Skip to main content

AI is just Data Science

Most of modern AI is just data science. In the early days when we did what we now refer to as  GOFAI (Good Old Fashioned AI), we tried to give computers the ability to match humans in solving problems that were strictly in the domain of human cognition.

Although so many powerful results were obtained and it convinced researchers that we were on the right path towards artificial general intelligence, it was still far from what a human child could achieve without any education.

The results of this age did more to help us build sophisticated computer systems than they helped to solve simple problems like image recognition, things we now take for granted these days.

The early AI researchers were incredibly smart people, they used the limited computer resources of the time to achieve many great feats but they were over-enthusiastic about the abilities of their machines and eventually when they were not able to show better results to the public and investors, we had two AI winters to contend with.

It was no fault of these researchers that they didn't come upon the kinds of simple algorithms we now use to solve basic issues in AI like vision and speech recognition. They had the brains but little computational muscle and of course no data.

Whenever I contemplate the history of AI I am shocked and humbled. Shocked at the fact that backpropagation was first published in a psychology journal and not a computer science journal and for the most part was largely ignored.

Another part that shocks me is that even though the intellectual foundations of deep learning were already laid in the 1970s the computational power and data was not available. But if you could time travel to the early days of AI enthusiasm in the 60s and ask a good researcher to give you an estimate of when AGI or strong AI would arrive they would say it would arrive in their own decade.

We grossly overestimate our current abilities most of the time. The AI researchers of the 60s were confident in the power of their machines to soon become better than humans in just a decade. Size too was not an issue because most of these computers filled entire rooms. If you travelled back in time and entered their scene with no memory of the future you too would be trapped in their illusion.

They didn't know that VLSI (Very Large Scale Integration) circuits were coming to change the entire computer architecture and make room for faster machines, many engineers mocked at the idea of VLSI when people like Mead Carver were proposing it.

There was nothing those researchers could do to bring about something like deep learning within the limits of their ambient technologies and it makes me think, do we currently have all we need to start even thinking of building AGI?

Without the internet and a large amount of publicly available data, it would be difficult to think of the ImageNet competition or all the large scale computer vision competitions we do. The AI researcher of the 60s and even 70s thought they had all it takes to get some AI system working and it was just a matter of some clever algorithm or the other and soon we would have AI systems taking over humans in every thinkable task.  Well, that didn't happen, and while we might feel that we are better than those people of the past we are not. We are still trapped in the illusion of our current technologies and we don't know for sure what combinations of new technologies will create the ambience that will enable us to achieve our high AI dreams.

This brings me back to the topic of this post: AI is just data science. Well more accurately modern AI is just data science. When I got into AI I focused mostly on symbolic systems. The logic-based systems didn't make much sense to me and I thought that since the human being is a symbol processor for the most part then building a synthetic symbol processor would be AI and all that was needed was just some super smart algorithm for constructing the perfect symbol processor and then boom! we have AI.

Well what I didn't consider was symbols were just data and when the crude data digesting systems called neural networks started becoming more prevalent, I initially refused to join the band wagon because I thought such wild black box systems were just trickery.

But we cannot deny the performance of neural networks especially the deep type because we can see them being adapted to many kinds of tasks in the modern world. This switched the entire focus of AI from clever algorithms to data. It was shown in several empirical tests that more data was always better than a clever algorithm and thus we have entered the data science-based AI age.

Many people getting into AI do not know that what they are doing mostly is just data science and I think this is because we have been overselling the word Artificial Intelligence because it sounds funky and garners interest. Data science is a cool profession but the title doesn't carry many sparkles and if you say that you are a data scientist nobody pops an eye. But the term AI just carries so much imagination of the terminator, Skynet etc. and thus when someone says, I am doing AI, it sounds more interesting but in reality when you sit on your computer and start training all those deep models you are actually doing data science.


Post a Comment

Popular posts from this blog

Next Steps Towards Strong Artificial Intelligence

What is Intelligence? Pathways to Synthetic Intelligence If you follow current AI Research then it will be apparent to you that AI research, the deep learning type has stalled! This does not mean that new areas of application for existing techniques are not appearing but that the fundamentals have been solved and things have become pretty standardized.

How to become an AI researcher

Artificial Intelligence is all the rage these days. Everyone is getting on the bandwagon but there seems to be a shortage of AI researchers everywhere these days. Although many people are talking about doing AI not many people are actually doing AI research.