The Lindahl Letter

Right before sitting down to work on this post I recorded the audio for week 108. My voice has almost recovered all the way from catching that pandemic thing in December. It’s interesting that sustaining vocal quality would be one of the last things to stabilize from that distinctly unwanted experience. You have probably been able to hear that within the last few recorded podcasts. I just did not have that totality of vocal control. I might go back and record that audio segment again, but at this point it is entirely possible that won’t happen based on time and the weight of other commitments this week. I was also a little concerned when Audacity, the software I use for this recording, required an update before the recording started, but it was apparently a small one and I was able to start the recording process within a couple of minutes. It’s very possible that this missive would go out without an audio edition. You however know that is not the case.

It turned out that I was able to sit down and record the audio version of this Substack post. Right now I’m working without my backlog due to some unavoidable things that disrupted my writing routine a bit. I’m hopeful that given a little bit of time a few weeks of backlog will build back up, but right now each post is as fresh as it will ever be as things are being written directly before being recorded as part of my weekend writing routine. Typically it is better to have the posts created within the 5 week planning and review cycle. At this point, that is not possible. Welcome to the now and being in the moment as the very freshest words are brought your way within this missive of consideration for Substack. 

For the most part households adopted some machines pretty quickly and in a sustained way. Depending on where you are, microwaves, refrigerators, freezers, dishwashers, washing machines, and dryers are a part of the technology footprint in households. None of those devices need to be connected to either WiFi or Bluetooth to be operational. They existed for many years without that type of connectivity. This essay happens to be about robots in the house and the aforementioned appliances are not really what people talk about in terms of modern robotics in the household. It’s something more mobile that gets a lot of attention. To be fair a lot of robot vacuum brands and companies now exist. They roam and clean, get stuck, and have to be rescued and maintained. 

I went out to find some papers that referenced the Roomba and they seem to have peaked between 2006 to 2007. Which I thought was a very interesting element to see within this literature review. They just sort of stopped in frequency. Here are three of them that were well referenced. 

Forlizzi, J., & DiSalvo, C. (2006, March). Service robots in the domestic environment: a study of the roomba vacuum in the home. In Proceedings of the 1st ACM SIGCHI/SIGART conference on Human-robot interaction (pp. 258-265). https://www.cs.cmu.edu/~kiesler/publications/2006pdfs/2006_service-robots-roomba.pdf

Tribelhorn, B., & Dodds, Z. (2007, April). Evaluating the Roomba: A low-cost, ubiquitous platform for robotics research and education. In Proceedings 2007 IEEE International Conference on Robotics and Automation (pp. 1393-1399). IEEE. https://citeseerx.ist.psu.edu/document?repid=rep1&type=pdf&doi=15cb2263caf26ac68906858093ae8d7749ad7827 

Jones, J. L. (2006). Robots at the tipping point: the road to iRobot Roomba. IEEE Robotics & Automation Magazine, 13(1), 76-78. https://www.researchgate.net/profile/Joseph-Jones-14/publication/3344755_Robots_at_the_tipping_point_the_road_to_iRobot_Roomba/links/5728aa7308ae2efbfdb7dce8/Robots-at-the-tipping-point-the-road-to-iRobot-Roomba.pdf 

None of these robots in the house are equipped with any conversational subroutines. Within the worlds created by science fiction writers it is not uncommon for the robots in the house to talk back and demonstrate some degree of personality. We currently have no legitimate AGI that would facilitate that type of exchange. It’s probably up next at some point. Now that both Microsoft (powered by OpenAI) and Google are trying to create chat-like interactions I’m guessing that chatting with the robots in the house will arrive as a feature. Right now in Google Scholar searches you can find almost 5,000 results for ChatGPT [1].

Links and thoughts:

Top 6 Tweets of the week:

Footnotes:

[1] https://scholar.google.com/scholar?hl=en&as_sdt=0%2C6&q=chatgpt&btnG= 

What’s next for The Lindahl Letter? 

* Week 110: Understanding knowledge graphs

* Week 111: Natural language processing 

* Week 112: Autonomous vehicles

* Week 113: Structuring an introduction to AI ethics

* Week 114: How does confidential computing work?

If you enjoyed this content, then please take a moment and share it with a friend. If you are new to The Lindahl Letter, then please consider subscribing. New editions arrive every Friday. Thank you and enjoy the year ahead.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.nelsx.com

Part of what made Twitter so interesting is the diversity of argument and the townhall nature of it being the first place things show up in the feed. I’m not sure any other company or social platform could attract the same amount of hyperactive content creation users geared at news and coverage of the moment. People are arguing and I’m sure papers will soon be arriving to describe the end of social media. This weekend I’m going to spend a bit of time reading a book by Robert Putnam of “Bowling Alone” fame called “The Upswing”.

Putnam, R. D. (2000). Bowling alone: The collapse and revival of American community. Simon and schuster.

Putnam, R. D. (2020). The upswing: How America came together a century ago and how we can do it again. Simon and Schuster.

It’s probably somewhere in the meta analysis between social capital and social media that a compelling story about why Twitter as a company would not happen today exists. During the course of this analysis you are going to receive two different lines of inquiry. First, I’ll consider the nature of Twitter and a few books related to it and silicon valley in general. Second, we will dig into some of the AI and sentiment analysis scholarly work related to that field of study to help keep the writing trajectory for the year on track. 

Books have arrived to tell the stories of what happened in Silicon Valley. A lot of unlikely things happened within the borders of the space described as silicon valley. Some of them will be a part of business courses for decades to come. It truly is an interesting thing that happened where so much creativity and output happen in such a relatively small area. 

Three of the books that I have enjoyed are listed below.

Bilton, N. (2014). Hatching Twitter: A true story of money, power, friendship, and betrayal. Penguin.

Frier, S. (2021). No filter: The inside story of Instagram. Simon and Schuster.

Wiener, A. (2020). Uncanny valley: A memoir. MCD.

You can zoom out a bit and grab some classic silicon valley reading like:

Isaacson, W. (2014). The innovators: How a group of inventors, hackers, geniuses and geeks created the digital revolution. Simon and Schuster.

A lot of scholars over the years have focused their attention on Twitter for a variety of purposes. You can imagine that my interest and the interest of those scholars overlap around the ideas of AI and sentiment analysis. Digital agents abound within the Twitter space and some of them are doing some type of sentiment analysis with what scholars are identifying as artificial intelligence. That second part of the equation makes me a little bit skeptical about the totality of the claims being made. We will jump right into the deep end of Google Scholar on this one anyway [1].

Papers from a search for “Sentiment analysis Twitter artificial intelligence” [2]

Kouloumpis, E., Wilson, T., & Moore, J. (2011). Twitter sentiment analysis: The good the bad and the omg!. In Proceedings of the international AAAI conference on web and social media (Vol. 5, No. 1, pp. 538-541). https://ojs.aaai.org/index.php/ICWSM/article/download/14185/14034 

Ghiassi, M., Skinner, J., & Zimbra, D. (2013). Twitter brand sentiment analysis: A hybrid system using n-gram analysis and dynamic artificial neural network. Expert Systems with applications, 40(16), 6266-6282. 

Giachanou, A., & Crestani, F. (2016). Like it or not: A survey of twitter sentiment analysis methods. ACM Computing Surveys (CSUR), 49(2), 1-41. https://arxiv.org/pdf/1601.06971.pdf 

Alsaeedi, A., & Khan, M. Z. (2019). A study on sentiment analysis techniques of Twitter data. International Journal of Advanced Computer Science and Applications, 10(2). https://www.researchgate.net/profile/Abdullah-Alsaeedi/publication/331411860_A_Study_on_Sentiment_Analysis_Techniques_of_Twitter_Data/links/5c78175ba6fdcc4715a3d664/A-Study-on-Sentiment-Analysis-Techniques-of-Twitter-Data.pdf 

I had considered some evaluation of searches for both “opinion mining Twitter artificial intelligence” and “artificial intelligence analysis of public attitudes” [3][4]. It’s possible some papers from both of those searches show up later. Generally, all of that argument and content could be broken down into two camps of intelligence gathering related to advertising and general opinion mining geared at understanding sentiment. One divergent thread of research from those two would be some of the efforts to identify fake or astroturf content. You can imagine that flooding either fake or astroturf content could change the dynamic for advertising or sentiment analysis. Advertising to a community of bots is a rather poor use of scarce resources.

Links and thoughts:

Top 5 Tweets of the week:

Footnotes:

[1] https://scholar.google.com/scholar?hl=en&as_sdt=0%2C6&q=twitter+artificial+intelligence&oq=twitter+artif  

[2] https://scholar.google.com/scholar?hl=en&as_sdt=0%2C6&q=Sentiment+analysis+Twitter+artificial+intelligence&btnG= 

[3] https://scholar.google.com/scholar?q=Opinion+mining+Twitter+artificial+intelligence&hl=en&as_sdt=0&as_vis=1&oi=scholart 

[4] https://scholar.google.com/scholar?hl=en&as_sdt=0,6&qsp=4&q=artificial+intelligence+%22analysis+of+public+attitudes%22&qst=ir 

What’s next for The Lindahl Letter? 

* Week 109: Robots in the house

* Week 110: Understanding knowledge graphs

* Week 111: Natural language processing 

* Week 112: Autonomous vehicles

* Week 113: Structuring an introduction to AI ethics

If you enjoyed this content, then please take a moment and share it with a friend. If you are new to The Lindahl Letter, then please consider subscribing. New editions arrive every Friday. Thank you and enjoy the year ahead.

Lindahl, N. (2023). The Lindahl letter: 104 Machine Learning Posts. Lulu Press, Inc. https://www.lulu.com/shop/nels-lindahl/the-lindahl-letter-104-machine-learning-posts/ebook/product-y244ep.html 

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.nelsx.com

This week it seemed like a good idea to take a look at some of the most highly cited AI papers. Back during week 81 of this journey into writing on Substack, I took a look at some of the most highly cited ML papers [1]. I was expecting a lot more overlap, but was pleasantly surprised at the differences. One of the papers really stood out based on the total number of citations and it’s up first. Intellectually I can accept that a paper has more than 100,000 citations, but in practice that is an awful lot of references for an academic paper to have and a representation of a degree of asynchronous interaction between researchers that helps bring the intellectual space called the academy to life.

Papers with over 100,000 citations:

* Kingma, D. P., & Ba, J. (2014). Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980. https://arxiv.org/pdf/1412.6980.pdf 

Papers with over 50,000 citations:

* Ren, S., He, K., Girshick, R., & Sun, J. (2015). Faster r-cnn: Towards real-time object detection with region proposal networks. Advances in neural information processing systems, 28. https://proceedings.neurips.cc/paper/2015/file/14bfa6bb14875e45bba028a21ed38046-Paper.pdf 

* Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., ... & Polosukhin, I. (2017). Attention is all you need. Advances in neural information processing systems, 30. https://proceedings.neurips.cc/paper/2017/file/3f5ee243547dee91fbd053c1c4a845aa-Paper.pdf 

Papers with over 20,000 citations: 

* Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A. A., Veness, J., Bellemare, M. G., ... & Hassabis, D. (2015). Human-level control through deep reinforcement learning. nature, 518(7540), 529-533. https://daiwk.github.io/assets/dqn.pdf 

* Ioffe, S., & Szegedy, C. (2015, June). Batch normalization: Accelerating deep network training by reducing internal covariate shift. In International conference on machine learning (pp. 448-456). PMLR. http://proceedings.mlr.press/v37/ioffe15.pdf 

Papers with over 10,00 citations:

* Silver, D., Huang, A., Maddison, C. J., Guez, A., Sifre, L., Van Den Driessche, G., ... & Hassabis, D. (2016). Mastering the game of Go with deep neural networks and tree search. nature, 529(7587), 484-489. 

* Arjovsky, M., Chintala, S., & Bottou, L. (2017, July). Wasserstein generative adversarial networks. In International conference on machine learning (pp. 214-223). PMLR. http://proceedings.mlr.press/v70/arjovsky17a/arjovsky17a.pdf 

* Kipf, T. N., & Welling, M. (2016). Semi-supervised classification with graph convolutional networks. arXiv preprint arXiv:1609.02907. https://arxiv.org/pdf/1609.02907.pdf 

Without question these levels of citation are an indication that the works are being read and actively referenced within the scholarly community. I’m referencing the citation numbers here to give you a sense of scale when it comes to considering the AI community and how many people are researching and considering the things happening in this space. This is a very crowded and vibrant place in the academy where a lot of time and effort are going into moving things along toward building very real and deployable technology in this space. Given the sheer volume of people working in this space it’s only a matter of time before somebody will shout “Eureka!” and we see practical deployments in production which will influence our daily lives.

What would ChatGPT create?

If you were wondering what ChatGPT from OpenAI would have generated with the same prompt, then you are in luck. I had that output generated over at https://chat.openai.com/chat by issuing a prompt.

Links and thoughts:

Top 5 Tweets of the week:

Footnotes:

[1] Week 81 of The Lindahl Letter:

What’s next for The Lindahl Letter? 

* Week 108: Twitter as a company probably would not happen today

* Week 109: Robots in the house

* Week 110: Understanding knowledge graphs

* Week 111: Natural language processing 

* Week 112: Autonomous vehicles

If you enjoyed this content, then please take a moment and share it with a friend. If you are new to The Lindahl Letter, then please consider subscribing. New editions arrive every Friday. Thank you and enjoy the year ahead.

Lindahl, N. (2023). The Lindahl letter: 104 Machine Learning Posts. Lulu Press, Inc. https://www.lulu.com/shop/nels-lindahl/the-lindahl-letter-104-machine-learning-posts/ebook/product-y244ep.html  

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.nelsx.com

You probably were wondering how long into the new year before a bunch of focus and attention were placed at the efforts of Hugging Face. You won’t have to wait any longer as this missive will dig into BigCode and some other efforts to use AI to build out code generating systems [1]. I was reading a TechCrunch article from Kyle Wiggers and wondering about how many different systems existed [2]. That article references 5 code generating systems that you could in practice elect to go evaluate. For completeness I’m listing Codex and Copilot separately in this last given that the interfaces are holistically different. 

* BigCode - Hugging Face & ServiceNow’s R&D division [3]

* AlphaCode - DeepMind [4]

* CodeWhisperer - Amazon [5]

* Codex - OpenAI [6]

* Copilot - GitHub (Codex based) [7]

One of the things you might be interested in learning about at this point would be a dataset called “The Stack” which happens to be a collection of 6 terabytes of permissive code data that covers 300 programming languages [8]. The permissive code part of the dataset is interesting. The GitHub archive was roughly 69 terabytes of data that they filtered by licensing which they considered permissive and ended up with that 6 terabyte collection. Understanding how the dataset that feeds the code generating system was built is very important. All my contributions on GitHub are intended to be MIT license which I think should be permissive [9]. You have to deeply consider that a lot of propriety code writers and corporations employing said coders would not have given permission to use their code in a code generation system. 

Generative coding systems will abound shortly and are in an early and developing state at the moment. We are getting to the point where you can instruct Codex to build something code related in terms of creating an application and you might get a great result. It’s not a universal code generation engine at this point. However, we are getting closer and closer to conversational code generation or some flavor of that outcome which I would classify as a generative coding system. It will be a seismic shift in code generation based on democratizing the creation of applications. 

What would ChatGPT create?

If you were wondering what ChatGPT from OpenAI would have generated with the same prompt, then you are in luck. I had that output generated over at https://chat.openai.com/chat by issuing a prompt.

Links and thoughts:

Top 5 Tweets of the week:

Footnotes:

[1] https://huggingface.co/bigcode 

[2] https://techcrunch.com/2022/09/26/hugging-face-and-servicenow-launch-bigcode-a-project-to-open-source-code-generating-ai-systems/ 

[3] https://www.bigcode-project.org/

[4] https://alphacode.deepmind.com/

[5] https://aws.amazon.com/codewhisperer/ 

[6] https://openai.com/blog/openai-codex/ 

[7] https://github.com/features/copilot 

[8] https://huggingface.co/datasets/bigcode/the-stack 

[9] https://github.com/nelslindahlx 

What’s next for The Lindahl Letter? 

* Week 107: Highly cited AI papers.

* Week 108: Twitter as a company probably would not happen today

* Week 109: Robots in the house

* Week 110: Understanding knowledge graphs

* Week 111: Natural language processing

If you enjoyed this content, then please take a moment and share it with a friend. If you are new to The Lindahl Letter, then please consider subscribing. New editions arrive every Friday. Thank you and enjoy the week ahead.

Lindahl, N. (2023). The Lindahl letter: 104 Machine Learning Posts. Lulu Press, Inc. https://www.lulu.com/shop/nels-lindahl/the-lindahl-letter-104-machine-learning-posts/ebook/product-y244ep.html  

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.nelsx.com

We are now starting the third year journey of The Lindahl Letter on Substack. You can find the first two years worth of that effort combined into one manuscript right here or at any other online retailer:

Lindahl, N. (2023). The Lindahl letter: 104 Machine Learning Posts. Lulu Press, Inc. https://www.lulu.com/shop/nels-lindahl/the-lindahl-letter-104-machine-learning-posts/ebook/product-y244ep.html  

During the course of this next year I’m probably going to write a book about how I would set up a modern polling company or to put it more bluntly, a book about how to monetize the sentiment measuring process in modern America. Part of that book writing effort could end up being a part of this series or it could end up being a stand alone set of content. Yes, that would mean 2023 would yield at least two books of exceptionally enjoyable reading content. 

Overall, this new writing direction is one of those things that I’m going to really dig into throughout 2023. This would be inline with my research trajectory and current five year writing plan. My thoughts have just become a little more focused on what exactly I would produce within that writing window. Naturally, that is a good thing to start forming up before the writing project begins. I’m going to need to sit down and write up the topics to include within my literature review on this one. That part of this analysis will be key and has to be pretty darn good at this point. 

Within that work effort I’m considering a few article titles to include in the writing plan:

* “Sentiment pooling: Applied multimedia polling”

* “Beyond the phone: A study of multimedia polling methods”

* “When nobody answers: Evaluating the efficacy of phone based polling”

* “Automated sentiment analysis compared to respondent based sentiment”

* “Beyond the paywalls of academic journals: how information is being shared more broadly.”

We will see which of those ends up getting produced as either book chapters or true stand alone articles. This last week of planning and review was actually pretty productive. I have sketched out a writing plan for about the next 25 weeks. This new Substack series now has a backlog to work against as posts get put into a five week planning and review cycle. As we move from 2022 to 2023, my primary weekly Substack based content creation is going to shift a bit from ML related coverage to a stronger AI focus mixed with more content on modern polling methodologies. 

You might be wondering why the combination of AI and sentiment analysis (polling methodologies) happens to be important to me as something to consider. As the intersection of technology and modernity occurs certain things will happen. Part of understanding civil society and figuring out what happens to civility during that intersection will require better active sentiment analysis. Critical thought and analysis into understanding the nature of measuring modern sentiment will be essential on a go forward basis based on the automation that AI will introduce throughout business.

I was looking at the proposed content from weeks 105 to 132 and realized that ethics as a part of the AI journey and process were left out as an early topic. I’m going to try to work in some evaluation of and references to academic works that consider ethics and responsible AI each week as we move forward. 

As we close out here at the start of this series it would be only fitting to share what I do consider to be a foundational AI book to consider as key reference material. Probably the best book about AI you could grab is a textbook called, “Artificial Intelligence: A Modern Approach,” by Stuart Russell and Peter Norvig [1]. The current version appears to be the 4th edition [2]. You can get a copy on eBay or some other used book proprietary for a lot less than the academic asking price. 

Some people prefer the findings of a certain alternative deep learning historian. I don’t want to deprive those readers of a link to what they might be looking for as fresh reading material. I am aware that Jürgen Schmidhuber has started writing some histories. It does present some alternative findings and different points of view. You can now access and read 75 pages of that content that have been recently published and are pretty easy to download from arXiv.

Schmidhuber, J. (2022). Annotated History of Modern AI and Deep Learning. arXiv preprint arXiv:2212.11279. https://arxiv.org/ftp/arxiv/papers/2212/2212.11279.pdf 

Outside of those two works a book that would be more conversational and easier overall to absorb about where AI is at current would be this one:

Kissinger, H. A., Schmidt, E., & Huttenlocher, D. (2021). The age of AI: and our human future. Hachette UK.

This is the start of a series that will last 52 weeks. It should be an interesting journey. 

What would ChatGPT create?

If you were wondering what ChatGPT from OpenAI would have generated with the same prompt, then you are in luck. I had that output generated over at https://chat.openai.com/chat by issuing a prompt.

Links and thoughts:

All-In Podcast “E110: 2023 Bestie Predictions!”

Top 5 Tweets of the week:

Footnotes:

[1] http://aima.cs.berkeley.edu/

[2] https://www.pearson.com/store/p/artificial-intelligence-a-modern-approach/P100000291856/9780137505135 

What’s next for The Lindahl Letter? 

* Week 106: Code generating systems

* Week 107: Highly cited AI papers

* Week 108: Twitter as a company probably would not happen today

* Week 109: Robots in the house

* Week 110: Understanding knowledge graphs

If you enjoyed this content, then please take a moment and share it with a friend. If you are new to The Lindahl Letter, then please consider subscribing. New editions arrive every Friday. Thank you and enjoy the week ahead.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.nelsx.com

My appreciation for you reading this sentence right now is very real. That holds true in the moment of writing and within my intentions moving forward. At the start of this project a bit of writing on Substack happened two years ago now. During the course of starting that effort, I had not considered the amount of effort sustaining it would require. Probably the single most sustained writing project I have ever completed was my doctoral dissertation [1]. That writing process took less time than producing the 104 posts that are currently a part of The Lindahl Letter. I know some future writing projects could involve more effort, but that seems unlikely given the amount of time that was devoted to this one.  

You can take a moment and reflect on the realization that we have reached the point in the program where 104 weeks of content creation has occurred. Over time the writing process ended up following a planful approach to getting things done. The structure ended up being a block of writing, links/thoughts, top five tweets, and footnotes.  

* Weekly topic coverage. This is the heart of what is happening. Five weeks in planning or review is what makes the magic happen. For the most part, I ended up with a backlog with a bunch of planned posts. You should always keep a writing backlog of topics that deserve attention but might not have been advanced before. This method of keeping a writing topic backlog let me work on a couple of different weeks of content at any one time. Believe it or not, this is really important to keep things moving along. Momentum in writing is a real and present element of the creative process. If you run into a bit of writer’s block, or the inspiration is not showing up for a given topic, then you can move ahead and shift to a topic that allows the creation to continue. I’m not one to sit back and brew some word tea, but switching topics and trying to mix things up with a bit of stream of consciousness is certainly an option.  

* Links and thoughts. During the course of the week, I end up selecting some links and from research. Depending on where I’m at during the planning and review process, these links are either very timely or could have been pulled a week or two before. 

* Top five tweets. This section is really just for my own amusement. Based on the data from the last two years, nobody ever clicks on any of these tweets. I’m getting so close to having published 10,000 tweets. It will probably happen toward the end in the next couple of months.  

* Footnotes. Most writers in the academic space are comfortable with footnotes. Sometimes it takes people a bit of reading to get used to the links, references, and footnotes that I drop into the things being published. It is something that I plan on continuing with to help provide the context that we stand on the shoulders of giants in terms of our intellectual library of thoughts and considerations. The academy lives based on the interaction of scholars. That is something that cannot be forgotten or set aside to write prose without acknowledgement of the contributions of others.  

* Podcast. Adding the podcast element changed my posture to working to stay a week or two ahead. A part of that change was that I had to now plan ahead for vacation windows. That five-week planning and review cycle was key to keeping the publishing streak alive for what is approaching two years.  

After Post 104, I’m going to switch things up and focus more on sharing the content I find interesting in the AI and technology space [2]. Part of that is planned and was noted out in my five-year writing plan. In case you were wondering, I have included my five-year writing plan as of March 3, 2022: 

Year 1: Keep a heavy machine learning focus for the rest of 2022. Finish writing a collected series of machine learning/AI essays on Substack and combine them into a manuscript, The Lindahl Letter: 104 Machine Learning Posts. This manuscript should include both Years 1 and 2 of the Substack series. 

* Keep writing weekly Substack posts. 

* Take time for the manuscript generation process at the end of the year. 

* That manuscript will need to be edited by a professional before the print edition goes live. 

* Rework last year’s speaking engagement talks into academic papers. This could be one combined paper or potentially five different papers depending on how the initial effort shapes up. 

* “What Is Machine Learning Scale? The Where and the When of Machine Learning Usage” 

* “The Machine Learning Scale Problem: Thinking About Where and When to Use Machine Learning, ROI Models, Synthetic Data, Repeatable Frameworks, and Teams” 

* “Applied Machine Learning ROI: Understanding Machine Learning ROI From Different Approaches at Scale” 

* “Demystifying Applied Machine Learning: Building Frameworks and Teams to Operationalize Machine Learning at Scale” 

* “Figuring Out Applied Machine Learning: Building Frameworks and Teams to Operationalize Machine Learning at Scale, V3” 

* Rerun the MLOps GitHub research and turn that content into a paper. 

Year 2: For 2023, I want to pivot into studying sentiment analysis and modern polling methodologies. At this point, I will have written 104 essays on machine learning/AI and should probably refocus on a specific topic that is material to machine learning/AI, but adjacent to it as an area of research. It’s possible by 2023 that quantum computing will be a huge topic for research and will end up getting some attention as well. 

* Complete work on an automated sentiment analysis paper. 

* Write some sentiment analysis and machine learning essays for Substack. 

* Work on publishing modern polling methods essays for Substack. 

* Finish up writing the breakdown of modern polling paper. 

Year 3: 2024 will include a return to writing about local government administration and technology. It will be 20 years since earning my master of public administration degree. By this time, my writing should be as crisp and focused as it will ever be, and my perspective on technology will be well considered from my previous work on machine learning/AI. A few topics will be considered: 

* Technology and local government administration 

* The intersection of public administration and technology 

* How technology influences the practice of governing  

* How government uses machine learning/AI technology 

Year 4: 2025 will probably be the year where quantum computing has broken down modern encryption frameworks. A few topics will be covered here: 

* Changes and uses in encryption technology 

* Encryption and society 

* Quantum encryption 

Year 5: 2026 is going to be a year where my backlog should be highly full. The previous four years of this writing plan should have created a ton of leftover writing works. A few topics that I plan on highlighting include these: 

* A reflective work on machine learning/AL 

* Did open source MLOps technology survive? 

* Did the serverless trend pan out in the cloud? 

Pretty much everything that was a part of Year 1 of my five-year writing plan was worked on and completed. You can find a lot of that content contained in this 104-week writing project on machine learning. The future of writing things is wide open, and I’ll be starting work on Year 2 of that writing plan shortly. The last thing I’m about to work on will be moving the content from all 104 Substack posts over to a manuscript to send off to be edited.  

Footnotes: 

[1] Lindahl, N. P. (2010). Responsive e-government: A study of local government e-feedback methodology (Publication No. 3426428) [Doctoral dissertation, Walden University]. ProQuest Dissertations and Theses Global. https://www.proquest.com/openview/5fd815b75e8efa09a64814d0e290389b/1.pdf?pq-origsite=gscholar&cbl=18750  

[2] https://trends.google.com/trends/explore?date=today%205-y&geo=US&q=technology,artificial%20intelligence,machine%20learning,AI,ML 

What’s next for The Lindahl Letter? A totally new writing format… 

If you enjoyed this content, then please take a moment and share it with a friend. If you are new to The Lindahl Letter, then please consider subscribing. New editions arrive every Friday. Thank you and enjoy the year ahead.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.nelsx.com

You probably have gotten a sense of the tremendous and overwhelming flood of publishing that is happening in the machine learning space. So much content is being created right now that nobody could possibly consume it all. The flood of content is real and overwhelming. That is one of the reasons that I work really hard to distill the complex topics I select into a readable format for people to consume. I really try to provide a pathway to people who want to be a part of this journey. Maybe those two observations are inherently in conflict, but I think helping people navigate the deluge of information to focus on key things helps break down the conflict.  

As I sit down to rethink the future of machine learning, my thoughts are circling back to some of the original things I was writing about use cases and how people select what they want to work on based on ROI. Right now, machine learning is being built into everything, and we are seeing a creative explosion of people using machine learning to generate things on the fly that otherwise would have never been possible. Some of the models within the image- and video-creation spaces are really changing how we interact with the world. My honest guess about the future of machine learning is that it will become seamless within the background of our daily lives. It will be ever present and constantly just on the edge of how we perceive and interact with the world around us.  

I grabbed five papers where the future of machine learning was discussed from 2001 to 2018. 

Zhou, Z. H. (2016). Learnware: On the future of machine learning. Frontiers in Computer Science, 10(4), 589–590. https://www.lamda.nju.edu.cn/publication/fcs16learnware.pdf 

Mjolsness, E., & DeCoste, D. (2001). Machine learning for science: State of the art and future prospects. Science, 293(5537), 2051–2055. https://www.researchgate.net/profile/Eric-Mjolsness/publication/11789794_Machine_Learning_for_Science_State_of_the_Art_and_Future_Prospects/links/09e415147695a42e12000000/Machine-Learning-for-Science-State-of-the-Art-and-Future-Prospects.pdf 

Choy, G., Khalilzadeh, O., Michalski, M., Do, S., Samir, A. E., Pianykh, O. S., Geis, J. R., Pandharipande, P., Brink, J. A., & Dreyer, K. J. (2018). Current applications and future impact of machine learning in radiology. Radiology, 288(2), 318–328. https://doi.org/10.1148/radiol.2018171820  

Obermeyer, Z., & Emanuel, E. J. (2016). Predicting the future—Big data, machine learning, and clinical medicine. The New England Journal of Medicine, 375(13), 1216–1219. https://doi.org/10.1056/NEJMp1606181  

Handelman, G. S., Kok, H. K., Chandra, R. V., Razavi, A. H., Lee, M. J., & Asadi, H. (2018). eDoctor: Machine learning and the future of medicine. Journal of Internal Medicine, 284(6), 603–619. https://doi.org/10.1111/joim.12822  

What’s next for The Lindahl Letter?

* Week 104: That 2nd year of posting recap

I’ll try to keep the what’s next list forward looking with at least five weeks of posts in planning or review. If you enjoyed this content, then please take a moment and share it with a friend. If you are new to The Lindahl Letter, then please consider subscribing. New editions arrive every Friday. Thank you and enjoy the week ahead.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.nelsx.com

You end up with people who are working in a field and people who study the field of inquiry academically. Sometimes and especially within the machine learning space you end up with people who are actively working as a practitioner. Those very same practitioners of the craft of machine learning are publishing academic articles at a rate never before seen within any field of study at large within the academy. The best way to describe that effort would be to call them the pracademics of the machine learning space. Intellectually it’s probably good to have people write about things who really understand how they are occurring. You certainly get solid firsthand accounts of what people are creating.  

Research is the process of digging into things, investigating, studying, or maybe just seeking to understand the things better. Original research is often brought out to describe novel inquiry. Some of these pracademics are pushing things into a new frontier for machine learning. The DALL-E-2 AI system was introduced and changed the way people think about how an AI system would create realistic images and art based on a prompt [1]. 

During the course of working on this Substack post, you can imagine I was surprised to find a webpage called “AI Brain Drain” [2]. That site opens up with a clear, “Welcome to the AI Brain Drain Index.” They pretty much are tracking the number of AI faculty that have left academic areas to go work within industry. You can see charts and other figures that sort of run from 2004 to apparently 2018. The researchers must have stopped making charts in the last few years, but the ones they made are pretty nice. All of that research seems to have yielded a paper you can read [3]. According to the SSRN website where I downloaded the paper, 928 downloads have occurred on this one.  

Gofman, M., & Zhao, J. (2022, July 31). Artificial intelligence, education, and entrepreneurship. Journal of Finance [Forthcoming]. https://doi.org/10.2139/ssrn.3449440 or at SSRN: https://ssrn.com/abstract=3449440  

An article from Inside Higher Education called “AI Academy Under Siege” from author Oren Etzioni took a look at AI experts leaving institutions of higher education and what might be some potential solutions to that situation [4]. You could find more from the work of Professor Michael Gofman [5]. You could check out an editorial published in Springer by Lars Kunze called, “Can We Stop the Academic AI Brain Drain?” [6]. Outside of those sources, one of the articles that I really liked was from Ben Dickson titled, “What Is the AI Brain Drain?” [7]. You could pivot to an article that I enjoyed less called, “Brain Drain of AI Researchers: Academia vs Industry” [8].  

Let’s close this one out with an interesting look at, “AI Brain Drain to Google and Pals Threatens Public Sector’s Ability to Moderate Machine-Learning Bias” [9]. That article links back out to a paper that was interesting. 

Jurowetzki, R., Hain, D. S., Mateos-Garcia, J., & Stathoulopoulos, K. (2021). The privatization of AI research(-ers): Causes and potential consequences. https://regmedia.co.uk/2021/02/04/drain.pdf  

Top 5 Tweets of the week:

Footnotes:

[1] https://openai.com/dall-e-2/ 

[2] http://www.aibraindrain.org/

[3] https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3449440

[4] https://www.insidehighered.com/views/2019/11/20/how-stop-brain-drain-artificial-intelligence-experts-out-academia-opinion

[5] https://simon.rochester.edu/blog/deans-corner/brain-drain

[6] https://link.springer.com/article/10.1007/s13218-019-00577-2 

[7] https://bdtechtalks.com/2019/09/26/artificial-intelligence-brain-drain/ 

[8] https://medium.com/codex/brain-drain-of-ai-researchers-academia-vs-industry-8e385e8fd517

[9] https://www.theregister.com/2021/02/04/ai_brain_drain/

What’s next for The Lindahl Letter?

* Week 103: Rethinking the future of ML

* Week 104: That 2nd year of posting recap

I’ll try to keep the what’s next list forward looking with at least five weeks of posts in planning or review. If you enjoyed this content, then please take a moment and share it with a friend. If you are new to The Lindahl Letter, then please consider subscribing. New editions arrive every Friday. Thank you and enjoy the week ahead.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.nelsx.com

One of the core topics within the machine learning space that I have studied happens to be ROI or, more to the point, an examination of just how well spent money would be in the space. You can certainly spend money on machine learning and AI related efforts as a part of a think tank, an independent lab, or a pure research institution. Within corporate spaces, pure research and development is one thing, but generally to expend the precious capital resources of an institution, organization, or group, you want to know that some type of return will be accumulated from those activities. It’s inherent to the nature of the venture into running a corporation compared to running some other type of organization. 

ROI for machine learning is a topic that deserves consideration. You can evaluate a variety of potential machine learning use cases to solve problems. Going from being a special product to an operationalized business process that utilizes technology to get things done will be dependent on the ROI associated with the use case. If it costs more to do it, then you could reasonably expect that use case is going to make most responsible actors pump the brakes on the project. However, it appears that a lot of use cases went forward anyway.  

I spent some time trying to find a good accounting of how much money has been spent on machine learning projects overall and how many of them have actually yielded solid ROI. You probably will not be all that surprised to learn that very few rigorous studies exist of successful ROI in the machine learning space. If some of those types of studies exist and I just missed them during my search, then by all means feel free to share them in the comments and let me know [1]. It won’t hurt my feelings or anything, and I’d actually be a little bit relieved that research on the subject exists.  

Before we conclude here, I do want to share one paper that does seem to be directly addressing this question. It has only been cited by three other papers since 2020. I had hoped it would lead me to a cluster of academic research. That was not the case. The search for solid academic research on machine learning ROI is still ongoing.  

Mizgajski, J., Szymczak, A., Morzy, M., Augustyniak, Ł., Szymański, P., & Żelasko, P. (2020). Return on investment in machine learning: Crossing the chasm between academia and business. Foundations of Computing and Decision Sciences, 45(4), 281–304. https://sciendo.com/article/10.2478/fcds-2020-0015 

Links and thoughts:

“We Talked To A VP At Microsoft - WAN Show December 23, 2022”

Top 5 Tweets of the week:

Footnotes:

[1] https://scholar.google.com/scholar?hl=en&as_sdt=0%2C6&q=%22return+on+investment%22+%22machine+learning%22&btnG=

What’s next for The Lindahl Letter?

* Week 102: ML pracademics

* Week 103: Rethinking the future of ML

* Week 104: That 2nd year of posting recap

I’ll try to keep the what’s next list forward looking with at least five weeks of posts in planning or review. If you enjoyed this content, then please take a moment and share it with a friend. If you are new to The Lindahl Letter, then please consider subscribing. New editions arrive every Friday. Thank you and enjoy the week ahead.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.nelsx.com

You might remember back to Week 75 when I tried to explain the nature of overcrowding within the machine learning space and what exactly that is doing to engineering colleges. Back in 2020, Igor Susmelj wrote a piece in Towards Data Science, “How to Keep up With the Latest Research and Trends in ML” [1]. Within that paper, Igor raised awareness of a chart from Jeffrey Dean that showed machine learning arXiv papers versus Moore’s law growth rate [2]. A quick Google Scholar search of that paper will show that it is cited by 72 papers at the moment [3]. I had hoped that a batch of papers would help me find more academic work about overcrowding and machine learning, but that did not really turn out to be the case. I have been trying to figure out the right words to describe intellectual overcrowding and machine learning, but for the most part I have not been able to figure out the secret decoder ring settings to locate a bunch of papers on the subject.  

I noticed a chart on page 10 of a paper called, “An Overview on Applications of Machine Learning in Petroleum Engineering,” which showed the number of publications with AI/machine learning within the SPE OnePetro digital library [4]. Just like the chart from Jeffrey Dean, it showed a clear takeoff point after 2010, where the terms just skyrocket like a hockey stick. That is probably a fairly consistent trend, and things will shake out over time to a key set of academic articles that get referenced a lot and a core set of topics that are covered within the AI/machine learning academic community. At the moment, however, we are at the peak of the inflection point, where intellectually everybody rushed to be first in the pool and kept on swimming.  

With a rush of academic focus in the area of machine learning, we will see both a great deal of progress and a potentially calamitous fallout from intellectual overcrowding. Only so much progress and ultimately only so many faculty positions are going to exist within engineering programs. I do think we will see a pretty significant oversupply in the number of people seeking those faculty positions at the more prestigious set of academic institutions. Right now it is a lot more lucrative for people who are able to get jobs within industry to do that and to enjoy some pretty solid compensation. That is probably the element that has made the overcrowding less problematic during the rise of machine learning implementations. People being able to work in the private section and people working within academic spaces have been able to find work. You can find a lot of articles about brain drain within academic institutions related to both AI and machine learning. Some of those pieces make some interesting arguments. A lot of private organizations have run labs that are doing things of a more academic nature than perhaps applied use case development.  

Links and thoughts:

Top 4 Tweets of the week:

Footnotes:

[1] https://towardsdatascience.com/how-to-keep-up-with-the-latest-research-and-trends-in-ml-a45a356b1001 

[2] https://arxiv.org/ftp/arxiv/papers/1911/1911.05289.pdf 

[3] https://scholar.google.com/scholar?cites=3848695121612936760&as_sdt=4005&sciodt=0,6&hl=en 

[4] https://www.researchgate.net/publication/339952951_An_Overview_on_Applications_of_Machine_learning_in_petroleum_Engineering 

What’s next for The Lindahl Letter?

* Week 101: Back to the ROI for ML 

* Week 102: ML pracademics

* Week 103: Rethinking the future of ML

* Week 104: That 2nd year of posting recap

I’ll try to keep the what’s next list forward looking with at least five weeks of posts in planning or review. If you enjoyed this content, then please take a moment and share it with a friend. If you are new to The Lindahl Letter, then please consider subscribing. New editions arrive every Friday. Thank you and enjoy the week ahead.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.nelsx.com