What is needed to program artificial general intelligence



Ben Goertzel: There are aspects, yes. AGI has aspects of computer science, mathematics, engineering, philosophy of mind, linguistics, neuroscience. It is quite interdisciplinary and the education system is not really that way. It is more so in the US than anywhere else in the world. That is a strength that the US has. As an undergraduate you can at least follow courses at each department. And in many countries that is not true.

But even in the US, the educational system is not as interdisciplinary as it should be to struggle with a problem like AGI or say quantum computing or nanotechnology or many other advanced things.

So what that means is that if someone really wants to work in one of these cutting-edge topics that has the greatest opportunity to transform the world, if they want to work on these things in the core capacity, they have to take their time to do some other study fields that they have not learned at school. And that also takes time. You can not do that by reading a blog post. I mean, you have to take a book of neuroscience and go through it step by step. And not everyone has the patience for that.

But again, some people do that, and I would say Coursera, Udacity and MIT, the many universities that have put their course material online, are a huge, huge asset in this process because they help people learn the information from everyone. disciplines they need to attack something like AGI. We found these online courses incredibly useful in what we did in Ethiopia.

So in 2013, together with two others, I established Ethiopia's first AI and development company for robotics. So we do some original R & D, a few projects aimed at helping the African situation. Then outsource a lot of software and robotics. The company is called iCog Labs in Addis Ababa. And we have an internship program that we use for recruitment.

So we take dozens of students every year with an undergrad study and what we do is give them practical lessons in OpenCog and various other AI tools. We also have each of them as seven Coursera courses. And they go through them very quickly and they teach them neuroscience, computer linguistics, bioinformatics, machine learning, a lot of subjects that are not offered at the university.

And this works much better than having a lot of textbooks read because it gives them a process and a community to go into. It not only teaches them information, but also obscures people who have no perseverance to manipulate things from different disciplines and to really stretch their brains in a deeper, interdisciplinary way.

So yes, I would say, as with everything else, pluses and minuses are all confused, right? I mean, the modern way of doing things in a certain way eliminates the attention span of people because nobody has to think for themselves. They immediately search for the answer on the internet or download something instead of solving a problem themselves.

On the other hand, there is so much high-quality educational material available, along with supportive communities for people who want to deepen and deeper and have a more fundamental understanding.

But what we are doing in OpenCog is that we have developed a system where each of the cognitive processes can help the other if it gets stuck in some combinatorial explosion problem. So when a deep neural network tries to perceive things, it gets confused because it is dark or it looks at something it has never seen before, well maybe the reasoning machine can come in and make some inference to break that confusion.

If logic reasoning gets confused and does not know which step to take next, because there are just so many possibilities and there is not much information about it, well, maybe you're fishing in your sensory-motor memory and using deep-learning something to visualize you saw earlier, and that gives you an idea of ​​how to go through the many possibilities that the logic engine sees.

Now you can mathematically model this kind of cognitive synergy with the help of a branch of mathematics called category theory, something that I've been working on lately.

But what's really interesting is to build a system that manifests this and generates general intelligence, and that's what we do in the OpenCog project.


Source link

What is needed to program artificial general intelligence



Ben Goertzel: There are aspects, yes. AGI has aspects of computer science, mathematics, engineering, philosophy of mind, linguistics, neuroscience. It is quite interdisciplinary and the education system is not really that way. It is more so in the US than anywhere else in the world. That is a strength that the US has. As an undergraduate you can at least follow courses at each department. And in many countries that is not true.

But even in the US, the educational system is not as interdisciplinary as it should be to struggle with a problem like AGI or say quantum computing or nanotechnology or many other advanced things.

So what that means is that if someone really wants to work in one of these cutting-edge topics that has the greatest opportunity to transform the world, if they want to work on these things in the core capacity, they have to take their time to do some other study fields that they have not learned at school. And that also takes time. You can not do that by reading a blog post. I mean, you have to take a book of neuroscience and go through it step by step. And not everyone has the patience for that.

But again, some people do that, and I would say Coursera, Udacity and MIT, the many universities that have put their course material online, are a huge, huge asset in this process because they help people learn the information from everyone. disciplines they need to attack something like AGI. We found these online courses incredibly useful in what we did in Ethiopia.

So in 2013, together with two others, I established Ethiopia's first AI and development company for robotics. So we do some original R & D, a few projects aimed at helping the African situation. Then outsource a lot of software and robotics. The company is called iCog Labs in Addis Ababa. And we have an internship program that we use for recruitment.

So we take dozens of students every year with an undergrad study and what we do is give them practical lessons in OpenCog and various other AI tools. We also have each of them as seven Coursera courses. And they go through them very quickly and they teach them neuroscience, computer linguistics, bioinformatics, machine learning, a lot of subjects that are not offered at the university.

And this works much better than having a lot of textbooks read because it gives them a process and a community to go into. It not only teaches them information, but also obscures people who have no perseverance to manipulate things from different disciplines and to really stretch their brains in a deeper, interdisciplinary way.

So yes, I would say, as with everything else, pluses and minuses are all confused, right? I mean, the modern way of doing things in a certain way eliminates the attention span of people because nobody has to think for themselves. They immediately search for the answer on the internet or download something instead of solving a problem themselves.

On the other hand, there is so much high-quality educational material available, along with supportive communities for people who want to deepen and deeper and have a more fundamental understanding.

But what we are doing in OpenCog is that we have developed a system where each of the cognitive processes can help the other if it gets stuck in some combinatorial explosion problem. So when a deep neural network tries to perceive things, it gets confused because it is dark or it looks at something it has never seen before, well maybe the reasoning machine can come in and make some inference to break that confusion.

If logic reasoning gets confused and does not know which step to take next, because there are just so many possibilities and there is not much information about it, well, maybe you're fishing in your sensory-motor memory and using deep-learning something to visualize you saw earlier, and that gives you an idea of ​​how to go through the many possibilities that the logic engine sees.

Now you can mathematically model this kind of cognitive synergy with the help of a branch of mathematics called category theory, something that I've been working on lately.

But what's really interesting is to build a system that manifests this and generates general intelligence, and that's what we do in the OpenCog project.


Source link

Leave a Reply