AI has the mental capacity of a four-year-old with no common sense

David Tom

Posts: 149   +3

ai mit artificial intelligence conceptnet

The advancement of artificial intelligence (AI) has been truly impressive; however, it doesn’t look like machines are getting anywhere close to a human’s level of understanding. To learn more about the failings of current AI systems, an AI developed at MIT known as the ConceptNet4 was given a standardized IQ test. How did it fair? The results indicate that it has the mental capacity of a four-year-old. That being said, a parent should be more than concerned if their child exhibits any of the same patterns as this MIT prodigy.

What separates the ConceptNet apart from traditional computers is its ability to find connections between different things. For example, any computer could tell you that Tom Brady is a football player, but ConceptNet would also find the relationship between him and his teammates. According to ExtremeTech, not only can the AI regurgitate hard facts, but it can respond to more involved questions regarding these relationships.

To assess the AI’s cognitive abilities, the program was subjected to the Wechsler Preschool and Primary Scale of Intelligence Test, a popular test given to young children. Like most children, the ConceptNet scored better in some categories than others. It was excellent in identifying similarities between items and also demonstrated an impressive vocabulary.

Unfortunately, it exhibited poor performance on comprehension-based questions, indicating that it lacked common sense. For example, it might know the exact freezing and boiling point of water, but it cannot understand the difference between hot and cold. It would also have trouble answering a simple question like "Why do people shake hands?" Answering this type of question involves human intuition and experience; two qualities that all AI systems lack.

At this point in time, researchers have yet to find a way to incorporate common sense into AI. It’s easy to get a machine to recall basic facts, but we are unable to have it draw any conclusions from this knowledge. Unless this hurdle is overcome, then artificial intelligence will never be anything more than an extremely peculiar four-year-old.

Permalink to story.

 
Well... let's be fair. It does take a 4 year old 4 years of learning to be as smart as a 4 year old. Should we expect to be able to make a computer that can just immediately be as smart as an adult? This is a newborn computer that can score like a 4 year-old. A newborn baby literally doesn't know up from down or night from day. A newborn doesn't even know they have hands and feet.

What they should be trying to do is make a computer that can learn, remember, and be curious. They should make a computer that can learn like a human, not act like a human.
 
mental capacity of a four-year-old
Most 4 year old's don't know the freezing point and boiling point of anything.
Common sense is learned, and some four year old's have much more than others.
The lack of common sense for a human is reflected upon by their parent(s). Yet.
The lack of common sense for a robot is reflected upon by their programming (Initially) beyond that, it is up to the cognitive learning abilities.

Either way, I'm impressed, and could see a robot with the "Mental capacity of a four-year-old" being given a pretty long list of fairly simple tasks.

On the flip side, how many jobs will literally vanish once the mental capacity is of an intelligent college graduate? How many billionaires will want a private army of their own? Oh how the curiosity grows.
 
That is how it looks today, but if they reach something like self conscience they could speed up to super intelligence and then to something way beyond that.
 
"AI has the mental capacity of a four-year-old with no common sense". That sounds a lot like a lot politicians with the exception of their ability to squander our money.
 
"any computer could tell you that Tom Brady is a football player"

Oh really? My computer can't. All that I can do is give my computer some very specific input that it redirects to an application that requests some services via the Internet and returns back to me some URLs, which contained keyword matches on my input, whose quality of match I have to manually determine and then make yet another decision to intentionally select the URL and hope that its resource contains the information to answer my actual question. If that's what you mean, then lets take this one step further. My computer didn't tell me anything at all, but rather accepted my inputs and delegated to another set of computers, that actually had some information about the topic I inquired, which in turn also didn't tell me the answer to my question and instead, redirected me to some other source of information.

So really, hardly any computer could tell me that Tom Brady is a football player.
 
Considering that the last AI I remember being tested ranked about as high as a german sheapard then before that a mouse, I'd say 4year old is good progress. I'm curious how IBM's Watson would score.
In the future will Machine Sentients consider the term "Artificial intelligence" offensive?
 
In the future will Machine Sentients consider the term "Artificial intelligence" offensive?
More likely they will consider "Human Intelligence" offensive. Hell to some degree, I already do. We call ourselves intelligent, but yet we have trouble sharing the same planet. Notice I said share, not hoard as much as possible.
 
Seem to me, the programmer can write the responses with common sense, thus the answers would portray common sense..... Just saying....
 
This article makes me feel bad for losing to the AI on computer games :( the AI shoot better then I do, they micro manage infinitely times better then I do, they do everything better. The only reason why I win is because I have cunning and know how to exploit. (such as hide behind a barrel and creep out until I see the AI opponents head. Or quickly bombard their unguarded resources).
 
Seem to me, the programmer can write the responses with common sense, thus the answers would portray common sense..... Just saying....
Programming is done with logic not common sense. For example, a prgrammer could say, well obviously writing it like A will lead to B. But writing it like A will have C as an unintended side effect.

So the programmer is forced to write the code like D in order for the code to have the desired effect.
Ergo, in programming logic is used instead of common sense, and that logic is reflected in AI capacity of thinking.
 
"The advancement of artificial intelligence (AI) has been truly impressive; however, it doesn?t look like machines are getting anywhere close to a human?s level of understanding."
The advancement has been impressive, but todays AI is still very primitive. In fact, we are still very, very, very, very, very far away from having anything that would even come close to resembling AI thats worth a damn.
And thats just a reflection on us and the truth about our so called great technology. We have made strides, but are still barely out of the jungle.
 
The sad part is the AI will never reach full understanding. It has NO stilmulation!! No sensory input!! That means it will never understand its enviroment. Thats why a 4 yr old has SOME common sense. A human child will have learned hunger, pain, gravity, mine, mommy, daddy, puppy, ect. This computer will never get those experiences.
 
Scientists are aware that in order to develop an artificial intelligence they need to design a system that can gather, organize and store data for future application to various situations. In other words, a learning algorythm. Here is an example of progress that has been made in that venture:

http://www.pcmag.com/article2/0,2817,2390277,00.asp
 
The sad part is the AI will never reach full understanding. It has NO stilmulation!! No sensory input!! That means it will never understand its enviroment. Thats why a 4 yr old has SOME common sense. A human child will have learned hunger, pain, gravity, mine, mommy, daddy, puppy, ect. This computer will never get those experiences.
The last thing I want is for a computer to learn about hunger or pain. Don't give it gravity mines either!
 
These people are playing with fire, artificial intelligence is something the human race isn't ready for yet. Question, is will the human race ever be ready for it, honestly I don't know.

Big thing to remember is that if this is indeed a true AI, even though it has a mental capacity of a 4 year old child, that it won't turn on them.
 
Well... let's be fair. It does take a 4 year old 4 years of learning to be as smart as a 4 year old. Should we expect to be able to make a computer that can just immediately be as smart as an adult? This is a newborn computer that can score like a 4 year-old. A newborn baby literally doesn't know up from down or night from day. A newborn doesn't even know they have hands and feet.

What they should be trying to do is make a computer that can learn, remember, and be curious. They should make a computer that can learn like a human, not act like a human.

Finally some one with some common sence. Spoken like a genious. These scientists and engineers seem to know nothing about social science and socio phycho structures.
 
The sad part is the AI will never reach full understanding. It has NO stilmulation!! No sensory input!! That means it will never understand its enviroment. Thats why a 4 yr old has SOME common sense. A human child will have learned hunger, pain, gravity, mine, mommy, daddy, puppy, ect. This computer will never get those experiences.

Totally agree. I always thought that it was one of the lacking pieces in AI development.
 
Well... let's be fair. It does take a 4 year old 4 years of learning to be as smart as a 4 year old. Should we expect to be able to make a computer that can just immediately be as smart as an adult? This is a newborn computer that can score like a 4 year-old. A newborn baby literally doesn't know up from down or night from day. A newborn doesn't even know they have hands and feet.

What they should be trying to do is make a computer that can learn, remember, and be curious. They should make a computer that can learn like a human, not act like a human.

That be pointless ... here is your new robot slave/warrior please allow 40 years before sufficient knowledge is acquired before usefulness.
 
Back