MOUNTAIN VIEW, CA—As the push towards automation continues, more efforts are being made to make robots and machines capable of performing human tasks and appearing more lifelike.
After the success of the question-answering system, Watson, programmers have sought new areas of human service and interaction where artificial intelligence can take the place of other areas normally dominated by people. These include with problem solving in high-stress environments, such as airline customer service or automated sales calls. Additionally, common online arguments could potentially be entirely overtaken someday by artificial intelligence.
One area that has continued to baffle AI programmers is attempts to make a program that is logically consistent and also permissive of the pro-choice position concerning abortion. This problem has already reportedly cost tech companies $3.4 million in time and equipment, but they explain that it's always the same problem.
"Well, we begin with the basic premise that anyone who dislikes the pro-choice position hates women," said Deter Hamlin. "That coding is not hard at all. However, we really start to run into logic problems whenever the bot starts to define or value the fetus."
Hamlin shared that the artificial intelligence program is capable of repeating and citing semantic arguments, but that they just haven't been able to keep a machine from shutting down once it starts simultaneously making arguments like a fetus is a parasite, and it's not alive until its first breath.
"Watch, I'll show you," he said, turning on a speaker and microphone on a small computer. "Hey, abortion kills a living human."
The speaker then played a response originating from the machine, "A fetus isn't alive, man."
"Then why does abortion require killing the fetus?" responded Hamlin.
At this point, the speaker began playing several beeps and digitized screeching noises, and said repeatedly, "Does not compute! Does not compute!" and then began to melt where it stood before ultimately going silent.
"Yeah, they do that every time," said Hamlin. "I once got one to last for a full 8 minutes, but then it tried to say that a human is a human only if it is wanted, and then it combusted immediately. Good thing I had my safety glasses on for that one."
This challenge appears likely to continue as programmers struggle to maintain a coherent and consistent set of rules. However, Hamlin did share a promising work-around. "We may be able to bypass all of the logic problems if we have the program resort to insults and ad hominem attacks just at the point of losing to the facts and reason. Or another idea is to make all of the bots operate from a nihilist perspective, where nothing actually matters at all. That sometimes helps."
However, as the struggle continues, it appears that the pro-choice position will need to be argued by humans who are capable of operating without logical foundations.
Comments