Soon a robot will ask you if you want fries with that. Computer screens already do, in some locales. How does that make you feel?
Hungry?
That could be taken more than one way. Hungry because you like fries and relish the idea of them rolling out of the fat efficiently and quietly. Or hungry because you lost your job asking the question the robot is now asking.
The Pontifical Academy of Sciences convened “Power and Limitations of Artificial Intelligence“, and its partial purpose was to ask hungry-like questions. For instance, if we could invent AI-robots to do the work of whatever it is those poor slobs do in soul-sucking cubicles and in the meeting rooms of dreary office buildings, we will have done mankind a great service.
But never mind economic questions, which are easy to sort. Let’s ask the harder questions, presented in the same order as on the Vatican’s site.
What is the state of the art in AI software and machine learning?
Machines don’t “learn”; they just adjust values of parameters in man-created equations. The adjustments themselves flow along pre-determined lines. They are not alive.
Can all aspects of brain function be mimicked by artificial systems?
Not in the sense that a machine will have an intellect and will as we do. And it’s even doubtful we can replicate the fine behavior of the quantum mechanical interactions happening when neurons meet. We can simulate some level of detail, of course. But simulation isn’t replication.
Will machines soon surpass us in all domains of human competence?
No. Emphatically no. Not only no but…listen. Our intellects and wills, as was proved many times in this series, are incorporeal. They are not bodies. They are not made of physical stuff. They therefore cannot be created by us. Machines can never possess rationality as we do. Machines can’t be us.
On the other hand, machines have, from day one, surpassed human competence. That’s why we have machines! You cannot run as fast or as far an automobile, nor calculate digits of π as quickly as a calculator, and on and on.
What is the proper form of mathematics that may capture the operation of minds and brains?
Hint: linear regression isn’t going to be it. Since the intellect and will and incorporeal, and thus without material dimension, and thus indestructible, we’re dealing with some form of infinity. Which? Counting infinities? The continuum? Something higher? I have no idea. Equations that can simulate gross and approximate brain and nerve chemistry have already been found, and will improve. But what equations show how the incorporeal interacts with the corporeal? Can they even be found? I doubt.
What is consciousness? Could a machine be endowed with an artificial consciousness?
Animals have consciousness, and so do we, and plants and rocks do not. But we have more than consciousness. We have intellect and will, which are above (if you like) mere consciousness. The Catholic Encyclopedia says consciousness “cannot, strictly speaking, be defined. In its widest sense it includes all our sensations, thoughts, feelings, and volitions–in fact the sum total of our mental life”; it goes on to discuss these aspects.
See also this series by Feser: Nagel and his critics.
What would it take for a machine to possess a sense of self?
A miracle. Or an angelic fraud. Unless the questions means sense of self of the same sort animals have. It may be possible to create a biochemical machine which is an animal. But that sounds like genetic engineering. Cross a dachshund with a dove. Result? A flying sausage with a sense of self.
Will intelligent machines soon pose a danger to humanity?
Again, from day one, machines have been a help and hindrance. Cars crash, calculators give wrong results, and so on. So it won’t be a surprise to have a programming glitch in the new robot kitchen helpers that mistake the range of allowable things to chop.
Is it possible to design and construct an intelligent robot endowed with an artificial sense of ethics?
No. It will only do what it’s told. That can result in complex programming which fools some into thinking they are dealing with an ethical machine. The Turing test is no great thing. To have ethics, there must be morals, and morals imply sin, and sin is something that roosts in the will and intellect. So we’re back to the same problem.
How can we enhance the humanitarian uses of artificial intelligence and robotics, in particular in the field of education, health and emergencies?
Through massive government spending, of course.
Undoubtedly many people much more learned than I will weigh in on the particulars here, but I want to focus on just one phrase in Matt’s piece:
“…sin is something that roosts in the will and intellect.”
I finally now understand the passage, “before the cock crows, you will deny me three times.”
Bah-DUM-bump.
Speaking from the perspective of a former research scientist, professor, and scholar in Artificial Intelligence, I can say with confidence that if the gnomes in the Vatican read Roger Penrose’s book, The Emperor’s New Mind, they will discover they have little to worry about.
RE: “Soon a robot will ask you if you want fries with that.”
Or, it will say you can’t have fries with that: http://www.youtube.com/watch?v=bdqIJabjTJk
What your calling “incorporeal” could also be called a “projection.” But it’s still physically explainable. A movie is a projection of film, without the film and the projector, no movie, but the movie still exists, as you would say, in potential. And it is, of course, perfectly explainable, photons and screens and context and so forth. Certainly, someday there will be robots who can experience novel projections that we call thinking.
JMJ
…” What is the proper form of mathematics that may capture the operation of minds and brains?
Hint: linear regression isn’t going to be it. Since the intellect and will and incorporeal, and thus without material dimension, and thus indestructible, we’re dealing with some form of infinity. Which? Counting infinities? The continuum? Something higher? I have no idea. Equations that can simulate gross and approximate brain and nerve chemistry have already been found, and will improve. But what equations show how the incorporeal interacts with the corporeal? Can they even be found? I doubt.” …
Is there a reason why you have totally ignored the most famous book on this subject, Hofstadter’s ‘Godel, Escher and Bach’?
I only ask because he gives complete answers to all your questions in it, which renders your item rather redundant…
Back in the 1970s I studied this and we used to joke that we saw lots of genuine stupidity but very little artificial intelligence.
Dodgy Geezer:
“Is there a reason why you have totally ignored the most famous book on this subject, Hofstadter’s ‘Godel, Escher and Bach’?
I only ask because he gives complete answers to all your questions in it, which renders your item rather redundant…”
If they gave COMPLETE answers to those questions there would be algorythms in the book that the current AI world would be using. There are not. They gave something you interpreted as complete. Maybe you could take a stab at telling us why those complete answers have not resulted in useable software?
kuhnkat
If they gave COMPLETE answers to those questions there would be algorythms in the book that the current AI world would be using. … Maybe you could take a stab at telling us why those complete answers have not resulted in useable software?
Define ‘usable’. There are practical applications of neural networks embodying the first two levels described by Hofstadter (neuron and interconnections between them). It currently is not possible to reach the complexity of the brain (even simple ones) in both number of neurons and the number of interconnections. Just as it is not possible (currently, anyway) to have anything but small solutions to the travelling salesman problem. Does this limitation imply the mathematics behind the TSP is incomplete?
The answers are complete but not all questions have been answered. There is a difference between complete answers and complete sets of answers.
“And it is, of course, perfectly explainable, photons and screens and context and so forth.”
If by “perfectly” you mean it explains everything *except* the ideas, story, intended meanings, etc – which IS the movie – then I agree. The film, projector, protons, screens are not the movie.
I can’t resist chiming in here.
1. What evidence do you have that real-time quantum level interactions are required to mimic/recreate intelligence?
2. Machines can learn. Sure, they are based off of equations– so what? They learn to do some task, and continue to learn as the environment (and task) changes. The proof is in the pudding Briggs.
3. Machines can/will outperform humans in most cognitive tasks. It’s not impossible to get a computer to learn the English language– you just need a really powerful computer to make it practical.
4. The proper form of mathematics (for computers) is Boolean logic. It’s the only type computers work with. The programming languages and calculators are only there to let people better work with the computers.
5. Self aware/sense of self– easy. if (self.running == true). Self aware in a way you would recognize as human? Prove to me that you are self aware, and then we’ll talk tests.
6. Ethics– yes. Probably better than most people. Most people do not understand their own sense of ethics, let alone other peoples. Human level language is too ambiguous to even talk about it effectively. Just look at how bizarrely worded the law is.
7. Dangerous? For sure– but no more dangerous than the programs being executed the old fashioned way by bureaucracies. Do you expect empathy and ethics from the tax man? Laws / Rules are no different than computer programs– the processing medium is all that changes.
Again, these are just tools, and like all tools are easily misused and misunderstood.
The modern AI/Machine Learning landscape (if you ignore the buzzwords) looks like Bayes and Brieman fell into a blender and created something magical. Very effective, but not overly good at explaining ‘why’ to use mere humans. If accuracy is good enough though (better than an expert human), why becomes less important.
No assumptions of underlying distributions necessary.