Posts tagged artificial intelligence

“Artificial Intelligence is useless subject ” – A Classmate

Is it really the popular belief? On more than one occasion I’ve heard students complaining about AI and it’s uselessness as a field of computer science.

Is it really a dead-end of a Computer Science field?…

This XKCD says a lot:

and they both react poorly to Showers.

To assess the current state of Artificial Intelligence we need to first define what is Artificial Intelligence?

As my professor of Artificial Intelligence puts it : “Artificial Intelligence is a field of computer science in which we study, how to make computer do things which at the moment, humans do better”.

As the definition suggests, this covers a broad number of topics from Natural Language Processing to Computer vision.

To properly study A.I. Computer scientists divided things or tasks humans are good at into three task domains :

1) Mundane or simple tasks
- Perception (vision, speech, smell)
- Natural Language (translation)
- Robot Control

2) Formal tasks or tasks with moderate difficulty
- Games (Chess)
- Mathematics (Geometry)
3) Expert Tasks or difficult tasks
- Medical Diagnosis
- Engineering (Design)

Initially scientists thought that since a Child learns simple tasks first then moves on to learn more difficult task, it will be easy to make computers to mundane tasks, like sound recognition or NLP. Turns out these tasks are much more complex than formal tasks like Designing, Predicting a pattern or Medical Diagnosis.

Current Applications of AI includes :

1.) Game Playing : When you play Chess or any game  (Assassins Creed counts too) against a computer, it involves a fair amount of AI but computers play well against humans mainly via brute force computation methods–searching for thousands of moves and selecting the best one according to the criteria programmed. In my opinion that’s not really AI.

2.) Speech Recognition : It was touted to be the next big thing 90s, now we know it is just an EPIC FAIL! The  sorry state of speech recognition is proven by the fact that you can type faster using just your left hand than you would if you use a Speech to Text software. It is still used in mobiles and other such gadgets to give simple voice commands. It has reached a practical level for limited purposes.

3.) Computer Vision : Microsoft’s Kinect anyone?

Microsoft's Kinect - A breakthrough in AI

Kinect is a real breakthrough in computer vision and real world object tracking (Kinect’s AI breakthrough explained ). Kinect is a motion sensing input device which looks like a cheap web-cam but employs a range of camera technologies which interprets 3D scene information from a continuously-projected infrared structured light.

4.) Understanding natural language : Currently if you want to give instructions to computer, you need to write computer programs in a specific specialized programming language like Assembly, C or Java (or any other). Computers are bad in understanding language which we humans speak. Google is working on projects in this area.

there are many more applications of AI. Companies like IBM with their Watson and Google with their gazillion projects are major players in this field.

IBM collaborating with American universities have developed a Computer Chip which is being hailed as a move towards “cognitive computing”.The chip contains 256 neurons, connected to each other in a way that mimics patterns of human brain. Now this is nowhere as efficient and powerful as human brain which has billions of neurons, but this is a step.

So the Big Question :

Is there a promising career in Artificial Intelligence?

Unfortunately, the answer is no. At least not in India. Though there’s a lot of demand for the techniques that someone competent in AI should know. A lot of these techniques are more appropriately classified as machine learning, but there’s a lot of overlap between these fields, so much so that sometimes machine learning is considered a subfield of AI (and I would assert that all those people are wrong, but hey, that’s just me). Machine learning has easy applications in e-commerce and data management, which together covers a huge swathe of the economy.

For those who do want to work in real “ARTIFICIAL INTELLIGENCE” related field but not research in AI, You may still be able to find a job doing “AI” for a game company, but it’s most likely going to be 90% programming other stuff, 5% looking up AI that could be applied to the game and 5% trying to implement it. In order to do real, interesting AI work, you’re more-or-less looking at university work / research work, no escaping it :p .

You first have to decide which “field” or “camps” in AI you’re going to join. At this point, if you’re not knowledgeable enough to make that decision, I would highly recommend taking AI classes that expose you to all different kinds of AI. There’s three main “camps” in AI:

1. Math-based: Simulate the intelligence and patterns of the human brain using complex, finely tuned algorithms, e.g. expert systems.
2. Statistics-based: By collecting enough data (knowledge) about one particular type of event you can predict the occurrence of that event. This is used in stock-market and has other financial / sports applications.
3. Biological-based: Simulate nature and the human brain in the computer, e.g., evolutionary computation.

Each major camp breaks down into even smaller specialized camps. There’s currently a huge disagreement among Researchers about which major approach is the correct approach. Researchers even within the same major camp often disagree with each other. Here’s my personal take on each camp:

1. The math-based approach was the first major approach pioneered by MIT beginning in the 50′s and 60′s. MIT promised all kinds of great things coming from this approach and everyone got really excited about AI in the 60′s. Money flooded in to fund the research projects. Then… nothing came out of it. Everyone was disappointed. AI fell on the back burner. Thanks, MIT! This approach has produced some interesting intelligent machines (e.g. Deep Blue, IBM’s Watson), but they won’t ever be able to produce real general intelligence.
2. The problem with the statistics-based approach is that it requires large amount of data beforehand. This is great for domains where there’s plenty of data from the past available, e.g. finance and sports, but what about domains where there’s no data available? We can’t just send 5,000 robots into space to crash and explode on Mars until we finally have enough data to predict how the robot needs to behave on Mars. Right?
3. The biological-based approach is what makes the most sense to me. If natural processes, e.g. natural evolution, produced human-level intelligence, then simulating those processes in the computer should eventually be able to produce the same level of intelligence. The problem with this approach is that the results are often unpredictable and even their own creators have difficulty understanding how and why said result came to be.

Don’t just take my opinion as fact though. I’m sure many people will disagree with me. The best thing you can do is expose yourself to all the AI concepts you can and make the decision for yourself.

Here are some Artificial Intelligence resources you can follow to better understand AI concepts (I will add more later, so do check back):
1.) Stanford University’s Free AI-Class : It is a free online intro to AI course. It starts in October 2011 and is ran by Stanford professor Peter Norvig who wrote one of the best text books on AI and algorithms.

2.) Free eBooks on Artificial Intelligence.

Good Luck guys, and remember contrary to popular saying among students – Artificial Intelligence won’t short circuit your natural Intelligence!