Terminator Syndrome

What is your first reaction to hearing about an advancement made to AI research that "brings us one step closer to machines that think like we do?" For me, it's "cool! Now how do I do one better?" But for many, that response seems to be "so is Skynet enabled yet?" or "so we're one step closer to the Terminator?" I see this as a problematic response. Yes, something like Terminator needs to be thought about, but it's a jump in technology that ignores the gradual, if rapid, steps that need to be completed for such a task to be accomplished.

                One thing people seem to assume is that an AI would find humans to be flawed and they need to be removed. AIs need to be written and tested, which implies that a predetermined action will be taken for a certain input. It’s not necessarily good for AI development, but it is something that can be improved over time. This means that if there is a chance that a AI might say "kill...any moving object" (an basic set of commands), there is a high chance for that to be disabled and for the command "kill" to require human input and/or verification. There is also that word, time. Developers and engineers instinctively try to reduce the amount of time they need to spend on something. For complex learning applications, this is done by in two ways: training (running through a set of predetermined input and output) or by pregenerated the data and reusing it each time.

                Unless a researcher wanted to kill everyone, the tests to be learned from and pregenerated data wouldn’t have something teaching AIs to kill people. I’ve always viewed AI training and implementation as something that provides a better understanding of the concept of learning. If people don’t like guns and violence, then instead of removing the weapons (as then it becomes a replace weapon X with weapon Y), determine why that was determined in the first place. Just as has been discussed with the accountability of self-driving cars, AIs and robots may be getting subconsciously viewed as “if it does something wrong, who is accountable” and if that AI determined to do something controversial, what does that make the AI? A human, an intelligent being?

                Should the reaction to improvements to AIs be “o no, will this become a Terminator and destroy mankind?” or “how do we handle the accountability of a machine that may plan out reactions in a way that it’s designers didn’t plan for and/or how can we prepare its initial ‘memory’ to not consider something as vile as killing another or even injuring it as a proper reaction?” It may be something better suited to sociologists and physiologists then scientists and engineers. Because, while the above post was scatter-brained, it is a legitimate thought to work through. If a movie like the Terminator is the only reason for society to fear AI and robot development, then the researchers simply need to provide proof that their AIs are trained to NOT look at killing and injuring as viable responses to events. If it is a more fundamental, if higher-thinking, fear of accountability and understanding thought processes, then the same “don’t injure or kill” AI training and data preparation should be done, but it should also be traced to determine why an AI might even go down that path. Both of which may provide a more fundamental response to why an individual might go down that same path and a possibility of determining a way to bring a person back from the brink. Is the AI the Terminator, or are we?