell Labs with the development of secure speech devices. µ
L'Inq
Telegraph
Comments
You dont need computers to fool humans
You can just repeatedly sell them the same software that doesnt do anything for them!When I was a kid you needed a typist to produce a document. Now everyone can make one and they're too busy writing theirs to read yours.
Productivity? FINR.
P. T. Warwick
"Warwick explained that while the machines weren't yet good enough to fool all the people all of the time, they were certainly capable of "fooling some of the people some of the time"."Showing that machines have become more proficient than Warwick in his main area of expertise. In other news it was reported that Warwick had failed the infamous "Robot Test": 0% of those asked believed him to be machine, robot, cyborg or anything other than some bloke tagged like a domestic pet pretending to be the Terminator.
This is one for the long haul
I heard one of the computers on Radio 4, and it was utterly pathetic. People cite the programs that have beaten Kasparov and other grandmasters at chess, but that - while very impressive - is a vastly simpler task. Once the algorithms are sorted out (as they have been) it's just a matter of watching the computers' superiority get steadily greater thanks to Moore's Law, etc.The Turing Test, in contrast, requires a grasp of colloquial natural language; humour; current affairs; human preoccupations such as sex, sport, money, and power; social interaction; and an immense array of "common knowledge". AFAIK there isn't even a promising strategy for doing all this yet. If a computer passes the Turing Test by 2020 I'll be surprised.
Even so
I wager that I'd be more comfortable chatting with the winning bot then chatting with a real live illiterate idiot smacking me repeatedly with LOLs and OMFGBBQ all the time.human characteristics
In order for a computer to act like a modern day human, it would have to be able to simulate arrogance, rudeness, stubborness, sarcasm, and the general desire to alienate and offend. If they did that, the goal would be achieved.If the person you are talking to is cooperative, helpful, kind, mild and thoughtful, how could they be human?
Not on the carpet! again!
Maybe some bright mind will adapt Elbot or the like into an Operating System or a Search Engine or DBMS.Then we'll leap forward in terms of human interfaces. But then, my desktop may start asking for a reach-around my Elbot.
It's easy...
...to fool 30% of humans, just make sure 30% of the humans are fools. I can throw a badly-trained MegaHal bot into #teenchat and watch it get kicked for its insolence, undetected, within 15 minutes.Anyway, a 25% success rate is pretty staggering. I'd love to see the thing in action - although the more realistic it gets, the less fun, I'm sure.
The old Voight-Kampf test comes to mind...
I wonder if the results would be any different if the testers were told to look for humans among the testees...Immagine the feeling of a human decoy testee when he gets tagged as a computer...
I stand corrected
That thing is embarrassingly bad - it managed to repeat canned phrases within a few lines. Barely better than Eliza, if that - in fact, the responses seemed almost completely randomized.How the hell this did anything but fail utterly I can't imagine.
What do you mean...?
..."Only" 25%?Remember Moore's Law? It's all about shifting intelligence so it's in Moore's domain. Seems to me we're most of the way there.
To the "chess is easy guy": funny how people move the goal posts all the time.
I remember the '90s when it was all "chess is more than algorithms" and "machines will never win against the world chess champion".
Now it's the exact opposite. I mean, if a *machine* can win at chess, then it must, by definition, be easy after all, right...?
What do you want to bet the same process took place in the case of every single field where AI rules today?
The simple fact of the matter is that there is nothing mystic about thought and sentience. There is fundamentally nothing to stop either from being automated. It's just a matter of time. And it would seem the shift could occur in our lifetime, after all.
End Times
This is a bit off-topic, but I'd like to comment about all the people (admittedly, none actually posed about it here) who think the Armageddon of the Bible will happen because of machines.I've been reading a bit about evolution and the idea that some humans(and, by extension, intelligent machines) could be considered better, and therefore have more of a right to lead than other people. But when you dig deep into the philosophy, survival of the fittest being the core of it, you realize that it's a load of garbage. If machines ever truly get the ability to reason, without prejudice, then they will either discover a reason for existence that has nothing to do with evolution or intelligent design, or they will decide intelligent design is the true answer and desire to discover the source of the designer, who doesn't necessarily have to be the same as the God of Jesus or the Islamic God.
The thing to truly fear is a computer taught to behave in the same selfish way as his creator.
Turing test
is useless as it is a test of humanity rather than intelligence. Designing a new sentient species, why don't we make it without copying all our flaws while we're at it.Don't hold your breath, even at our current rate of technological development it would take centuries, and I wouldn't be so sure we have that much time left, after all, the only thing pushing us forward and keeping us alive is our greed.
Ah hah
25%, good enough to be placed on the Republican ticket.drop the bombe
Turing did not invent the bombe, the polish did.http://ed-thelen.org/comp-hist/NSA-Enigma.html
eventually
Computers don't usually have typing errors in the chat box, while most humans will. Plus what human answers; "How was your weekend?" with "Please wait, while I research that."There is no fundamental reason preventing machinces from eventually passing the Turing Test; however, with each iteration, the expections move. But that's ok, they should be moving.
It seems that software development is lagging hardware development by quite a bit (maybe as much as 10 years). With Moore's law and Intel performing research on 80 core processors, and IBM claimig that with todays technolgy they could take today's room full of super computers and in 10 years time build a computer with equivalent processing power that would fit on the head of a pin; there is no doubt that one day computers will have vastly more processing than the human brain for the same volume.
I can't possible speculate on timeframes, but I bet it happens sooner than we expect...provided the ecconomy doesn't collapse into nothingness.