Tuesday, 24 April 2018

Week 4 [23.04-29.04.2018] Google’s new AI algorithm predicts heart disease by looking at your eyes.

Hi!

In February, The Verge posted an article that describes the findings in the paper published in the "Nature" journal "Biomedical Engineering".

https://www.theverge.com/2018/2/19/17027902/google-verily-ai-algorithm-eye-scan-heart-disease-cardiovascular-risk

As we may observe, the direction of development diagnostic procedures is right now strictly related to new findings in the field of computer vision and deep learning analysis. The idea seems to be promising and it is developing dynamically.

However, if we take into consideration the report posted in 2013 (Frey C. B., Osborne M, A.: "The Future of employment: how susceptible are jobs to computerisation?", it could be searched easily if someone is interested) we can see that researchers at the University of Oxford were predicting that by 2050, nearly 50% of current professions will disappear as a result of broadly understood computerization, but this will affect most jobs that require automated, repetitive tasks rather than advanced thinking, intuition and problem solving.

Do you think that findings from this report are still actual? Or maybe it is now a little bit expired if we take into consideration that this prediction was made in the time when "deep learning revolution" in the October 2012 just begun? Going further: do you think that programming profession could be replaced, at least on some level, by the AI? I'm asking about that because sometimes we can meet the perception of a programmer being a craftsman and not an artist.

And finally, are you interested in the field of machine learning and/or computer vision? If you are using them in your professional or scientific activity, share with us your favourite toolchain. In particular, I'm interested in the languages/libraries, along with some information about why you chose the components you did and how they all fit together into your workflow.

Cheers,
Artur

55 comments:

  1. Obviously there are major changes since 2012 and I can not give the exact percentage number of current professions that will disappear by 2050, but I would say that it might be more than 50%. I think that developers are somehow replaced, when it comes to website design or front end development and the example is WIX platform. People know that it is much cheaper to use some wizard and spend some time to do it be themselves than to pay few thousands to developer that have some communication problems and at the end he will not get what he wanted.

    Of course I am interested in field of machine learning and computer vision, but rather as observer rather than doer. What is more, I guess all the blue collars will be replaced. If we start counting how many people work in manufacture, as drivers, cashiers and so on we will be surprised how large part of society can have problems with finding new jobs. Those people usually are unskilled and as we know ministry of education want to lower the number of students and higher education graduates. Personally I do not feel threaten because we already have more power than those people, but what will happen to them is a good open question.

    ReplyDelete
    Replies
    1. Thank you for your opinion. Yes, I agree that developers are replacing themselves by creating tools like WIX. Another platform that could be an example is logojoy.com where the algorithm is creating nice looking logos by the simple form of AI, which is potentially a threaten for lower-skilled designers.

      Delete
  2. I agree with the previous speaker. It's difficult at this time to determine how many and which occupations will be replaced (eliminated) by the automation of works. Earlier, people were folding cars. Today, cars are folding machines. Once the cashier counted the shopping bills, today the computer counts using bar codes. Will it ever be more than 50% time will show. I think that today it will be difficult to prophesy so accurate numbers!

    Yes, I look at systems that learn and replace people but at the moment I do not use them. I'm just an observer.

    ReplyDelete
    Replies
    1. Yes, I agree with that. Last year report called "The next era of human machine partnership" written by Dell Technologies with Institute for the Future (IFTF) says that as many as 85% of occupations of the future have not yet been created, so the numbers right now are very rough.

      Delete
  3. Thank you for an interesting article. We have been seeing the dynamic development of artificial intelligence. However, in my opinion, the situation is not so bad. When weaving machines became popular in the nineteenth century, the profession of a weaver was unnecessary. However, greater machine efficiency lowered fabric prices, which increased demand. As the result, the increase in employment in the textile industry was 4 times bigger than previously.
    I agree with the diagnoses described in the article that the number of administration employees will be reduced. Generally, the development of artificial intelligence will contribute to the revolution in the labor market. This will result from automation - now at a higher level. Certainly, in many professions there will be a significant drop in employment. How will it be with programmers? It is difficult to predict. The nature of the work of programmers will definitely change. The quoted text shows how artificial intelligence can improve the diagnosis. You can see that this is a risk of losing.

    ReplyDelete
    Replies
    1. Thank you for your comment. I agree with your sentence which says that the nature of the work of programmers will definitely change. I think that at the very beginning the AI would rather be a "helper" and not the "replacer".

      Delete
  4. I've heard some colleagues discussing the article you've mentioned - one of the opinions was, this research was getting dangerously close to iridology (i.e. "alternative" medicine).
    Following that, I proposed to predict a persons future, based on a 3D scan of their palm...

    After seeing it first hand, I have a bit different impression. An angiography with blood vessel segmentation can generally be correlated with blood pressure, and this is already a medical metric. However, it may not be quite enough to give a diagnosis based on just that.

    Now, about job elimination: I also think that 50% is a number with no meaning whatsoever. Does it mean "50% of the list of professions in some legal system"? Or 50% of the working population has jobs which will not exist?

    I think a vending machine is not the same thing as a grocery store or a cafe. But it does not preclude the real thing being replaced with a cheap, artificial ersatz. The bad money drives the good money off the market.

    The society will be transformed all right. But not in a way, which can be predicted now. Take the car for example. It would seem to be an effortless alternative to the horse, with unimaginable speed etc. However, now it is just a sheet metal container for everyday commute, keeping people seated (with all its health effects) and stressed out because of traffic. It also keeps our hands occupied and kills thousands of people a year. So now we can dream of driverless cars, which promise the luxuries we didn't get the first time around.

    ReplyDelete
    Replies
    1. Thank you for your comment, especially about the biomarkers, those are very deep insights indeed. We always have to be careful when we draw the conclusions (especially in the medicine) and remember that correlation is not causality.

      Delete
  5. This is not the area of my deep thoughts. But it is quite interesting. With a smile I look at the forecasts from years ago. Flying cars, etc. I think that in such predictions there is no definite scale. In my neighborhood is the largest cement plant in Poland. When my grandparents worked there, the company employed about 400-600 employees. Two years ago I was there on a trip - We went the way from limestone to cement bags on pallets. I met only 7 people. (I also count on the concierge lady). and it is only 2018.

    ReplyDelete
    Replies
    1. That is true, the automatisation level in the industrial task is pretty impressive.

      Delete
  6. I think that they are not actual, nowadays everything is moving much faster, technology, society, political changes, etc. Nearly 5 year gap is a world changer.
    Yes, at some level programming profession can be replaced by AI. A lot of programming work is making exactly same things every time. A lot of code could be written automatically based on some initial data. But I don’t think that AI (at least now) could write an application from A to Z.
    Due to my specialization (augmented reality) I nee to be interested in computer vision, this is the core of many AR concepts. Currently I’m working mostly with JAVA and C#. For computer vision I’m using OpenCV because it is the most popular choice BUT I’m looking more into direction of Tensor Flow and their capabilities of object recognition based on the frames from camera

    ReplyDelete
    Replies
    1. OpenCV is a very powerful tool for image processing, I found that mixing that with dlib (dlib.net) for ML tasks gives very nice results. However, Tensor Flow approach gives more handy programming interface so I think that you'll find it very useful.

      Delete
  7. "University of Oxford were predicting that by 2050, nearly 50% of current professions will disappear as a result of broadly understood computerization" - thats a really bold prediction that virtually pumps up ratings of "university" and researchers, but believe me, no one in 2050 will take care of checking who said what 30 years ago. It is quite natural for names and definitions of professions to change or merge with each other and it's quite probable that in few years the way a programmer will program will differ in a changed (enchanced) level of abstraction.

    Considering I'm not an expert from this field, though taking into account info from the article, I'd say ~70% accuracy indicates that scanning retina + taking few additional factors isn't a hollistic-enough to be used as a perfect tool. Considering your question about replacing some aspects of programmers profession I find it just as OK as using sewing machines by tailors : you can do thing quicker and earn more money but some tailors that enjoy doing and learning their craft to the very core of understanding can make extraordinary hand made works. In the same manner I perceive a programmer as a blacksmith that - depending on will - can work on massive production of bread knives and/or craft some decent blades that cut stuff and have soul (encapsulated energy of a craftsman within every part of design for a specific ) so I wouldn't be bothered by names (craftsman-artist quarrel is old as f*ck) - but rather focus on exploration of the subject and eventually both sides of the same coin will emerge.
    Regarding the last part of questions, paradoxically I'm not interested as much in machine learning and/or computer vision as in the reverse usage of those fields. Not inventing tools but taking for consideration that as an independent human being I would profit much more by consciously exploring given algorithms within myself and checking if those ways of data manipulation work in a 24/7-available environment everywhere I am.

    ReplyDelete
    Replies
    1. Thank you for your comment. I think that I have the same vision about AI in the programmer's job - it would be rather a tool/extension for the environment which would speed up your tasks, rather than independent, abstract thinking singularity.

      Delete
  8. I think this article is already a bit outdated. Five years in the IT industry is a very long period of time. Currently, considering that the revolution with the use of machine learning was just beginning, I think that these predictions would be much higher.
    Recently I read about the replacement of programmers by artificial intelligence, and such research is still carried out. I think, however, that this will not happen for a long time at this stage. Perhaps it will only be a help for programmers in performing the most mechanical tasks.
    I'm interested in artificial intelligence and its more and more interesting applications. However, I didn't have the opportunity to use tools based on it.

    ReplyDelete
    Replies
    1. Thank you for your comment. In my opinion, the reality is moving so fast that this kind of approximations (about % of jobs that would be replaced by the machine) should be done at least annually.

      Delete
  9. I have to agree with Dawid P. This assumptions are not actual in nowadays. We can take a look into our are of interests. IT evaluates extremely fast and if you base on technologies that were efficient 3 years ago, you can not gain an advantage on the market.

    Part of our job (programming) is repeatable and I am sure that we can delegate this part of job to AI. Interpreter translates our code, so maybe AI can translate text to the code :)

    I am interested in machine learning. Especially in Deep Learning. Computer vision is part of this field but I am not an expert yet. Now I am focusing on Genetic Algorithm that will be able to predict events executed in parallel.

    ReplyDelete
    Replies
    1. Thank you for your comment, your research area sounds to be very interesting indeed.

      Delete
  10. I use machine learning to predict variables. I most often use binary classifiers. In my work, I investigate the relationship between healthy people and mental illnesses. The input data is the activity of a Brodman's area's I am trying to find a classification model that would distinguish healthy from sick people. Then I will try to classify the individual mental submersions. Each neural network uses indicators on the basis of which it trains. This article does not provide precise information on which indicators have been used for machine learning. I use AzureML to check the operation of the neural network. For more detailed analysis I use R

    ReplyDelete
    Replies
    1. Very attractive research field. Which Brodmann areas are you actually studying? Do you base your research on EEG records, fMRI scans or something else? What kind of mental illnesses are you researching? If you find some time, please elaborate because I'm also interested in this field.

      Delete
  11. Hmmm about actually of this kind of papers reports and others. When you look at this kind of documents from IT perspective, then you can say yes they are old. But when it comes to IT in Medicine answer is completly different. I am doing research on medical imaging so I am using ANN a lot. So when I am looking for somthing for example about training networks on a small batches most of valueble information will be from around 2015/2016 and later you will have some variations on previous stuff. Also some ideas were invented many years before paper was published (Capsule Networks) so I don't think that date may tells you something important.
    When it comes to my toolchain usually I use Python 3 (sometimes C++) from toolkits there will be Keras, some of my networks are written in PyTorch ( I am not very big fan of Tensorflow), sometimes I can use some libraries from Opencv and so on. So it depends on what I need to do.
    I don't think that in near future computer will start to program themselves (as in BattleStar Galactica) and after 11+ years of programming I must say that programming is craftsmanship not art.

    ReplyDelete
    Replies
    1. Thank you for your comment. In my opinion, there is nothing bad in taking well-known technology and applying it to the new task as long as it gives you results :)

      Delete
  12. I think that some jobs will be replaced by Artificial intelligence but the new ones will be created, not only requiring high creativity and special skills. For instance to train artificial network we need a lot of data. Of course data is everywhere but we need to collect it, tag and prepare to further analysis. For example tagging the data usually is long and not very creative process. Scientist very often hire people (or use crowdsourcing) to do that.

    The way we do programming is changing all the time. People started with assembler and now we have such languages like python which are quite easy to learn. I think that also artificial intelligence will make programing easier that replace it at all.
    In my work I use mainly Python, scikit learn, Keras and sometimes R. It depends on the project.

    ReplyDelete
    Replies
    1. Indeed, the majority of machine learning practitioners actually consider the data to be more important than the algorithms themselves.

      Delete
  13. I agree with this article, Many professions will be lost due to computer occupation, we can already see the appearance of vehicle with artificial intelligence, which are likely to soon take jobs from taxi drivers, public transport drivers and truckers. And unfortunately this will affect many other professions.
    But progress and automation were always, always some professions left, some came, maybe not as intensively as now. But I think we can all adapt.
    Also, I think the profession of the programmer is now not threatened, since it is the programmers who will be these intermediaries who will select other people's jobs

    ReplyDelete
    Replies
    1. I think yes, transportation automatisation would affect many jobs. However, I'm also waiting for the moment when transport would be fully automatized which hopefully expose in less fatal accidents on the roads.

      Delete
  14. It is really hard to predict what jobs will be available in 30 years. Who would think about drone operator, social media coordinator or SEO specialist 10 years ago. Perspective of 30 years is pretty long. For sure there will be large amount of jobs that will perish due to overall computerization and development in computer science field. But also there will be large amount of jobs that will be created.

    In my opinions programming should be treated as a craftsmanship (which is really hard to master). I do not think that jobs related to computer science will perish due to development in AI field, but I am pretty sure there will large changes in this matter.

    Af for toolchain
    Python as language of choice, PyCharm as development environment
    Anaconda as package and environment manager
    Tensorflow as low level library
    Keras as high level library
    Sk-learn for preprocessing
    Numpy, Pandas, MatplotLib for data analysis

    ReplyDelete
    Replies
    1. That is true, as I the report mentioned earlier says, many of future jobs are not even existing yet. Thank you for sharing your toolchain, I'm using very similar since few weeks.

      Delete
  15. I think it is highly possible that half of jobs could dissapears because of technology expansion till 2050 - if you look back in the past, you will notice that 50 years ago there was many jobs that dissapeared - but it doesnt means that people will have no work to do - they just changed their brand, besides that many new jobs have appeared, and still are appearing.
    I think that programmers will always have something to do - we are still humans and our desire for new applications, technologies will never be satisfied, and our imaginations have no limits. And I think we are still children who dont know anything about universe

    ReplyDelete
    Replies
    1. The sky is the limit so basically - there is no limit :)

      Delete
  16. Definitely, nothing can replace a human, even in programming ;-) Apart from that, I think you can be virtuoso in everything. AI can replace any profession, but in very narrow spectrum, I mean and I believe – simple activities can be executed by machines. Artistic, creative – it will always be human’s domain.
    Professionally I have nothing to do with this field of science or research, so I can’t answer in terms of technological content. But the issue is very interesting, has huge potential and I am afraid all of us will have to get more familiar with it, sooner or later, also professionally.
    That’s because AI & “deep learning” issues influence how we think or decide. I am sure the technology like social media can be powerful in changing human beliefs and behavior. For example, GoldenLine or LinkedIn send me jobs offers basing on the offers I reviewed lately. These new ones are not often even similar to the previous ones, but the system somehow knows how to direct my attention. This is a positive example, but there is a lot of less nice cases, when – especially advertisements – can make me nervous very fast, because they are appearing basing on old data, in spite I don’t need them.
    Thanks for the article. I have to underline, that I am always very happy, when I can see that science can improve our life in medical treatment aspect :-)

    ReplyDelete
    Replies
    1. That is true that the marketing automatons are very naive right now. I really don't like how for example booking.com makes follow up based on your historical search and not taking into consideration that you've already visited this place.

      Delete
  17. Hard to say. Certain professions may disappear. Currently in factories people are being replaced by robots and autonomous cars are tested on the street. Recently I found the article https://www.dobreprogramy.pl/Sztuczna-inteligencja-juz-dzis-pisze-kod-za-programiste-czy-jutro-go-zastapi,News,87796.html about AI who writes the code for the programmer. As you can see everything is possible. Artificial intelligence has great potential and can do more and more. Current IDE environments have smart hints that improve the work of the programmer. AI in the future can largely offload the programmer and maybe replace it. The only question is will it be safe for human?
    Machine learning is a very interesting tool with great possibilities. I don’t use machine learning every day but I've read about it. I tested libraries available in Python

    ReplyDelete
    Replies
    1. Thank you for your comment. Python is being very widely used by the data scientists nowadays thanks to wide range of handy ML toolkits.

      Delete
  18. First, I don't even know how to count this 50%. Let's take programmers: is it one profession or more like a dozen: system, webdev, game on mobile, game on consoles, research, automotive ... programmers?
    Second, we can look at professions in narrow way (truck driver - drives a truck) or wider (truck driver - drives, maintains vehicle, deals with customs officials, devises long term strategy, and many more things I don't know about because I didn't work as one). It might be easy to replace main activity associated with profession, but then there are all those hidden activities, that can't be automated/replaced.

    As for software developement, Andrej Karpathy (one of scientists behind current AI Spring) in his article, "Software 2.0" specifies "software 2.0" as complementary to old ways of "software 1.0", rather than replacing it. And I agree with this view: we can train/generate some parts of software, but other parts are still to be written as before.

    That is until we create Artificial General Inteligence, then it's all over.

    ReplyDelete
    Replies
    1. Thank you for your comment. I'm really curious about the birth of the singularity predicted by Kurzweil and its aftermath.

      Delete
  19. A human part in the process of software development is still very important and probably will be in a foreseeable future, even if that human is only a craftsman. Maybe AI can generate a working code, but only for well defined and specific task. Human, however, even if he's not a skilled programmer can collect requirements communicated in a natural language, often quite vague, understand them somehow and turn into code. Later, he can confront his work with expectations of a requester, improve his understanding of requirements, write a better code and so on. Basing on the current state of the art of machine learning we're not anything close to it yet.

    Yes, I'm interested in machine learning and computer vision. I use Python as a language of choice. As for libraries I use mainly OpenCV for computer vision as it has many popular algorithms implemented and easy interface for python. For machine learning I use scikit-learn, keras or tensorflow. From infrastructure perspective recently I'm a fan of google colab and google cloud platform.

    ReplyDelete
    Replies
    1. I do agree with you. If we take into consideration machine learning and when we cut the media hype, we're not anything close to it yet.

      Delete
  20. 1) Do you think that findings from this report are still actual?
    It’s hard to say because we are talking about the future in 30 years so it’s a kind of reading tea leaves. I agree that some professions will disappear but new ones will appear which will become new jobs. History repeats itself…

    2) Do you think that programming profession could be replaced, at least on some level, by the AI?
    Yes, everything tends to such situation. Of course it wouldn’t be possible to replace programmers at all. It seems to me that it will be the same as it is with painters - you can print everything but the art is still in the price.

    3) Are you interested in the field of machine learning and/or computer vision?
    Machine learning is a very interesting topic but it’s not my cup of tea. Nevertheless, I see that this is an increasingly popular issue that will undoubtedly be widely used in the future.

    ReplyDelete
    Replies
    1. As one said, you can replace one senior software developer by the finite number of IT students ;). Maybe the same situation would be with the coding by AI. You will achieve "some" results by the AI but the finishing touch will be done by the human.

      Delete
  21. Thank you very much for an interesting article. I agree with some of the previous speakers, I guess that even more than 50% of professions might disappear. Though, I don't agree that this phenomenon will affect all professions that are "repetitive" by their nature. I would say that those will be professions that require no or almost no qualifications from people. There are many professions very repertitive by their nature that in my opinion, won't disappear, as they require skills and such level of imagination and creativity that the representative of those jobs will never be replaces by machines, computers, AI etc. I am very interested in this topic and I am pretty sure that at certain point it will be hot issue for all the people around the globe.

    ReplyDelete
    Replies
    1. Thank you for your comment and for sharing your opinion.

      Delete
  22. So many indirect questions.
    Do you think that findings from this report are still actual?
    Why not? Even though this article is far from being a sufficient explanation.
    Or maybe it is now a little bit expired if we take into consideration that this prediction was made in the time when "deep learning revolution" in the October 2012 just begun?
    2012 is not that long ago. I would argue with findings from the 80s.
    Do you think that programming profession could be replaced, at least on some level, by the AI?
    Never, why? Because business will never learn to express their needs in a proper way. You need to communicate with them in an unpredictable way. It's not a new concept, companies are trying to make AI smart. But, there is always something. If we teach our AI based on blocks and so on, then there is a high chance that our smart AI can deliver something usable. However, based on my experience I can't imagine that interaction business requirements-AI will lead to something usable and close to actual needs.

    And finally, are you interested in the field of machine learning and/or computer vision?
    Machine learning, but only as a hobby. It sometimes helps with my study.

    ReplyDelete
    Replies
    1. You've put on the table very interesting problem. Yes, the business has a very large issue with expressing their needs to another person (I'll not mention about the problem of inner agreement). I also think that there should be still another person who would at least filter noisy data and replace it with requirements.

      Delete
  23. Thanks for such an interesting article. Artificial Intelligence has gone through a rapid development so 5 years makes it out-of-date. I think that today programming profession can be replaced by Artificial Intelligence in some fields. Code is can be performed by a machine when it is predictable, I think. I am thinking of the exactly the same pieces of code rewritten on and on. It is all about teaching these algorithms. I am interested in Artificial Intelligence and I look for opportunities for any promiscuous applications. It helps me in my research sometimes.

    ReplyDelete
  24. Of course, some professions will disappear and more will appear in their place. History likes to be repeated. It is difficult to determine whether the author is wrong or wrong, it is the diration of tea leaves. When it comes to the profession of programmers, it is already in many areas zapampiany, for example when creating websites through platforms created for it.

    Unfortunately, I do not deal with machine learning so I will not share the toolchain.

    ReplyDelete
  25. I'm convinced that big part of occupations will disapear. When 12 years ago I have started training for air traffic controler I was told that I will lost this job before retirement. Now I see the problem with software implementation is not with comlexity but stability.

    I think simpler parts of programers job will be automated. That is why we use modules/libraries :) I use the same tools as Ewelina Winska.

    ReplyDelete
    Replies
    1. Cool! I agree that stability is crucial and it is very hard to achieve the level of SLA=100%. Is that true that traffic control could be done when there is a blackout because every flight is managed not only by the computers but also it has parameters written on the paper?

      Delete
  26. Do you think that findings from this report are still actual?
    I think they’re still actual. There are many jobs which can be replaced by AI and robots. This is a natural course of events. Like we have washing machines and dishwashers when in the old times our ancestors were used to do they dishes by themselves. I hope soon we live to see autonomous cars instead of dishonest taxi drivers, who are a huge problem for tourist around the world.

    Do you think that programming profession could be replaced, at least on some level, by the AI?
    I think that is happening right now. Google said that their autoML system creates better AI system than Google’s engineers.

    And finally, are you interested in the field of machine learning and/or computer vision?
    Like Andrzej Kawiak said, in our job we use Azure Machine Learning for finding difference between control group and people with mental disorders where the input data is the activity of a Brodman's area.

    ReplyDelete
  27. Despite the great hopes associated with machine learning, I think that programmers will be irreplaceable. Of course, part of them job can be done by algoritm but not all of them. Algoritms are bad in create new, trendy model of pages. If all pages on the Internet would be similar, none of them wouldn't be remember by users. The biggest benefits come from pages which are original and unconventional.
    Other professions will be replaced-it's obvious. Now the car can drive safety and always choose the fastest way.

    I'm interested in machine learning, but to be honest I'm beginner. I using tensorflow to recognize shapes on video monitoring (I'm trying...).

    ReplyDelete
  28. Thanks a lot for sharing this amazing knowledge with us. This site is fantastic. I always find great knowledge from it. Coaching Institute in Mumbai

    ReplyDelete
  29. Thanks for this. I really like what you've posted here and wish you the best of luck with this blog and thanks for sharing. Stylist Course Modules

    ReplyDelete