4
$\begingroup$

This question already has an answer here:

As I understand it, the way our brain works is that in a given situation our brain is bombarded with several possibilities extremely quickly, and using experience and our own intelligence we choose one of the several possibilities.

Is this true? If so, then can we create an AI that will talk to us like we talk to each other?

$\endgroup$

marked as duplicate by Serban Tanasa, Magic-Mouse, TrEs-2b, Mazura, HDE 226868 Oct 30 '15 at 0:07

This question has been asked before and already has an answer. If those answers do not fully address your question, please ask a new question.

  • 1
    $\begingroup$ We've managed 3 answers in 10 minutes, all saying pretty much the same. Whoo.. $\endgroup$ – Joe Bloggs Oct 29 '15 at 15:47
  • $\begingroup$ You understanding seems misstated. The brain in constantly exposed to inputs not possibilities, the brain then reacts to them and my trigger nerve cells and cause a physical reaction. Your question is three questions you should split them up. Can a computer understand spoken language, natural language processing, can it carry a conversation, the classic Turing test, and can it speak, speech generation. $\endgroup$ – sdrawkcabdear Oct 29 '15 at 17:12
  • $\begingroup$ From an absolute standpoint, we should be able to create an appropriate full scale, fully functional model of the human brain in code once we understand the full scale and functionality of the human brain. It all eventually comes down to basic blocks of one sort or another acting and reacting in accordance with some basic rules -- and that's the level we need to get it down to before we can make AI that are truly on par with human minds in every way. Anything less is just a behavioral model. The problem is, we don't really have that knowledge currently. $\endgroup$ – Carpe CM Oct 29 '15 at 18:23
  • $\begingroup$ If you rate intelligence on the ability to play chess and Jeopardy... too late; we lost. $\endgroup$ – Mazura Oct 29 '15 at 20:03
  • 1
    $\begingroup$ That's what the microbes asked, about 500 million years ago, when they created multicellular life. $\endgroup$ – Serban Tanasa Oct 29 '15 at 20:42
10
$\begingroup$

This is a question that's taken decades of research by some very intelligent people and still has no good answer.

I'll start with your question about the way the human brain works. While your point about being presented with several possibilities and picking the best one is true of part of the brain, you completely miss the rest of the complexity involved, namely how the brain gets the information about where it is, and how it makes the decision about what to do (let alone how it stores previous information). All these other components lead me to make the following statement with absolute certainty:

However complicated you think your brain is; it's more complicated than that.

Now onto the question about AI. 20/30 years ago, when we'd managed to get computers to solve the 'hard' problems like playing chess and solving differential equations (please note that these are considered 'hard' problems because the human brain isn't optimised for them, and they played into the preconceptions of intelligence held by academia at the time), AI researchers said 'oh, anytime now we'll have AI. We've only got the easy problems like speech and image recognition left'

These were considered easy tasks because the human brain has had tens of thousands of years (more, if you consider our full evolutionary tree) of optimising the structures of the brain to do them. If you actually think about it though, even doing something simple like looking at a cat and thinking 'that is a cat' is insanely complex. We think it's easy, because we're built to do it, but in actual fact we've got millennia of trial and error behind us, driven by the small mammals that had to look at predators and go 'that is a dinosaur' to avoid being eaten. We've had millions of years. AI research has not had millions of years. This is known as Moravec's Paradox

But the answer to your question is actually fairly simple: Yes. We will. As long as humanity keeps driving towards having an 'intelligent' AI, we're effectively keeping up an enforced 'evolutionary' pressure on the various prototypes and AI models to develop. Researchers will combine various parts of models to create better AI's (roughly analogous to breeding) and the weak will be discarded, while the stronger AI's will continue to contribute to the next generation.

In that manner, we should quickly (for a somewhat geological value of quick) start to see AI's that exhibit traits that humans want, like being able to hold a conversation or make a decent cup of coffee martini. Whether we see anything else (like true free will) is a matter for a different debate.

$\endgroup$
  • $\begingroup$ Also worth considering that, even with all the functionality hard-wired into the brain, it takes many years of development to go from newborn to someone capable of holding an intelligent conversation. The first AIs that can rival general human intelligence are likely to require a similarly lengthy developmental period, with exposure to the same kind of breadth of input. $\endgroup$ – Dan Bryant Oct 29 '15 at 19:59
  • $\begingroup$ By the logic of computers not having as long to get to the point of intelligence that we are at and with more time they will get there, wouldn't that imply by the time they get there we too will evolve intelligence past that point. tldr; if time is the defining issue then wont we always be millions of years ahead of them? $\endgroup$ – DasBeasto Oct 29 '15 at 20:11
  • $\begingroup$ @DasBeasto Certainly not. While biological evolution has taken a long time, computational evolution has occurred over a matter of a few decades. Once we have the processing power to fully run a learning AI, it may well overtake our own intelligence in mere years. $\endgroup$ – phyrfox Oct 29 '15 at 20:20
  • $\begingroup$ So basically, computers are already better at us as "thinking about" just about all the stuff that requires heavy thinking, but they tend to get hopelessly lost on things that are so natural and fundamental to us that we don't even have to think about them. $\endgroup$ – Mason Wheeler Oct 29 '15 at 20:45
  • $\begingroup$ @phyrfox: Possibly. Or it may prove that an AI complex enough to think as we do will never be intelligent enough to comprehend it's own cognition, because it's using so much of it's resources just to think. We're having immense trouble making highly specialised brainlike processes occur, there's no reason to believe that an AI would be any more capable than us at doing that. It's an asymptotic thing, the closer something comes to being as intelligent as us, the harder we (and it) will find it to make anything more intelligent. $\endgroup$ – Joe Bloggs Oct 30 '15 at 9:59
3
$\begingroup$

While human brains are more complex in many ways than our most advanced modern computers, there is no scientifically justifiable reason to think that anything a human can do is simply impossible for a sufficiently advanced computer. We don't know exactly how long it's going to take, as we don't fully understand the human brain, but based on our advances to date we can get an idea that it could well be within 50 years that we get a a very interesting two-way conversational AI.

What helps this is that the human brain is not 100% for conversation. Much of the brain is used for every other thing a human does, from sight and hearing to movement to keeping us from falling over dead. While it should logically take far longer to replicate a human brain, there is no reason we we can't have various degrees of intelligence and behaviors far sooner than what it would take to have the first fully artificially created human brain equivalent device.

Indeed, in many ways we already have recreated many otherwise "human only" or even "beyond human" abilities. We have computers that can recognize faces even across dozens of years of aging, translate human language with some facility, navigate through 3D space, and many other highly difficult tasks - like winning on Jeopardy or at Chess.

The trick is that we don't have computer systems that can do all of that at once, weaving them all together seamlessly. That's just too much for our present level of development, and there's still plenty of things humans can do intellectually that have no equivalent computer expression. Some things we just haven't figured out, and may not figure out for many, many years.

Now, if you consider intelligence as the full range of human intellectual ability, that's going to take a whole lot of advancement. But if you accept the concept of "materialism" (the philosophy, not the "I love owning stuff" concept) as opposed to something like "dualism", then it's just a matter of time. If there is no magic pixie dust that makes humans intelligent, there's no reason a silicon-based thing can't be developed to have equal or greater intelligence.

How long is that going to take? Beats the heck out of me! It's not possible to know, because we don't know how hard the problems are to solve until we've pretty much solved them.

$\endgroup$
2
$\begingroup$

I do not think that we will ever develop an artificial AI that can rival human thought in its most important facet.

Mind you, we already have systems that can surpass humans in many areas such as have been mentioned by other posts here - solving integral equations, computation, applying game theory, etc. But these are all applications of deductive or even inductive reasoning. There is another type of thinking that the human mind can do which is beyond computational ability and which I think will be forever beyond the ability of computers. This is abductive reasoning, or creative thinking.

Think for example of Max Planck. Faced with the ultra-violet catastrophe, a problem that was baffling the physicists of his time, he came to a point where he said he was ready to forsake every previously held conviction about physics. It was only then that he came up with the equation that would give birth to quantum physics (E=hfc). This equation breaks with everything we knew to that point about physics. Its implication, as would later be worked out, is that energy naturally occurs not in a smooth gradient, but in chunks - quanta. That is, energy has particle properties.

That sort of leap that denies past knowledge and gives birth to a paradigm shift, it is the same sort of "thinking" or creating that goes on when an artist composes a strikingly beautiful piece of music or astounding painting. A conceptual entity was not there, and then it was, in the mind of its creator. This is where our mind goes beyond our brain. It is not a matter of being able to do more computations per second, or store more bits of information - it is not a quantitative difference. It is a qualitative difference between the human mind and the computer brain.

Think of it like this.... Consider an average human composer who has studied music theory and is able to consistently and mechanically produce acceptable music. Now consider Wolfgang Amadeus Mozart who composes page after page of flawless music, music that broke with tradition and yet is still admired centuries later. What is the difference? Will we ever be able to teach a machine to be a Mozart? We can't even teach a human to be a Mozart or a Max Planck; what would make us think we can build a machine capable of competing, much less surpassing them?

How do we impart the ability to create, to make an unwarranted logical leap, to create something out of nothing? This is the thing that will forever be beyond our ability to build into an automaton.

$\endgroup$
  • 2
    $\begingroup$ I think this answer presupposes that a machine AI is incapable of eschewing the knowledge we and experience give it. A machine AI can be programmed to question everything and form its own hypotheses - ("forsake every ... conviction...") and then question those. The human condition to think we know what the answer is does not have to be inherent in an AI. I would argue that allowing it to understand evidence and form its own conclusions (all the while gathering more evidence and evaluating that) is what will allow it to supersede our knowledge and help us learn more about the universe. $\endgroup$ – Tracy Cramer Oct 29 '15 at 20:21
  • $\begingroup$ @TracyCramer - It is easy enough to question what we know. The hard part is to then make the intuitive leap to the right answer, the right paradigm that provides an answer to a previously unanswered problem. Among the billions of combinations of notes that a composer could put together, and that mediocre composers do put together mechanically, what is it that allowed Mozart to jump, time and again, to the "right" combination without hesitation. This is the same type of creative thinking going on in Planck and actually in all of us, so often, we take it for granted. $\endgroup$ – user11864 Oct 29 '15 at 21:00
  • 1
    $\begingroup$ I agree that a computer just randomly putting together musical notes would not create a Mozart sonata. And I agree we don't understand how the creative process works - yet. But, I would offer that it is simply a matter of time before we understand the brain well enough to mimic all of it. Or, sooner, I believe, we will create a machine that can teach itself how to be creative in its own way, that's just mho though. $\endgroup$ – Tracy Cramer Oct 29 '15 at 23:44
  • $\begingroup$ @TracyCramer - You are assuming that the mind is a product entirely of the brain, or the human body. I assume we are more than the matter we are made of. And making a machine that can teach itself how to be creative is squaring the problem. We'd have to create a machine that can create creativity - an equal or tougher proposition. But, I like the way you think, and it will be exciting in the coming decades, to see if technology begins to tell us - can we create creativity? Does our power to create go that far? $\endgroup$ – user11864 Oct 30 '15 at 13:39
  • $\begingroup$ That's an interesting question. Another thing I just realized we were bringing into the conversation but not actually saying it was emotion. Mozart created music that appeals to us. Sculptures, paintings - they all have some aesthetic to them - something that makes us 'like' them. Having a machine do something that appeals to us may not be the point at all. I think your reference to Planck and scientific leaps are probably better suited to this question of creativity. Thanks for the discussion. $\endgroup$ – Tracy Cramer Oct 30 '15 at 16:25
1
$\begingroup$

No. It's not true. The reality of how our brain works is enormously more complicated than that. There are small subsections of rat brains that we have determined can be well modeled the way you describe, but the brain itself is actually so much more complicated that it almost defies capturing in words.

Create an AI that can talk to us? ABSOLUTELY. Compared to the greater challenge of making a thinking machine, conversation is actually a relatively simple topic. In fact, it was the subject of the most famous test of AI in the world, the Turing Test. Every year, many computer programs try their best to communicate with a panel, who try to guess if they are computers or humans (half of the "programs" are actually humans behind the scenes acting like programs). Often, some of them do remarkably well. Verbal intonation is still tricky. The Turing Test depends on text still, but we're already on the road.

Can we make an AI think? That's an open ended question. Science has yet to provide an answer, and science fiction has gone every which direction with the idea. We honestly aren't 100% in agreement as to what it actually means to "think," so it's been hard to extrapolate to see if its in the limits of science and engineering's capacities.

$\endgroup$
  • 1
    $\begingroup$ What about the brain is so complex that it defies capturing? A single neuron is complex, but we can fairly accurately describe what it does. Everything else is a matter of scale and connections. $\endgroup$ – ckersch Oct 29 '15 at 18:41
  • $\begingroup$ @ckersch Scale is tricky when you deal with interconnections. Its estimated that the brain has 86 billion neurons, which is a lot. The synapse count is estimated at 150,000 bililon synapses (150,000,000,000,000 synapses). Also, just identifying that the neurons are connected is not enough to really capture what the brain does. The analog timing within those synapses is often critical as well. If we were to fully capture the brain, there would likely be an answer to the philosophical question of physicalism vs materialism. That debate still rages to this day. $\endgroup$ – Cort Ammon Oct 29 '15 at 19:19
  • $\begingroup$ There's sections of the brain that operate as simply as you say. Then there are sections, such as the hippocampus, where we simply have to shrug our shoulders and say "we think it helps us with memory... somehow." That and we're starting to learn that there's important factors that aren't even captured in the synapses themselves, such as the recent studies suggesting some memories may actually be buried within a single neuron. $\endgroup$ – Cort Ammon Oct 29 '15 at 19:21
  • 1
    $\begingroup$ In practice, it's probably better to think of the Turing Test as something human interrogators routinely fail, rather than as something conversational AI routinely passes. $\endgroup$ – Russell Borogove Oct 29 '15 at 20:24
  • $\begingroup$ @RussellBorogove I brought it up because the OP explicitly asked about an AI that talks to us like we talk to each other. I could have dug into the much unknown parts of communication, or simply point out what we already do. A discussion on the complexities of linguistics and emotion from a computational perspective could take days. This took minutes =) $\endgroup$ – Cort Ammon Oct 29 '15 at 21:30
0
$\begingroup$
  • Yes, you can make an AI emulating the most people

  • No there always will be a person whose behavior is impossible to predict or emulate using a machine. You would need a hypercomputing device (which is physically impossible).

See Thomas Breuer "Impossibility or universally valid theories".

$\endgroup$
0
$\begingroup$

YES.

What you are describing is basically true, in other words I get the general idea of it. Lots of sensory goes it, it gets processed, then decisions come out.

Bioengineering erases the boundary between classical computers and the human brain. In the same way that scientists grow living tissues in laboratories for grafting and even organ replacement, we could grow brain tissue. And similar to how a mouse brain on a petri dish can fly a plane, lab-grown brain tissue could potentially be used to make intelligent computers.

The engineered brains will eventually become "better" than human brains, either by making them bigger, more efficient, more specialized, or a combination of all these things.

$\endgroup$
0
$\begingroup$

Yes. But not only will it rival our intelligence it will surpass us and quickly eclipse our ability to comprehend what it tells us.

An AI does not have to have the preconceptions and limitations we humans do.

The most important abilities an intelligence needs is to learn (input and memory), organize knowledge (relevant recall) and reconfigure its own programming.

We already have many algorithms for deduction and reasoning so once we figure out how to get a machine to do those things an AI will quickly emerge.

And once it learns how to learn better (and leaves behind our feeble efforts at programming), it will become an ASI (Artificial SUPER Intelligence).

$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.