Team-BHP - Artificial Intelligence: How far is it?

Thanks for the pointer to interesting book.

Quote:

Originally Posted by ZedMae (Post 4398645)

One is - All thinking (including conscious awareness) of the brain is mere computation. Very similar to how we calculate 2+2. Since it is mere computation, there will be a time when machines with appropriate computing power can entirely simulate our human brain (including - I repeat - being self aware)....

There have been some amazing advances that point in this direction :

Some of the ML was inspired by Human brain
Now to understand some brain defects, parts of ML models are used as examples

https://ai.intel.com/deep-learning-study-brain-improve-deep-learning/

I was attending a Workshop at Microsoft to learn about their cognitive service offerings and I got to have a look at their research project that is capable of reading people#s MRT data (during the MRT session) where they show the person some images. The algorithm would then try and predict from the MRT data what the person imagined. I was stunned at this as the results were almost accurate! This is so scary!

Quote:

Originally Posted by ZedMae (Post 4398645)

Hi locusjag, Give a shot at the book - Shadows of the Mind by Roger Penrose. The crux of the book is about 4 different viewpoints about the human brain and Artificial Intelligence.

Thanks! I'll surely look for it; I couldn't find it on Kindle, so will look for a paperback version perhaps.

As a return courtesy, I'll suggest "Surviving AI..." by Calum Chace. It coverss different scientists' approaches to AI, classifications (or types of AI), where we are and what could be. It was a lovely read.

Quote:

The other viewpoint (2nd of the 4 in the book) is - certain physical activities of the brain trigger or facilitate self awareness and this cannot be simulated computationally irrespective of the quantum of computing power at one's disposal. The author stands by this viewpoint.

Interesting view point.

Sounds similar (but not the same) as Moravec's paradox, which is also deemed as insurmountable by AI. This paradox is that the toughest tasks lend themselves to robotics/automation/AI, but the simplest tasks involving even the most basic motor skills are not counquerable by AI.

Gentlemen, I need a piece of advise from the experts. This particular field fascinates me and the things which fascinate me; I just learn them for the sake of knowing it. Now I want experts to guide me what is the way to tread on?

I mean, I am someone who is proficient with only MS Excel and the basics mathematics that is used in the field, but I don't know where exactly to go. What to study and how to implement? Any beginning tips will be a blessing for me. That said; I am currently a PG student with prior experience in business.

I know I have asked a noob question, but any person with as much know how as much I have will ask this type of stupid questions only.

Quote:

Originally Posted by VKumar (Post 4398963)

If you don't have any prior experience in programming, will suggest to at least take a course on basic programming fundamentals before you dig deeper into AI. There are plenty of them available online. Just google.

Post that, I suggest you to start looking reading books, blogs, online courses on machine learning. Udemy has a great online course on machine learning which can act as starter kit. You can begin with programming languages like Python & R.

Next step will be to think what problems can be solved in your line of profession using machine learning. This will help to understand the practical impact and maybe you can come up with a killer idea as well :)

Quote:

Originally Posted by VKumar (Post 4398963)

I'm sure others will go on to suggest courses on Pluralsight (with monthly subscription costs as low as Rs.2000), Coursera (costs per course), Udemy etc. There are also free course contents made open to the world by certain Ivy league colleges. I keep hearing that there are Youtube videos which are more than enough.

The one true solid nugget I can offer is, please visit Kaggle.com when you do start learning this stuff. Kaggle is to budding ML engineers/data scientists, what Team-Bhp is to auto enthusiasts; and moreover, there are free and awesome real-life data-sets available on that site to play with, when you're learning and experimenting with Python/R/SQL etc.

All the best!

Quote:

Originally Posted by VKumar (Post 4398963)

Statistics, Linear Algebra, Calculus, and programming are the prerequisites for machine learning, if you want to be technical about it.

Quote:

Originally Posted by VKumar (Post 4398963)

I mean, I am someone who is proficient with only MS Excel and the basics mathematics that is used in the field, but I don't know where exactly to go.

I suggest two pronged approach : (You can do both or only the second to just get a feel of ML/AI.)

1. Structured learning : Three courses that gradually bring you up to speed. All courses mandate code submissions every week (so learning is "active"). These courses do not assume that you are an expert in programming, you will learn programming gradually.

Andrew Ng's class. The de-facto place to start. He is a great teacher (in my opinion). This class prepares you with the basics and you know where to explore next. link
Geoffrey Hinton's class. He has been working on this for last 30 years. He pretty much invented current ML (he is author of back prop algo) link
Andrew Ng's deep learning classes. link

2. Practical learning by "tinkering" : Do you have any problem in mind that you want to solve ? If you have a problem, then one way is to just dive in and start solving it by various ways.

You can use various tools that enable ML without coding (including Excel tools). E.g.: https://www.youtube.com/watch?v=jS1cmzeiaDk

You can pick some beginner problems like :

Find if there is some pattern of car sales and months [I.e. Do hatchbacks sell more in summer/winter/rains ?]
Predict more probable time a user ill post on tam-bhp (Say user "xyz" will usually post in mornings and only if there is a new thread in 4x4 forum. Can you use some algo to detect this pattern)

Then some mid-level problems like :

Plot sentiment of car brands on team-bhp over time (Skoda might have a positive trend till the thread on Harish's car came up; does this assumption stand up to data produced by algo?).

Quote:

Originally Posted by VKumar (Post 4398963)

I have been reading through the so called ISL book. You can read it here for free: http://www-bcf.usc.edu/~gareth/ISL/

Hello fellow AI enthusiasts, if you've got a NetFlix subscription, I highly recommend this awesome documentary :thumbs up

If you come away from this documentary troubled & with a sense of disquietness, know that you will not be alone.

I work on the theoretical or "mathy" side of AI tech, still, some of the stuff I've seen on the applications side make me go sleepless some nights ! :Shockked:

I'm firmly behind Musk in his assessment of AI & the ultimate fate of humanity, nevertheless, the documentary is a definite must-watch.

Came across an interesting article on AI. Looks like AI can be biased.

Quote:

Amazon cans AI for hiring bias

Artificial intelligence promises to make us better by harnessing computing power to make everyday decisions more quickly and less subjectively than people can.

But as AI technology spreads, it's becoming apparent that it can't escape the biases of its makers.

Amazon Inc. is the just latest company to learn that lesson.

According to a new report today from Reuters, the Seattle tech giant was forced to scrap an AI recruiting engine after the company came to realize that the algorithm was excluding women.

The in-house technology was supposed to crawl the Web and identify candidates that the fast-expanding firm might hire for a variety of software development and technical jobs. But within months of its 2014 deployment, Amazon executives realized that it wasn't functioning as intended.

Source

This is not the only case

Quote:

Facial Recognition Is Accurate, if You’re a White Guy

Facial recognition technology is improving by leaps and bounds. Some commercial software can now tell the gender of a person in a photograph.

When the person in the photo is a white man, the software is right 99 percent of the time.

But the darker the skin, the more errors arise — up to nearly 35 percent for images of darker skinned women, according to a new study that breaks fresh ground by measuring how the technology works on people of different races and gender.

These disparate results, calculated by Joy Buolamwini, a researcher at the M.I.T. Media Lab, show how some of the biases in the real world can seep into artificial intelligence, the computer systems that inform facial recognition.

Source

More such examples on the original article. If only people take Elon seriously on AI.

Quote:

Originally Posted by Technocrat (Post 4476308)

Came across an interesting article on AI.
Looks like AI can be biased.

When it comes to AI, often times, what we construe as bias on their part stems from very logically derived conclusions - sometimes, to a fault.

For example, a team I know from work, develop what are called "generative" AI - basically AI used for "creative" tasks.

You can get a basic idea of this use of AI from the following TED talk.

https://www.youtube.com/watch?v=aR5N2Jl8k14

In any case, not unlike in the presentation, this team was using AI to develop monocoque chassis for passenger vehicles.

They soon started seeing a trend, that the structures being generated were structurally "firm" on only side !

This confused the team a bit, but they soon realised that this asymmetry was manifested in the designs only after constraints from crash testing data had been applied; this is quite common, the search space is "pruned" by specifying constraints, an unpruned search space can get quite unmanageable - even for AI.

I think you would've guessed by now what was happening...

It turned out, the AI was "optimising" the monocoques by strengthening only the driver's side ( crash testing is almost always carried out on the driver's side ).

The asymmetry in design wasn't some out-of-the-box thinking by the AI, it was just being logical lol:

I'm positive the AI folks at Amazon know pretty well why their AI went the way it did, it definitely wasn't misogynistic bias.

On a more somber note, most human designers "optimise" their designs for crash tests too, that's why the IIHS started carrying out passenger-side crash tests.

https://www.youtube.com/watch?v=72caLypmKCA
~