Artificial Intelligence Researcher: Modern AI is Not True Intelligence

“In a sense intelligence is the wrong word. Intelligence is a human attribute, but these are just computers. They’re doing exactly what they’ve been programmed to do.” source

Several times here at Catholic Answers Forums, I’ve seen people post questions about artificial intelligence and its implications for Catholic philosophy. In classic Catholic philosophy, intelligence is spiritual, not physical, an ability of the human soul that cannot be reproduced by physical systems. Since artificial intelligence has recently made leaps and bounds, some people are wondering if computers can eventually have human-level intelligence. This has even been put in the form of an objection to the spiritual: you Catholics always say that intelligence is spiritual, and therefore humans have a spiritual soul; but science has almost proven that physical systems can be intelligent too, so there goes one more argument for the soul!

As apologists, I think we need to develop resources to answer this kind of objection. And one of those resources may come from the above quote. It comes from Jonathan Schaeffer, an AI researcher. He’s not a nobody, either: his work has been so influential in modern AI research that the government of Canada has made him a national Chairman in Artificial Intelligence according to Wikipedia.

In the article I cited at the beginning of this thread, he was commenting on Google’s latest AI success: their DeepMind project has successfully beaten a top-level human at Go, which is arguably the most complex strategy game ever created.

This can help us show that modern AI researchers do not think they are replicating human intelligence. They are copying the brain in order to program their machines more intelligently, but this influential researcher emphasizes that the most sophisticated AI concepts are aiming at something different from human intelligence. They’re not just Beneath human intelligence, they are a different kind of thing altogether.

He’s not the only researcher to mention something like this. “Dr Simon Stringer, director of the Oxford Centre for Theoretical Neuroscience and Artificial Intelligence, said that AlphaGo and other deep learning systems are good at specific tasks - be that spotting objects or animals in photos or mastering a game. But these systems work very differently from the human brain and shouldn’t be viewed as representing progress towards developing a general, human-like intelligence.” source

Dr. Stringer apparently thinks that physical systems could theoretically have human-level intelligence, which makes him a witness for other side as well as ours. But he does point out that the best of modern AI research is well within the boundaries of typical computer programming and is not a low form of human-level intelligence.

If and when we ever talk to people about the spiritual nature of intelligence, we may soon find that people will start bringing up artificial intelligence as a counterpoint. They might ask about it because it troubles them and makes them think human intelligence is nothing special. In those occasions, it will be useful to have testimonies like these to show that we’re talking about different things and the best modern researchers know that.

Thanks for posting this. Computers lack any spiritual component and the creative abilities God gave man. However, self-learning systems can figure out most of the possible moves in a game like Go. The concern is that they will contain so much data storage and cross-referencing capabilities that they may be used to assist in human decision making. This would be a mistake since humans have other ways of thinking about solving certain problems. We have emotions and can be deceptive. AIs would be useful in solving certain types of problems, for example, in engineering. But they lack a will, desires or a sense of the future. Humans age. Technically, as AIs become more sophisticated, they will become smaller as breakthroughs in material science and optical memory storage are ongoing. They will know about time but have no personality.

Like any human invention, they could be used to solve problems that we would classify as bad, as well as good. The AI would not need to know why the problem exists. It only solves problems for whoever owns one.

Ed

The last time I worked with an artificial intelligence person (they now refer to themselves as cognitive) we ended up decided that anything that reacts to its environment and makes a decision counts as being ‘cognitive’.

So I showed him how we have code that uses an ‘If, then,else’ statement. He agreed that was cognitive.

news.bbc.co.uk/2/hi/technology/2226327.stm

Thank you. I will always complain when those funding this “who knows where the money comes from” for yet another useless social engineering project come up with a new fake. Fake animals, fake aquatic things. Fake is fake. It only does what it’s programmed to do. In the meantime, if fake robot cats were free, I wouldn’t take one. Ever.

bbc.com/news/technology-35321227

So this “accept our fakes as real” is a nonsense idea. Unless DARPA completes a Terminator, then I’ll consider the other implications.

Ed

I do not believe that intelligence is spiritual and I’m pretty sure it is not a teaching of the Church.

I’m basing that in part on what the Catechism says in two passages, assuming I’m understanding them correctly (which is always possible):

CCC 1705 is the most explicit: “[Man’s] spiritual powers [are] intellect and will.”

CCC 470 also seems to identify intelligence as a spiritual power distinct from the body: “Christ [had a] human soul, with its operations of intellect and will, and…[a] human body.”

See also CCC 330: “As purely spiritual creatures angels have intelligence and will.”

And also, remember that we will continue to have intelligence even when we (temporarily) no longer have a body.

How can intellect be a spiritual power unless intelligence is spiritual? How can it be distinct from the human body unless it is spiritual? How can pure spirits have it unless it is spiritual?

That’s why I said what I did. Unless I’m misunderstanding it, Church teaching holds that intelligence is spiritual.

I don’t agree that the CCC section you quoted say/teach that intelligence is spiritual.

Can you teach that something is a spiritual power without teaching that it is spiritual?

I don’t teach anything. The Church teaches and it does not teach that intelligence is spiritual.

Right now that statement is true, as we dont have a true form of AI yet, but who knows what the future holds, there are many R&D labs trying to find this holy grail, eventually we will probably see some form of pretty close AI be realized, we would not know if its spiritual until it could be asked the question, in fact, I dont think we could even speculate what its answer may be.

The question is, what if one day an AI is created that can so completely beat the Turing test that no one could tell it was an AI—or maybe it wasn’t even aware it was an AI itself? Would we all have to become atheists?

I only say this because there were a lot of things in the past that many Catholic thinkers would have rejected as possible because they seemed to violate what the Church taught, like Heliocentrism, Evolution, or the Big Bang (i.e. that it happened 14 billion years ago) that are now seen as being compatible with Catholicism. I’m not saying it will happen, but it’s worth noting that when when the people have said “This outcome is an impossibility” in relation to natural science on the basis of Aristotelian metaphysics, they’ve bee burned before.

I don’t think so, and one reason why is because the Turing test is not a test of true intelligence. Simple programs have already beat the Turing test by imitating adolescents, who are stereotyped to say unexpected things. Aristotle developed a theory of intelligence that is arguably a truer test than the Turing test: abstraction. In his theory, if I understand it correctly, abstraction is the ability to take imperfect things and then imagine perfect ones based on them. A classic example is that whenever we see a bent line we are comparing it against an imaginary straight line that has never actually existed, not even in a computer. The perfect line is abstracted from the approximations we see around us. Aristotle used this to argue that intelligence is spiritual: physical things have limitations that prevent them from holding infinite things in them. Only a spirit can conceive the infinite because it doesn’t have material restrictions. People can do that, while animals and computers can’t.

I think Aristotle’s test of intelligence is much more appropriate than Alan Turing’s, and it’s also much older and has been studied much more.

I only say this because there were a lot of things in the past that many Catholic thinkers would have rejected as possible because they seemed to violate what the Church taught, like Heliocentrism, Evolution, or the Big Bang (i.e. that it happened 14 billion years ago) that are now seen as being compatible with Catholicism.

Those things were seen as compatible with Catholicism both now and then. Catholics were the ones who laid the foundations of those theories – unless you count Philolaus’ heliocentric model, since he was before Christ. St. Augustine and St. Thomas Aquinas both said that animals evolved from slime and earth into their current forms over time, Copernicus and Galileo both had the support of the Church in developing their scientific theories, and the Big Bang was a Catholic priest’s theory before it was anyone else’s.

I’m not saying it will happen, but it’s worth noting that when when the people have said “This outcome is an impossibility” in relation to natural science on the basis of Aristotelian metaphysics, they’ve [been] burned before.

Point taken. At least at the present time, it seems as if the scientists are clarifying that their programs aren’t in the same category as real intelligence. The hardware they are using is approximating more and more of the brain’s structure, but the brain is not the same as the mind. And the scientists behind this are clarifying that what’s running their hardware is a program, not an intelligence. Several of the best in the business seem to be doing the necessary work to clarify the difference.

I don’t think so, and one reason why is because the Turing test is not a test of true intelligence. Simple programs have already beat the Turing test by imitating adolescents, who are stereotyped to say unexpected things. Aristotle developed a theory of intelligence that is arguably a truer test than the Turing test: abstraction. In his theory, if I understand it correctly, abstraction is the ability to take imperfect things and then imagine perfect ones based on them. A classic example is that whenever we see a bent line we are comparing it against an imaginary straight line that has never actually existed, not even in a computer. The perfect line is abstracted from the approximations we see around us. Aristotle used this to argue that intelligence is spiritual: physical things have limitations that prevent them from holding infinite things in them. Only a spirit can conceive the infinite because it doesn’t have material restrictions. People can do that, while animals and computers can’t.

I wasn’t speaking of the Turing test per se. I’m talking about the idea that you could put an AI in some sort of advanced robot body and said AI was able to interact with humans and display absolutely no noticeable difference—maybe they wouldn’t even be able to recognize or accept they were an AI and not an actual human. Even if such a thing isn’t what researchers are trying to make now, it isn’t inconceivable they might try at some point—there aren’t any laws against it.

Those things were seen as compatible with Catholicism both now and then. Catholics were the ones who laid the foundations of those theories – unless you count Philolaus’ heliocentric model, since he was before Christ.

The Church certainly didn’t seem to think so—though I admit they were incorrect

St. Augustine and St. Thomas Aquinas both said that animals evolved from slime and earth into their current forms over time

I believe there was an idea that certain animals could emerge from inorganic species, but that’s not the same as evolution, which posits that over billions of years, tiny changes lead to species changing into different species. Not to mention that at least the human body, under theistic evolution, can be seen as being derived from earlier animals (though the idea is that the human form was the intended end of all those before), and then infused with a soul as an act of special creation.

Copernicus and Galileo both had the support of the Church in developing their scientific theories,

Copernicus did dedicate his book to the Pope, but Galelio, no matter how you shake it, was imprisoned because he insisted that Heliocentrism was true, something which the Church couldn’t accept----and they might have been right that he couldn’t show there was a stellar parallax, but it doesn’t change the fact that they jailed him because it seemed his theory challenged the Bible.

[and the Big Bang was a Catholic priest’s theory before it was anyone else’s.

True enough, but I was referring to the fact the of the universe being very old (though admittedly it wasn’t a good example since geological records in the 19th Century already showed the Earth, at least wasn’t 6000 years old.

Point taken. At least at the present time, it seems as if the scientists are clarifying that their programs aren’t in the same category as real intelligence. The hardware they are using is approximating more and more of the brain’s structure, but the brain is not the same as the mind. And the scientists behind this are clarifying that what’s running their hardware is a program, not an intelligence. Several of the best in the business seem to be doing the necessary work to clarify the difference.

That may be true, but I wouldn’t rule out the possibility of some trying to do create a real intelligence in the future just to see if they could.
[/quote]

At the end of the day, I don’t really care if AI is “real” in the sense that “it does things the exact same way we do.” I don’t even think that is a particularly philosophically interesting distinction.

The interesting part, I believe, will be that it will lay bare the reality that religious people don’t have a meaningful test to determine the presence of a soul. If they did, then no machine would ever be able to pass as a human, since the soul test would always find them out.

Catholic theologians have long held that certain capabilities, so far unique to humans, indicate the presence of a rational soul. (Since the soul is not matter or energy, we should not expect that there is a way to “detect” it except via the behavior of rational beings.)

Should a machine ever display all the capabilities of a rational being, I believe we would have to conclude (just as with a sapient alien of biological origin) that God has granted that creature a rational soul. While some believers might be offended at the notion, there is nothing that limits the granting of souls to biological beings. It’s not our parents that provide human souls when they conceive us, but God in a special creative act. While it is so far reasonable to believe that God would not grant a soul to a machine, and therefore no machine would ever be able to replicate salience fully, we can hardly proclaim that He cannot. Therefore, should such a machine arise, the most reasonable conclusion for Catholics would be that He has.

Usagi

I agree that yours is a reasonable conclusion in the face of such a machine, however I believe that it will not be popular. I think that kind of machine will put the church between a rock and a hard place:

On the one hand they could make your conclusion. But I think that many people would reject it on the grounds that 1. humans are special and 2. there are a ton of theological questions that immediately need answers (e.g. is the machine’s soul in a state of sin? If you make a complex machine but don’t put the AI software on it, does it still get a soul? If you remove the AI software, has the machine died and been separated from its soul, or is it merely like a comatose human? Can the AIs get married if they invent a way to produce little AI offspring?)

On the other hand the Church could invalidate the previous ideas that souls give us capabilities. They would instead simply assert that we get souls because we are humans, and revelation says humans have souls, end of story. I think this will be more palatable to Joe Catholic, but it has the problem I previously mentioned. It means that we only know we have souls because of revelation, and that we therefore can’t know about the soul-havitude of things not mentioned by revelation. I’m not sure what happens to animal souls in this scenario. In any case, I think that this position would make the whole concept of souls seem an awful lot like a God-of-the-gaps in that it was forced to retreat from “souls explain certain capabilities” to “that whole capabilities thing was never official doctrine, we know about souls because religion says so.”

Let’s give that theoretical type of AI a name: Jarvis AI. I think the name is appropriate to take from the Iron Man movies, in which Jarvis starts out as sort of an enhanced chatbot voiced by Paul Bettany. In the early films, the character is able to carry on conversations that could pass for human. In the later films, Tony modifies the Jarvis program and puts it into an artificial body – the Vision. At that point, the program has become so person-like that no one can tell the difference, and the Vision becomes another superhero and an ally of the other Avengers.

Even if [Jarvis AI] isn’t what researchers are trying to make now, it isn’t inconceivable they might try at some point—there aren’t any laws against it.

That sounds reasonable enough: someone might try. And in fact, if my understanding is correct, a chatbot Could theoretically be good enough to use as a legitimate conversation partner. Here’s why I think so:

It seems to me that Siri and Google Now can figure out what we are looking for when we phrase our sentences a certain way. I think their foundation is solid enough to add an additional function: we could say something like, “Tell me one of Thomas Aquinas’s arguments for the existence of God.” (Just to use one example of a thing we could ask about.) Then the AI could search a database including his books to find his arguments for God’s existence, and select one at random to repeat back to us. Then we could say, “Tell me what some scholars have said about it.” And it could repeat that process with the new information, selecting two scholars’ thoughts on the subject at random and giving us the opportunity to hear more.

I think a chatbot could even be programmed to select sentences about a given subject without being directly prompted, so that it seems to be coming up with unique thoughts like people do. For example, suppose you asked about one of Thomas Aquinas’ arguments and then about what some scholars say about it. After hearing one, you say: “That seems reasonable to me.” A chatbot could, I think, use grammatical rules to determine what that sentence refers to, and select an appropriate response in a similar way to how it selected from a database of writings.

For example, it could say, “It seems reasonable to William Lane Craig too. He says…” and then continue the conversation. Chatbots already have the ability to stay on topic. If we ask who Hilary Clinton is, and then ask “How old is she,” they’ve been programmed to respond with info about Hilary again. So if we ask for certain information, and then say a reaction to it, I think they could be programmed to select a new sentence that stays on topic but adds something new, and thus appear to be coming up with unique thoughts. In reality, they would only be pulling from what other people have said, but in my opinion that would be “close enough” to to how we interact with people. In that way, I think chatbots could pass for acceptable conversation partners.

The Church certainly didn’t seem to think [that heliocentrism was compatible with the faith]—though I admit they were incorrect… Copernicus did dedicate his book to the Pope, but Galelio, no matter how you shake it, was imprisoned because he insisted that Heliocentrism was true, something which the Church couldn’t accept…they jailed him because it seemed his theory challenged the Bible.

It may be true that Galileo was imprisoned for insisting that heliocentrism was true, but consider this: it seems to me that Galileo’s theory was promoted by the Church until he got in trouble, as well as Copernicus’s theory. Lectures on it were favorably received by the Vatican, and Vatican officials helped promote Galileo’s theory. This was Before he got in trouble. After he got in trouble, I think they made it clear that they were willing to accept his theory if he could prove it by observation and mathematics. St. Robert Bellarmine is an example – this book cites his comments on the subject. To me, the Church’s behavior both before and after Galileo got in trouble together show that the Church didn’t think the heliocentric theory was incompatible with the faith. I think it was Galileo’s dogmatizing that got him jailed, not his theory by itself.

I believe there was an idea [among the Church Fathers] that certain animals could emerge from inorganic species, but that’s not the same as evolution…

I think I see what you mean: the Church Fathers’s theory was different from inter-species evolution. That seems true, but the difference doesn’t seem to be in the category of matters of faith. They believed that each species developed slowly from slime – then I think they Could have believed that dogs, for example, developed slowly from wolves. I doubt they would have condemned anyone for that seemingly minor modification.

geological records in the 19th Century already showed the Earth, at least wasn’t 6000 years old.

Once again, I think Catholics laid the foundations for that, and plenty of them believed in an older earth without getting in trouble.

I wouldn’t rule out the possibility of some trying to do create a real intelligence in the future

I agree. For myself, my philosophical beliefs convince me that true AI is impossible, and my understanding of Church doctrine agrees. Of course, my understanding might be wrong.

We “don’t have a meaningful test to determine the presence of a soul” (I think you meant to indicate detection of rationality instead of soul) in the same sense, in which we “don’t have a meaningful test to determine if the program ends or runs forever”.

That is, yes, we can afford to assume that we do not have an algorithm that would infallibly find out if something is rational. For likewise, we do not have an algorithm that finds out if an arbitrary program will run forever (and it has been proved that such an algorithm does not exist). But it does not mean that there is no difference between programs that end and programs that run forever - and in most practically important cases we are quite capable of finding out what is going to happen with a given program. And similarly, there is a real difference between rational and non-rational, and in most practically important cases we can easily see which is which.

The real issue is whether an AI could ever truly grasp a concept, understand the “form” of something the way a rational mind can, and not simply carry out algorithms.

DISCLAIMER: The views and opinions expressed in these forums do not necessarily reflect those of Catholic Answers. For official apologetics resources please visit www.catholic.com.