AI/Machine Learning, Bharat and Bhartiya IT Industry

ArjunPandit · Post by **ArjunPandit** » 27 Mar 2019 06:14

fanne wrote:Arjun, thanks for the input (particularly what course to take, otherwise it was an ocean and even with that link, one would not have made any progress. Where do you start? You cannot do all - nearly 1000-5000 hours of learning, mostly repeating and perhaps confusing). Follow up question - I see most classes have videos, it looks free. Is it all complete? Like if I needed an engineering degree in computer science, there are perhaps 10,000 colleges offering it, is the Oxford one complete in itself (to the extent it can be, the field is evolving every day) and fully available in that link?

I wont be able to comment on the video lectures as I am more of a reading person. You have to make first step. I would suggest oxford course. It's pretty light. If you get stuck, I can help you. Should we have a separate thread for that or change the thread title?

NRao · Post by **NRao** » 27 Mar 2019 07:27

fanne,

As Arjun mentioned, "AI" is too broad. And, for most of us who figured out a particular language (Java?) and made a good living, "AI" does not fit that model.

AI includes Machine Learning (ML), Natural Language Understanding (NLU), Network Analysis (NL), Mining (Text), etc.

First, decide which area interests you the most (there are a lot more than I have listed). Some examples here

Next, pick a language - Python is the most commonly used, R is not too far behind.

Based on these two, you will need to specialize, within those languages, in libraries that pertain to a specific field within "AI". Examples for Python here

Other areas that would be of immense help: Statistics (ML uses this area a lot. Here is a sample). Data Integration (DI), moving data around. Data sets being real, real large, you will need to understand how to manage data (Python).

One thing for sure, just learning a language, like Python, is not going to cut it.

There is a LOT more to this topic but hope this, along with Arjun's URLs, helps a wee bit.

UlanBatori · Post by **UlanBatori** » 27 Mar 2019 07:59

Hay, I AM serious.. Do you know any AI preachers in desh who might take interest in the items I mentioned?

UlanBatori · Post by **UlanBatori** » 27 Mar 2019 08:04

ArjunPandit wrote:On a serious note, does the Kublai khan have a code of Hammurabi for a credible n write up

What is a credible n writeup, hain?

Bliss to send Brivate Message Onlee for more Oben Sesame keys. Ramana knows. Also, for starters on serious things, I assume u speak LaTEX?

siddhu · Post by **siddhu** » 27 Mar 2019 08:07

titash wrote:I'm actually not opposed to a BR startup that develops AI technologies for use by the Indian Armed forces - perhaps something that will help us win in Kashmir, or robotics/UAVs etc. Anyone interested? We should do meetups and brainstorming.

Please count me in.

NRao · Post by **NRao** » 27 Mar 2019 08:17

UB,

Not sure who you are addressing. I will take a stab at your Q.

Assuming the following is the post that you are referencing, here goes:

UlanBatori wrote:Calling all Faithphool:

I have a serious interest in this dhaga. Here is my take:

1. Legal System
AI can be the breakthrough technology that breaks through the convoluted mess of India's Legal System. Think of a case filed: The parties would have to use the AI app (on a serious computer) to describe their problem/case precisely, but also with the more subjective disputes described.
AI digs out the relevant Precedents, laws etc in a flash. This will take time, as the Precedents are gradually entered into the System, but at the start at least a title search followed by text search is possible soon).

Bottom line: the work of Law Clerks taking months is done by AI in seconds. Then the clerks can step in and clean up what AI cannot.
Then the Judge is presented with the AI/clerk based summaries and references to data. The Judgement is most of the time written by AI, for the Judge to review and approve, with a probability-based Certainty Rating (CR) assigned for the Judge to decide how much time to spend.

Where the CR is say, below 85%, the Judge has to spend more time reviewing, but the data collection is done very swiftly.

HP bought a (Britsh!!) company by the name of Autonomy that is just too good at Such things. Autonomy was founded around 1995. The algoes, BTH, is based on Bayesian theorem. It is an excellent tool and very widely used in the legal community. It understands NLU, in MULTILE languages. You can train it to learn in Hindi too - if it has not been yet done

So, in short, that is already a done deal.

2. Medical delivery/ Telemedicine
Here AI has a number of roles to play, from communications to diagnosis.

There are innumerable other specific applications of course.

What I need is to find someone(s) who will
(a) help develop a credible writeup
and (b) come to a conference in July (11-14) in Chennai and give a presentation and chai-biscoot with experts in other fields, That's right: not other experts in same field, but experts in other fields.

VERY SERIOUS request, Meanwhile I will be happy to steal stuff from here to use in trying to appear knawlidgeable.

Unfortunately "diagnosis" is just too wide a term. TODAY AI (seems to) work very well in IDing cancer cells. But it fails when it is consulted for a common cold or fever. A lot of work needs to be done in this area.

AI is being used, extensively, in Pharma - to come up with new concoctions.

BTW, "AI" started in the medical field. BUT that AI was very different than what we have today.

Hope that help.

____________________________________________________________

As a side note: Google for jobs in AI by nations. India has recently slid a wee bit but still employs one of the largest numbers of people within "AI".

Extremely unfortunately, MOST (I estimate 95%+) work for foreign companies.

I estimate there are at least about 500K employed in AI in India.

TandavBrahmand · Post by **TandavBrahmand** » 27 Mar 2019 09:43

First, let us get the terminology right. AI is too broad. ML is probably we all are talking about here. ML pertains to systems which learn (supervised or otherwise) over a period of time and get better than humans. AI, on the other hand, is mimicking human intelligence and has a broader or esoteric goal.

Few comments based on past few posts:

1. Autonomy is too old school. A high-schooler can write a NB classifier these days over a weekend.

2. Text summarization (Legal or otherwise) is very much within the grasp of the current NLP. Text understanding( and Translation) is well understood but summary generation with content is still WIP for serious applications. However, there are serious, production-ready applications that are being worked upon by ML folks in the industry. The scale of the application you are talking about is huge. # of documents and text to be processed would be huge. Also, first they would have to be digitized (e.g. FIRs are still handwritten) and then read as text. As of today, no ML framework (TF, Azure, PyTorch) can scale to that level. However, they will be in 2-3 years.

3. Medical diagnosis using x-rays, CT, MRI images via trained Neural Nets (CNN) are being attempted as well. Mostly in research labs. A phone picture based glaucoma or other retina anomaly detection etc have been attempted. Not commercial AFAIK.

Given the sensitivity and compliance issues (such as HIPPA) real applications are hard to develop without training data at scale. Hospital Networks, Medical mfg, Pharma with big money are indeed working on these. Again, we should see scalable applications in 2-3 years.

I am in Unkil's country so not sure how local you want the help to be.

banrjeer · Post by **banrjeer** » 27 Mar 2019 10:10

Nvidia came out with this yesterday.

https://nvidianews.nvidia.com/news/nvid ... 0005468164

hanumadu · Post by **hanumadu** » 27 Mar 2019 12:01

fanne wrote:Arjun, thanks for the input (particularly what course to take, otherwise it was an ocean and even with that link, one would not have made any progress. Where do you start? You cannot do all - nearly 1000-5000 hours of learning, mostly repeating and perhaps confusing). Follow up question - I see most classes have videos, it looks free. Is it all complete? Like if I needed an engineering degree in computer science, there are perhaps 10,000 colleges offering it, is the Oxford one complete in itself (to the extent it can be, the field is evolving every day) and fully available in that link?

If you are completely new to ML, you can probably start the Machine Learning course by Andrew Ng on coursera. It gives a quick and fast moving introduction to a range of applications of ML. You can explore further based on what holds your interest.

ArjunPandit · Post by **ArjunPandit** » 27 Mar 2019 13:10

UlanBatori wrote:
ArjunPandit wrote:On a serious note, does the Kublai khan have a code of Hammurabi for a credible n write up
What is a credible n writeup, hain? Bliss to send Brivate Message Onlee for more Oben Sesame keys. Ramana knows. Also, for starters on serious things, I assume u speak LaTEX?

Interesting... Seems like caves have tunnels... As for latex .. Little experience in cyber space .. Real world is different...
As I said earlier, that's how the ISI agent ended up in raw...sir Zaid misses him..I miss sir zaid..
is it necessary to come? can a humanoid with a bhediya mask with deep throat not give a presentation on skype?

ArjunPandit · Post by **ArjunPandit** » 27 Mar 2019 13:12

hanumadu wrote:
fanne wrote:Arjun, thanks for the input (particularly what course to take, otherwise it was an ocean and even with that link, one would not have made any progress. Where do you start? You cannot do all - nearly 1000-5000 hours of learning, mostly repeating and perhaps confusing). Follow up question - I see most classes have videos, it looks free. Is it all complete? Like if I needed an engineering degree in computer science, there are perhaps 10,000 colleges offering it, is the Oxford one complete in itself (to the extent it can be, the field is evolving every day) and fully available in that link?
If you are completely new to ML, you can probably start the Machine Learning course by Andrew Ng on coursera. It gives a quick and fast moving introduction to a range of applications of ML. You can explore further based on what holds your interest.

I'm completely against anyone spending a dime on these things

ArjunPandit · Post by **ArjunPandit** » 27 Mar 2019 15:31

For those looking for a structure in AI/ML. Very simply said, my definition of AI: that can mimick human intelligence in decision making.
1. I rule out RPA from it.
2. Primary split for ML is traditional v/s non traditional
a. Traditional Supervised: Regression, parametric distribution fittings, traditional unsupervised: Clustering
b. non traditional: Supervised
i. Decision tree, ensemble
ii. Neural Net: CNN/RNN
c. Non traditional: Unsupervised
i. KNN, look alike etc
3. I overlay the AI with
a. NLP-NLU & NLG
b. Image Processing

The applications we see below are a combination of the above approaches

ArjunPandit · Post by **ArjunPandit** » 27 Mar 2019 15:42

banrjeer wrote:Nvidia came out with this yesterday.

https://nvidianews.nvidia.com/news/nvid ... 0005468164

I am very curious abt the hardware part of it. Can you please guide me about how h/w supports AI. Can someone point more details? even huawei claimed AI camera phones. Not sure what part they did in h/w. For all we know it might be s/w doing the job
..got this on wiki
https://en.wikipedia.org/wiki/AI_accelerator

ArjunPandit · Post by **ArjunPandit** » 27 Mar 2019 15:45

I have a better slide than the above one..will post it tonight...

UlanBatori · Post by **UlanBatori** » 27 Mar 2019 17:08

NRao wrote:
So, in short, that is already a done deal.

2. Medical delivery/ Telemedicine
Unfortunately "diagnosis" is just too wide a term. TODAY AI (seems to) work very well in IDing cancer cells. But it fails when it is consulted for a common cold or fever. A lot of work needs to be done in this area.
AI is being used, extensively, in Pharma - to come up with new concoctions.
BTW, "AI" started in the medical field. BUT that AI was very different than what we have today.
Hope that help.

NRaoji, is it a "done deal" as in today it is used in the legal system? I suspect that I know the answer to THAT one (Ayodhya verdict out yet?) I am not talking about writing basic research proposals, but IMPLEMENTATION. As in having See - Near IAS ppl involved. HUGE task, IMHO.

If text recognition and interpretation is indeed a "done deal" (I agree that it can't be that hard to automate the intelligence of the average ambu-chaser) then it is great news: it is time to "Bridge Research Results to Implementation" which is the point.

On Med delivery, sounds like again your focus is on the basic research aspect. Ppl are welcome to do that, but what I am trying to see is how to bring mass delivery of medical care to every village. So the role of AI is to try and do what a doctor can do in 5 seconds of "diagnosis". Presumably at a station where there is a qualified medical assistant who can operate a stethoscope, BP gizmo, thermometer, oxygen indicator, and say

SAY "AAAAA!"

and shine a Pen Torch into someone's (nose).

Like I said, serious request.

APji, Sky-Pee is fine if the Brejenter is a geezer who can't come. But much more important is to contribute knowledge to the distillation of knowledge B4 the Conpherenj.

ArjunPandit · Post by **ArjunPandit** » 27 Mar 2019 17:23

^^My evil manager said many eons back, ideas are cheap as your sperms, getting a kid, and a good one, out of them is a the main thing

NRao · Post by **NRao** » 27 Mar 2019 21:26

UB,

1) Autonomy:

as in today it is used in the legal system?

* As in 180 moons ago. I just checked, they (HP Autonomy now) still have the "Litigation Readiness Archiving" and "Legal Content Management". It is an expensive product, so only large firms, that provide litigation across many areas, can afford it. Someone dealing with just injury cases, local to a city/county can never afford this product. BUT, back 180 moons ago, "Legal" was the core to sales at Autonomy.
* The product is based on patterns and therefore everything had to be digitized (not sure if this is currently the case, but I expect it to be). So, Pt. Bhimsen Joshi's old cassette tapes will not do. They ALL have to be digitized
* Once digitized, everything is fair game - text, audio, visual, streaming, whatever. No problems searching for info in any format
* Since it is digital based, languages do not matter
* Biggest users used to be ESPN (use the audio + visual search features, very extensively) and BBC,com (the core to their multi-language web site is Autonomy). Again, that was 180 moons ago. not sure now

Best I can suggest is to have HP India demo the product. Make dead sure to get a great SE.

Back then Microsoft and Oracle shut down their efforts in this area. Each had a cluster of 50-60 PhDs in "search". I think it is a better product than Google. BTW, I tried to get in touch with old hands, to see if they could geolocate the downed F-16 (based on pics while they were loading the truck). All had moved on. But, Autonomy should be able to do that.

2. Medical:

Try this. Taja, taja. It is a huge challenge - as far as I hear (not my area of interest).

Hope that helps.

ArjunPandit · Post by **ArjunPandit** » 27 Mar 2019 21:38

NRao wrote:
Try this. Taja, taja. It is a huge challenge - as far as I hear (not my area of interest).

Hope that helps.

This is where i have a problem, our data, they refine algos and sell it to these countries as CSR, they might charge us for it

NRao · Post by **NRao** » 27 Mar 2019 21:40

banrjeer wrote:Nvidia came out with this yesterday.

https://nvidianews.nvidia.com/news/nvid ... 0005468164

Nice find, thx.

Here is more info:

https://www.sparkfun.com/products/15297

GPU 128-core Maxwell™ GPU
CPU Quad-core ARM A57
Memory 4 GB 64-bit LPDDR4 | 25.6 GB/s
10/100/1000BASE-T Ethernet
Storage microSD (not included)
Video Encode 4Kp30| 4x 1080p30| 9x 720p30 (H.264/H.265)
Video Decoder4Kp60| 2x 4Kp30| 8x 1080p30| 18x 720p30| (H.264/H.265)
Micro-USB 5V 2A
DC power adapter 5V 4A
USB 3.0 Type A
USB 2.0 Micro-B
HDMI/DisplayPort
M.2 Key E
Gigabit Ethernet
GPIOs, I2C, I2S, SPI, UART
MIPI-CSI camera connector
Fan connector
PoE connector
Dimensions: 100mm x 80mm x 29mm

Data sheet

Getting started with Jetson Nano

NOTE: The list of nations this cannot be shipped to is larger than those countries it can be shipped to!!!!

RohitH · Post by **RohitH** » 27 Mar 2019 21:49

ArjunPandit wrote:
NRao wrote:
Try this. Taja, taja. It is a huge challenge - as far as I hear (not my area of interest).

Hope that helps.
This is where i have a problem, our data, they refine algos and sell it to these countries as CSR, they might charge us for it

Here too it's Google in the fray..

NRao · Post by **NRao** » 27 Mar 2019 21:51

ArjunPandit wrote:
NRao wrote:
Try this. Taja, taja. It is a huge challenge - as far as I hear (not my area of interest).

Hope that helps.
This is where i have a problem, our data, they refine algos and sell it to these countries as CSR, they might charge us for it

Amy Webb has tried to address this real concern of yours: Who owns the "data". Yesterday she appeared on WBEZ Worldview (they may have it as a podcast). And, she mentioned two things:

1) Intrusiveness, where companies like Amazon are trying to invade your private space to collect info passively. So, Amazon, as we type, is peddling a real cheap microwave oven. She feels that this is a way for Amazon to gather data passively (in addition to the Echo). Once collected, she expects, Amazon to control you. As in the microwave will refuse to pop your popcorn because Amazon has determined you have already consumed too many calories for the day/week/month. And, if the weather is great, Amazon will disconnect your car, so you are forced to walk to work a mile away. This brings up the issue - what you brought up - who owns the data.

2) She mentioned Apple's watch may have more cardio info on you than your own doctor

She has suggested resolving these problems, which of course, has made her the target of the 6 companies in the US. She hates Facebook (yeah) the most.

Point being India has the time to and should regulate "data". There is some movement in the area - keep data local, but that I do nto think is sufficient. And, traditionally India does not move quick enough.

ArjunPandit · Post by **ArjunPandit** » 27 Mar 2019 21:56

^^NRao, should we at BRF make this proposal to our local legislators?

Haresh · Post by **Haresh** » 28 Mar 2019 00:59

Just curious
How many companies in India manufacture IT Comms cabinets?
Are there any Indian manufacturers?
Or do they all get imported from China/West?

UlanBatori · Post by **UlanBatori** » 28 Mar 2019 01:12

ArjunPandit wrote:^^NRao, should we at BRF make this proposal to our local legislators?

Data ownership is a HUGE issue. To be discussed in conpherenj.

ArjunPandit · Post by **ArjunPandit** » 28 Mar 2019 01:28

^^I am sure you know the mirch by RBI data localization. In my opinion data is like natural wealth. I wont be surprised if Winter solider pods capability exists in limited scope

UlanBatori · Post by **UlanBatori** » 28 Mar 2019 01:29

[quote="TandavBrahmand"]AI is too broad. ML is probably we all are talking about here.

Much obliged. This is exactly what I need: discussion on what is possible today, what is on the horizon, to be placed in front of the Implementors - and the actual Villager residents.

You are right: For the vast majority of human tasks in the workplace of today, no special intelligence is required. Just careful attention to a procedure, with some sensor/detection of anomalies.
In corporations, dealing with anomalies requires "escalation" to involve Higher Authorities.

So it absolutely true that 99.99% of the need is basic "ML" or even simpler forms of routine automation.
Back in a desi university recently I saw this gizmo (Siemens product and co-development) to, say, train plumbers on the force needed to bend pipes. There was a mechanism with the kinematics needed for various bend angles, and a computer and sensors. Pretty simple but very useful. You learned to bend pipes very swiftly by learning the correct rates of acceleration/deceleration to hit a target - without messing up any pipes. Just an example.

1. Autonomy is too old school. A high-schooler can write a NB classifier these days over a weekend.
That may be, and it's good to hear, but with a HUGE market for autonomous UAVs and self-driving gizmos, this may be a huge demand area too.

2. Text summarization (Legal or otherwise) is very much within the grasp of the current NLP. Text understanding( and Translation) is well understood but summary generation with content is still WIP for serious applications. However, there are serious, production-ready applications that are being worked upon by ML folks in the industry. The scale of the application you are talking about is huge. # of documents and text to be processed would be huge. Also, first they would have to be digitized (e.g. FIRs are still handwritten) and then read as text. As of today, no ML framework (TF, Azure, PyTorch) can scale to that level. However, they will be in 2-3 years.

If See-Near IAS Aphsars and Judges agree that this has breakthrough potential it can reduce litigation delay from the present 10 years down to 2 days, at which point the deterrent effect of the law becomes real, leading to better bijnej growth due to certainty of enforcement and level playing field. Of course big entities will then "game" what the System will do and find loopholes which is where AI really comes in: doing the chess moves to stay ahead of the Bad Ones.

3. Medical diagnosis using x-rays, CT, MRI images via trained Neural Nets (CNN) are being attempted as well. Mostly in research labs. A phone picture based glaucoma or other retina anomaly detection etc have been attempted. Not commercial AFAIK. Given the sensitivity and compliance issues (such as HIPPA) real applications are hard to develop without training data at scale. Hospital Networks, Medical mfg, Pharma with big money are indeed working on these. Again, we should see scalable applications in 2-3 years.

This is where desh needs come into the picture. Take the 65% of India where "medical care" is many miles away. A lot of things that are taken for granted in Bhesht are simply not there. So it's not CAT scans, it is "Abdul fell down and is sweating and has a lump on his head. Speech is slurred".
Can you get Abdul in front of a camera so that Nurse Assistant Saira, 20 miles away can do a quick diagnosis and ask those present there to treat for concussion (... I was watching M*A*S*H on Ulan Bator TV yesterdin and Hawkeye was self-diagnosing, but I had to leave before the cure was done).

And then because Saira is an intelligent person, she types her observations into her system, which detects some anomalies in the combination of symptons. Her report is flagged to be seen by Dr. Gundappa Rao Singh in the best neuro facility in desh, who notices something strange - and orders timely intervention. Result: Top quality care reaches Abdul within an hour, in time to save his life.

IOW, terminology does not matter. Real-life application to make major practical breakthroughs matters.

UlanBatori · Post by **UlanBatori** » 28 Mar 2019 01:30

APji: FYI: I know nothing.

ArjunPandit · Post by **ArjunPandit** » 28 Mar 2019 01:36

UlanBatori wrote:APji: FYI: I know nothing.

UBji I think you should be banned for masquerading as a what you are

. Before I joined brf i used to read posts, there were 5 posters whose posts were on a diff league. You're one. Your post above tells a lot about nothing. (-1)chaiM. Nevertheless, in the spirit of no gravity waves:
https://barandbench.com/rbi-mandates-da ... -services/

UlanBatori · Post by **UlanBatori** » 28 Mar 2019 01:52

Thanks. Some ppl close to the top of the Planning in desh are very worried about the data aspect. Consider that most desh companies apparently base their servers in Singapore or Pindi or Shanghai or wherever. Best solution may be to flood them with garbage "data" to make them useless. Like UBCN reports filling a whole thread on Indian Defence Forces Order of Battle and Force Dispositions.

But I think this is for the Polis to worry about. I want to see much more focus on application of the basic capabilities to the villages. Because believe me, the situation there is what keeps me from rah-rahing about Present Guvrmand Achievements. Vast gulf between claims and reality, just like the "India Shining" days before Election 2004.

VKumar · Post by **VKumar** » 28 Mar 2019 02:38

In 1989 we had written an expert system using PROLOG. This was for the Indian insurance industry, for underwriting diamond exports.

UlanBatori · Post by **UlanBatori** » 28 Mar 2019 04:34

That is very interesting. I have written a Hello World in Prolog so I guess I am an AI expert!

ArjunPandit · Post by **ArjunPandit** » 28 Mar 2019 04:56

UlanBatori wrote:That is very interesting. I have written a Hello World in Prolog so I guess I am an AI expert!

the way current things are in industry, not just India, but also across the globe. I think you already are. But as I already posted, I think you are a manifestation of Alien tech....

ArjunPandit · Post by **ArjunPandit** » 28 Mar 2019 05:00

VKumar wrote:In 1989 we had written an expert system using PROLOG. This was for the Indian insurance industry, for underwriting diamond exports.

wondering why you didnt use SAS, python, all these have been available for long? i handnt touched a computer or as a matter of fact pen at that time...
on a serious note, did you include political connections or suranme as a factor? might have been useful to PnB

ArjunPandit · Post by **ArjunPandit** » 28 Mar 2019 05:01

UlanBatori wrote:That is very interesting. I have written a Hello World in Prolog so I guess I am an AI expert!

AIyo

UlanBatori · Post by **UlanBatori** » 28 Mar 2019 08:01

What we are trying to do, is to bring "gee-whiz" tech people (AI, Space, Climate Change, River Network for flood/drought control, economists, energy specialists.... ) together with social scientists and IAS types, to drive towards some really bold solutions. One is - imagine it - a swift and certain Justice system, avalable to all. Another is a universal wellness/healthcare system that integrates and synergizes Ayurveda/ traditional medicine with modern diagnostics, emergency care techniques, and cures - using tele/video presence to reach everywhere in India.

This is a fabulous opportunity for anyone who wants to contribute and participate, starting at the stage of developing Working Group Documents that define the problems, solutions, roadmap - and - Pilot Projects. With any luck we might get Neta/Baboon to actually pay attention and bring in big resources. At minimum we will have got the nerds to sit with the IAS types - and the village leaders - to get better understanding of what is needed, what works, and how to do it.

The Working Group Documents are the no-BS products. I know who is developing them.
No time to waste.

ArjunPandit · Post by **ArjunPandit** » 28 Mar 2019 08:07

Ok so what's the next stop??

UlanBatori · Post by **UlanBatori** » 28 Mar 2019 08:22

Working Group and Document. Email. Maybe we will post drafts here for ppl to expand?

TandavBrahmand · Post by **TandavBrahmand** » 28 Mar 2019 08:41

[quote="UlanBatori"][quote="TandavBrahmand"]AI is too broad. ML is probably we all are talking about here.

UBji,

Thank you - I always search and read your posts irrespective of topics, so it is great to get a response to a newbie like me!

Autonomy is too old school. A high-schooler can write a NB classifier these days over a weekend.
1That may be, and it's good to hear, but with a HUGE market for autonomous UAVs and self-driving gizmos, this may be a huge demand area too.

I was talking about Autonomy as in a company which was acquired by HP. Autonomy (HP) once had a technological edge in 90s that it has lost out and now no longer in the AI scene. I agree that autonomous (vehicles, drones, IED detection/removal etc are very much required and are in demand)

I am in full agreement about our needs are very unique and that provides us a huge opportunity to clean-up our systems and leapfrog the adoption of AI in legal, medical, rural settings. An excellent example is a Gurugram based agri startuphttp://www.imagoai.com/ - (Disclaimer - I have no investment/stake in except pointing an example of something very applicable to our needs)

I have one rant though - As a country, we are woefully lagging behind in AI. Cheen is investing heavily in AI at a massive scale. If we don't bridge the expertise gap, we will be at a massive disadvantage. They will overcome their declining population and fill that with AI-based competitive advantage. We have got to beat them. Sadly, we are not currently thinking as a 5-10 trillion dollar economy. My only hope is that our free-thinking spirit will enable innovations which cheeen will not be able to produce.

UlanBatori · Post by **UlanBatori** » 28 Mar 2019 08:51

^^ Thanks! This is shared by thinking people, and part of the objective is to develop an indigenous market push to drive massive efforts in AI.
From my limited observation, what has changed are (a) neural networks theory and technology (b) massively parallel, and cheap computation using graphics cards (NVIDIA), and (c) ability to connect millions of devices and people through fast Internet and 4G/5G etc.

Back in the 80s AI was visualized as millions of Rules and Case statements, with languages like LISP and PROLOG (I am BSing here, I happen to know someone who passed grad courses in these things..). We used to imagine developing Expert Systems for such things as troubleshooting intricate laser instruments, but the pain of writing up all the experience in these languages was not worth it. Also, Expert Systems were way too rigid.

Then came Fuzzy Logic and implementations of other Math. Plus highly nonlinear Neural Networks. Plus the NVIDIA video game cards. Plus the Cloud.

So it is true that India should be able to develop some exotic indigenous applications. For Indian markets first, then others. Perhaps finding the way to a given place is one example. In other countries you have maps with street names. In India you have directions like "near XYZ shop". Maybe Indian location services should go by such intelligence rather than hard GPS locations.

TandavBrahmand · Post by **TandavBrahmand** » 28 Mar 2019 09:10

^^ You are talking to one such dinasoor who has managed to survive! I did programming in LISP and did build a knowledge-based system for my thesis

. I also worked on India's first DOE-funded machine translation project using panini grammar as a notational representation language!

Neural nets existed in 80s but in addition to factors you have mentioned, the availability of massive data in text/visual (in future IoT signals) which allows these statistical learning systems to get trained (supervised or unsupervised) have made the difference.