AI: Threat or panacea?
-
It obviously hasn't now, but it will, and it won't remotely require being 'intelligent' in any strict sense that we might require to consider it an equal. So it'll happen long before that threshold is crossed. It doesn't take any real 'intelligence' to put an 'AI' in charge of weapons or weapons response systems. They just need to be able to take a lot of inputs and reach some level of confidence that something needs to be done and make it happen, very quickly. Some folks would argue that could be done now, and it could, but not in the same way. I could write a conventional program to recognize faces or speech, but it would be brutal and wouldn't likely compete with a DNN based system, where you need to deal with information that is incomplete and fuzzy. These types of systems, I would think, will be more likely to be 'trusted' with such jobs specifically because they don't depend on the programmed in prejudices of a team of software engineers. But that means that, like us, they can misinterpret the input and come to the wrong decision.
Explorans limites defectum
-
You said: "It obviously hasn't now, but it will". I'm not as sure as you are that "...it will". Before "...it will" we need to understand how we extract meaning from data. You might even have to explain what "life" is.
No, I meant it will be PUT INTO a position to do things detrimental to us. Humans will allow to do so. I won't have to take over, it'll apply for the job and get approved.
Explorans limites defectum
-
I've read all the replies. I thank everyone for their perspectives! I will give you some context and my answer to what camp I'm in. (NOTE: This became much longer than I anticipated so I don't mind if your reaction is TLDR.) I read a book back around 1980 entitled, "The Adolescence of P1". P1 is a reference to "memory Partition 1" - the privileged operating system partition. Thumbnail of the book: Computer Science student attending the University of Waterloo creates a program, giving it a mission to gain control of the operating system, hide itself, seek out routes to other computers, and gain access to "information". Said student submits the program and it immediately throws up an catastrophic exception and fails. Except that it hadn't failed. That was a smoke screen necessary to fulfill its directive to hide itself. The student assumes the failure is legit, gives up on his project and gets on with his life - graduating and eventually landing a job in the U.S.. Time passes, P1 carries on, follows the networks, expands the number of computers it controls, assimilates all the "information" it encounters, infects the computer at IBM that creates the operating system images sent by IBM to its customers, and P1 gains more and more resources and "information." Somehow (the process is never fully explained), P1 gains enough "knowledge" that it spontaneously becomes a "conscious entity." It does nifty things like detect that the U.S. authorities are onto it, and it infects the air traffic control computers and crashes a plane which kills the investigator. Eventually it finds its creator, and reveals itself to him. Further merriment ensues. It was a great story and it sparked in me the naive goal of replicating the university student's achievement. So my point is, I've been thinking about thinking and AI ever since. I have a book (not finished) entitled, "Insights on My Mind" in which I am in the process of writing down all that I've learned and the conclusions I've reached SO FAR. I'm not here to sell anyone anything. I'm just explaining how I've gotten to this point. Theologically speaking, I'm an agnostic. So I have proceeded with my AI research all these years based on the assumption that I cannot invoke metaphysical answers to the hard questions. That means that every element of my study has to be grounded in physical reality. The consequence has been that, if we are truly going to replicate human-level "intelligence" in a physical entity such as a digital or analog or hybr
Very interesting topic! People are a threat to themselves. When people have control over objects that can harm them then they better be careful and focus on what they are up to. This applies as much to AI as to a gun, knife, or a lathe. I regard AI currently as more of an advanced pattern recognition system and since I have witnessed first hand how the average software developer struggles to even get CSS to jump through the correct hoops I am not too worried about some self-conscious AI going berserk. Of course, if those same programmers are going to be fiddling with code that launches tactical nukes then I would be a bit more worried. I will also be driving my own car for now, thanks Elon. As you have alluded to there are more fundamental issues that we need to solve before even getting to anything that is going to approximate awareness or, heaven forbid, self-awareness. We know we have matter and we know we have consciousness. If consciousness is as a result of some configuration of matter then it is something we can cook up in a lab. However, if matter was somehow "created" by consciousness or is somehow "experienced" as "real" then it is a whole other affair. A simple concept such as "size" would seem to me to be problematic. If some mean-spirited self-aware AI were to create robots to annihilate us then exactly how "big" would these be? It would need to understand something that we all take pretty much for granted. It is a similar conundrum with the evolution of wings: how on earth would wings sprout out of no knowledge of how "thick" the air is and how "big" the wings need to be in order to lift the bird? If it is a matter of chance then what records this monumental event in the DNA that produced "wings" that could have the bird fly and then also keep those same wings around in the same configuration? Would another pair of wings not be even better? I mean, we have this in software development: "Oh, a 5 page document resulted in a successful system... then 100 pages would be even better!" For now I'm quite happy to have AI spot faces and listen to requests for stuff. Especially the voice recognition is handy for kids that can't yet write/type what they are after but they know that they would like to see a "fan collection".