Afternoon with Tomasso on the limitations of Artificial intelligence’s application

For us to be able to successfully apply artificial intelligence on any domain, the following needs to be true

  • The behavior the system to be modeled must not be stochastic
  • The state of the system must be decipherable by the data scientist
    • it should be possible to understand the state in which the system is at through interpretation of data gathered
  • The domain can be modeled
    • the parameters for modeling the domain must be well defined

Only when all three premise are true can we determine where the adjustment should be made when a model fails to predict an outcome

The financial markets is stochastic  in the short run.

The underlying parameters are constantly changing and thus hard to model due to the emergent nature of impacts caused by human activities. The data is qualitative and thus hard to convert into clean quantitative datasets.

While the price movements are obvious it is hard, it is hard to attribute impact to the various parameters.

As such, it requires human neural networks that consumed all these qualitative data to perform the prediction/decision making.

The case for hastening the replacement of workers with AI

If the issue of aging population is an inevitable affliction of all industrialized countries and majority of countries will become industrialized within the next 30 years, then we should be expecting our population to collapse by 2050. Based on this premise rather than being worried that majority of workers will get replaced by Robots and made irrelevant, we should instead be worried that robots are not replacing tasks handled by forthcoming retirees fast enough,

Related References

https://www.bloomberg.com/amp/news/articles/2019-07-24/u-s-truck-driver-shortage-is-on-course-to-double-in-a-decade

https://amp.businessinsider.com/elon-musk-reiterates-global-population-is-headed-for-collapse-2019-6

 

Book summary AI super-powers, China, silicon valley and the new world order by Kai Fu Lee

The difference waves of AI

  • Internet AI – Facebook, Netflix, Google search
  • Business AI – Palantir
  • Perception AI – Tesla cars
  • Autonomous AI – Tesla self driving, Google self driving

Key locations

  • Silicon Valley
  • Zhong Guan Cun – Beijing

State of the Union

  • We are in the stage of implementation/application as opposed to RnD
    • having access to more data is more important than have expertise to do more RnD
    • having solid AI engineers is more important than AI researchers
  • We are still far from general AI
  • Key ingredients
    • data
    • computing
    • maybe work of strong AI algorithms engineers

Key differences between eco-systems

  • Silicon Valley businesses are mission and core values driven while Chinese businesses are pragmatically focused on profitability.
  • Silicon Valley businesses stay in bits and binaries offloading the brick and mortar to external vendors vendors while Chinese businesses extend their business model into the brick and mortar (online to offline)
  • Silicon valley prefers one size fit all strategy, Chinese businesses utilized localized solutions often investing/acquiring in local startups
  • Americans treat search engines like Yellow Pages (come and leave fast) while Chinese treat search engines like shopping mall (come to linger around long)
  • Silicon Valley is adversed to copying preferring to be unique Chinese business copy the heck out of each other

Chinese Advantage

  • Abundant data – quality and quantity aided by their online to offline initiatives
  • hungry entrepreneurs
  • AI scientist
  • AI friendly policy environment – strong emphasis by Chinese government
  • Hardware manufacturing know how – Shen Zhen
    • unparalleled supply chain flexibility – XiaoMi

Silicon Valley Advantage

  • Microchip manufacturing know-how

Trends within the Chinese eco-system

  • Darwinian eco-system has lead to extreme levels of competition
  • Chinese companies have already moved past the stage of clone Silicon Valley business models
  • Businesses innovate to build a defensive moat around themselves. Local businesses have advantage, with no timezone differences to deal with, decision making is relatively faster.
  • Online to offline
    • an essential ingredient to building strategic moats
    • caused the decline of cash use
  • Chinese government information systems will be able to leap frog US government information systems

Policy approaches

  • Google – impeccable safety
  • Tesla / China – trial by fire
  • key to winning the Autonomous AI race
    • is the bottleneck technology (Silicon Valley) or policy (China)?

Key concerns

  • having cheap labor is no longer going to be a source of advantage in a world heavily powered by automaton.  Developing countries hoping to employ this well tested strategy to progress will not be able to do so anymore
  • Estimated 60% potential job loss worldwide barring policy interventions
  • Job loss probability assessment
    • physical labor
      • environment – unstructured versus structure
      • tasks nature – level of dexterity versus high dexterity
    • cognitive labor
      • social – high versus low
      • cognitive – optimization based versus creativity/strategy based
  • AI replacement approach
    • single tasks approach
    • ground up rethink re-imagination
  • A population of irrelevant (no longer employable) as opposed to unemployed

Tackling Key concerns

  • Silicon valley – reduce, retrain and redistribute
  • Kai Fu Lee – stipends for care, service, education

New promise

  • Humans freed up from repetitive tasks can now focus on becoming more human oriented

Related readings

  • Disruptor, Zhou
  • www.Arvix.org – an online repository of scientific papers
  • Folding Beijing – Hao JingFang

Mark Zuckerberg chats with Yuval Noah Harrai on the Future of AI

Key take aways

  • Spread of inequality where some countries have the ability to harness AI while others don’t
  • AI based recommendation systems moving from being just an oracle to becoming a sovereign
  • AI as a tool is an amplifier
    • concerns that it will benefit totalitarianism more than democracy leading to totalitarianism becoming a more favorable governance model worldwide
    • surveillance
    • psychological manipulation – the inability to know your true self through your thoughts
    • what happens if morality and expediency diverge when it comes to governance
  • Effectiveness of curbing the negative effects of AI by encoding values within policy frameworks governing these AI based systems
    • Companies based in Democratic countries will encode democratic values within their systems vice versa for Totalitarian countries
  • Personalization versus Fragmentation
    • when everyone in a country chooses his own community that is mainly online there is no longer a glue holding the local community together
  • Long term versus short term
    • The long term benefits might come sooner than expected when taking a short term trade off

 

Book summary: Everybody lies by Seth Stephens-Davidowitz

Signals from Search

  • What people search for is in itself a signal
  • The order of keywords in which they search is also a signal
  • Quality of google search data is better than in Facebook because
    • You are alone with no fear of being judged
    • You have an incentive to be honest

On Big data

  • the needle is still the same size but the haystack has been getting bigger
  • Be judicious by cutting down the sample size of the data to be used

Data science

  • Trust your intuition as the initial signal but verify quantitatively to avoid narrative bias
  • Correlation is most often sufficient for utilization purposes – often the explanation of why the model works comes after the fact
  • critically assess the actual data underlying the narrative. At times it might tell a very different story than narrative presented
  • Clustering of groups of people helps predicts behavior – Netflix and baseball
  • AB testing to discover causations

Social Impact

  • great business are found on:
    • secrets about nature
    • secrets about people
  • Modeling
    • Physics – utilize neat equation
    • Human behavior – probabilistically via Naive Bayes classification

Managing angry people

  • Lecturing them will provoke their anger
  • Provoking their curiousity will cause their attention to be diverted causing anger to subside

Related readings

  • Zero to one, Peter Thiel

Book summary – The Signal and the noise

Risk versus uncertainty

  • Risk can be mathematically modeled to yield a probability
  • uncertainty cannot be mathematically modeled

Conditions for quality data

Why google’s Search data is better than Facebook profile data

  • subject feels she has privacy privacy
  • subject feels she is not judged
  • subject sees tangible benefit from being honest

The hedgehog versus the fox

  • The hedgehog approaches reality through a narrative/ideology while the fox thinks in terms of probabilities
  • The hedgehog goes very deep in an area while the fox employs multiple different models
  • The fox is a better forecaster than the hedgehog
  • The fox is more tolerant of uncertainty

Big data

  • More data does not yield better results and predictions
  • Deciding the right kind of data from the abundance available
  • To do prediction it is important to start from intuition and to keep model simple
  • qualitative data should be weighted and considered
  • Be self aware of your own biases

Prediction

  • Similarity scores – clustering in Netflix and baseball
  • Be wary of confirmation biases
  • Be wary of overfitting using small sample size – Tokyo earthquakes and global warming
  • Correlation does not equal causation
  • short hand heuristics to reduce the computational space – for example chess

Related references

  • Irrational exuberance, Robert Shiller
  • Expert political judgement, Philip E. Tetlock
  • Future shock, Alvin and Heidi Toffler
  • Principles of forecasting, J Scott Armstrong
  • Predicting the unpredictable, Hough

Insights from Klaren’s birthday

Conversations with Yi (EverString)

The forthcoming trend for engineering

Machine learning is increasingly becoming commoditized. DevOps becomes more important. Demand for specialized service where DevOps is encapsulated will further increase as demand for engineering tasks further outstrips engineering supplies.

On lead generation market

Companies in the lead generation space have need for scalable web crawlers. This helps offset the cost of retaining three in-house engineers.

Lead generation space has consolidated. There were priorly 120k such companies. There is 7k companies in operation. Majority of players are generating leads by scraping LinkedIn.

Consumer space require constant development of new features. Enterprise space requires service heavy. Enterprise space requires not just lead generation but entire channel marketing service suit (physical mail, online advertising, email marketing)

Lead gen hard to retain. The list becomes less valuable once it’s been used. 80% yearly churn is normal. One company reduces yearly churn to just 10% this by reducing second year subscription from USD800/yr to USD200/yr. further discount to USD100/yr if they don’t like. Recurring service is for grabbing fresh leads from same data source.

On Tele conference

Zoom’s product team compared with UberConference has developed a better understanding of the true conference needs of their users in various context. They have worked harder to ensure their product work seamlessly in identified scenarios. A typical example is the ability to join s conference bybthe press of a button on their mobile phone while driving instead of having to type the typical 4 pin digits.

Insights from the week

From Connie (Edmodo)

  • the key to consulting is to organize data into high level mutually exclusive buckets to allow easy defeating by decision makers

From Tim (Edmodo)

  • Kano model

From Val (Totango)

  • Company is concerned with increasing revenue and profitability. This will drive higher valuation during further exit

From Yip (ATT)

  • Analytics from Facebook page comments and twitter hashtag
  • need to balance customer support demand and cost of running department:
    • customer support hotline
    • Direct comments from influencers  which trigger negative sentiment to support staff
  • Business analyst reads comments manually to get qualitative needs and understands business needs
  • Data scientist explores data might not know the business needs
  • business analyst have problems working with data scientist
  • tools to help business analyst get directly at the insight instead of via data scientist
  • build model to predict call support volume by category
  • build model to quantify feature demand level needs
  • correlation of weather and commodity prices

Highlights from The Future of Humanity by Michio Kaku

  • Organisms on earth eventually will meet one of three fates, leave, adapt or die. Earth has already sustained 5 extinction cycles.
  • The threats we face are largely self inflicted
  • Scientific revolution comes in waves often stimulated by advances in physics
    • 19th century
      • mechanics and thermodynamics: locomotive and industrial revolution
    • 20th century
      • electricity and magnetism bring forth the electric age
    • The forthcoming wave
      • nano-technology
      • AI, neural networks
      • Quantum computing
      • CRISPR revolution
      • Transhumanism – the need to deal with ethical questions
  • Technological regression occurs when the population becomes complacent,
    • Admiral Zheng He and his fleet under subsequent rulers
    • US space program after the cold war is over
  • Interesting phenomena worth exploring
    • wormholes
    • rogue planets – planets that do not orbit any particular stars
    • caloric restriction and increased life expectancy
    • falling birthrates and education of women
    • uploading and downloading of consciousness (Transcendence and Mnemonic Johnny)
    • achievement of super strength on new planets
    • artificial enhancement of body, seamless interfacing with machines (telekinesis)
    • big bang happening over and over again and the universe does not grow only in one direction
  • Civilization categorization method 1:
    • energy based
      • Type 1: utilizes all the energy of the sunlight falling on the planet
      • Type 2: utilizes all energy its sun produces
      • Type 3: utilizes energy of an entire galaxy
      • Type 4: utilizing energy beyond the galaxy
    • information consumption based

Navigating the trough of sorrow

While I was reading through most of the success stories that were published on IndieHackers.com, it occurred to me that my project GetData.IO really took longer than most others to gain significant traction, a full 5 years actually.

The beginning

I first stumbled upon this project back in December 2012 when I was trying to solve two other problems of my own.

In my first problem, I was trying to identify the best stocks to buy on the Singapore Stock Exchange. While browsing through the stocks listed on their website, I soon realize that most stock exchanges as well as other financial websites gear their data presentation towards quick buy and sell behaviors. If you were looking to get data for granular analysis based on historical company performance as opposed to stock price movements, its like pulling teeth. Even then, important financial data I needed for decision making purposes were spread across multiple websites. This first problem lead me to write 2 web-scrappers, one for SGX.com and the other for Yahoo Finance, to extract data-sets which I later combined to help me with my investment decision-making process.

Once I happily parked my cash, I went back to working on my side project then. It was a travel portal which aggregates all the travel packages from tour agencies located in Southeast Asia. It was not long before I encountered my second problem… I had to write a bunch of web-scrapers again to pull data from vendor sites which do not have the APIs! Being forced to write my 3rd, 4th and maybe 5th web-scraper within a single week lead me to put on hold all work and step back to look at the bigger picture.

The insight

Being a web developer, and understanding how other web developers think, it quickly occurred to me the patterns that repeat themselves across webpage listings as well as nested webpages. This is especially true for naming conventions when it came to CSS styling. Developers tend to name their CSS classes the way they would actual physical objects in the world.

I figured if there existed a Semantic Query Language that is program independent, it would provide the benefit of querying webpages as if they were database tables while providing for clean abstraction of schema from the underlying technology. These two insights still prove true today after 6 years into the project.

The trough of sorrow

While the first 5 years depicted in the trend line above seem peaceful due to a lack of activity, it felt anything but peaceful. During this time, I was privately struggling with a bunch of challenges.

Team management mistakes and pre-mature scaling

First and foremost was team management. During the inception of the project my ex-schoolmate from years ago approached me to ask if there was any project that he could get involved in. Since I was working on this project, it was a natural that I would invited him to join the project. We soon got ourselves into an incubator in Singapore called JFDI.

In hindsight, while the experience provided us with general knowledge and friends, it really felt like going through a whirlwind. The most important piece of knowledge I came across during the incubation period was this book recommendation?—?The Founder’s dilemma. I wished I read the book before I made all of the mistakes I did.

There was a lot of hype (see the blip in mid-2013), tension and stress during the period between me and my ex-schoolmate. We went our separate ways due to differences in vision of how the project should proceed shortly after JDFI Demo Day. It was not long before I grew the team to a size of 6 and had it disbanded, realizing it was naive to scale in size before figuring out the monetization model.

Investor management mistakes

During this period of time, I also managed to commit a bunch of grave mistakes which I vow never to repeat again.

Mistake #1 was being too liberal with the stock allocation. When we incorporated the company, I was naive to believe the team would stay intact in its then configuration all the way through to the end. The cliff before vesting were to begin was only 3 months with full vesting occurring in 2 years. When my ex-schoolmate departed, the cap table was in a total mess with a huge chunk owned by a non-operator and none left for future employees without significant dilution of existing folks. This was the first serious red-flag when it came to fund raising.

Mistake #2 was giving away too much of the company for too little, too early in the project before achieving critical milestones. This was the second serious red-flag that really turned off follow up would-be investors.

Mistake #3 was not realizing the mindset difference of investors in Asia versus Silicon Valley, and thereafter picking the wrong geographical location (a.k.a network) to incubate the project. Incubating the project in the wrong network can be really detrimental to its future growth. Asian investors are inclined towards investing in applications that have a clear path to monetization while Silicon Valley investors are open towards investing in deep technology of which the path to monetization is yet apparent. During the subsequent period, I saw two similar projects incubated and successfully launched via Ycombinator.

The way I managed to fix the three problems above was to acquire funds I didn’t yet have by taking up a day job while relocating the project to back to the Valley’s network. I count my blessings for having friends who lend a helping hand when I was in a crunch.

Self-doubt

I remembered having the conversation with the head of the incubator two years into the project during my visit back to Singapore when he tried to convince me the project was going nowhere and I should just throw in the towel. I managed to convince him and more importantly myself to give it go for another 6 months till the end of the year.

I remember the evenings and weekends alone in my room while not working on my day job. In between spurts of coding, I would browse through the web or sit staring at the wall trying to envision how product market fit would look like. As what Steve Jobs mentioned once in his lecture, it felt like pushing against a wall with no signs of progress or movement whatever so. If anything, it was a lot of frustration, self-doubt and dejection. A few times, I felt like throwing in the towel and just giving up. For a period of 6 months in 2014, I actually stopped touching the code in total exasperation and just left the project running on auto-pilot, swearing to never look at it again.

The hiatus was not to last long though. A calling is just like the siren, even if somewhat faint sometimes, it calls out to you in the depths of night or when just strolling along on the serene beaches of California. It was not long before I was back on my MacBook plowing through the project again with renewed vigor.

First signs of life

It was mid-2015, the project was still not showing signs of any form of traction. I had by then stockpiled some cash from my day job and was starting to get interested in acquiring a piece of real estate with the hope of generating some cashflow to bootstrap the project while freeing up my own time. It was during this period of time that I got introduced to my friend’s room mate who also happened to be interested in real estate.

We started meeting on weekends and utilizing GetData.IO to gather real estate data for our real estate investment purposes. We were gonna perform machine learning for real estate. The scope of the project was really demanding. It was during this period of dog fooding that I started understanding how users would use GetData.IO. It was also then when I realized how shitty and unsuited the infrastructure was for the kind and scale of data harvesting required for projects like ours. It catalyzed a full rewrite of the infrastructure over the course of the next two years as well as brought the semantic query language to maturity.

Technical challenges

Similar to what Max Levchin mentioned in the book Founder’s at work, during this period of time there was always this fear in the back of my mind that I would encounter technical challenges which would be unsolvable.

The site would occasionally go down as we started scaling the volume of daily crawls. I would spend hours on the weekends digging through the logs to attempt at reproducing the error so as to understand the root cause. The operations was like a (data) pipeline, scaling one section of the pipeline without addressing further down sections would inevitably cause fissures and breakage. Some form of manual calculus in the head would always need to be performed to figure out the best configuration to balance the volume and the costs.

The number 1 hardest problem I had to tackle during this period of time was the problem of caching and storage. As the volume of data increase, storage cost increase and so did wait time required before data could be downloaded. This problem brought down the central database a few times.

After procrastinating for a while as the problem festered in mid-2016, I decided that it was to be the number 1 priority to be solved. I spend a good 4 months going to big-data and artificial intelligence MeetUps in the Bay Area to check out the types of solutions available for the problem faced. While no suitable solutions were found, the 4 months helped elicit corner cases to the problem which I did not previously thought of. I ended up building my own in-house solution.

Traction and Growth

An unforeseen side effect of solving the storage and caching problem was its effect on SEO. The effects on SEO would not be visible until mid-2017 when I started seeing increased volume of organic traffic to the site. As load times got reduced from more than a minute in some cases to less than 400 milliseconds seconds, the volume of pages indexed by bots would increase, accompanied by increase in volume of visitors and reduction in bounce rates.

Continued education

It was in early-2016 that I came across an article expounding the benefits of reading widely and deeply by Paul Graham which prompted me to pick up my hobby of reading again. A self-hack demonstrated to me by the same friend, who helped relocated me here to the Bay Area, which I pursued vehemently got me reading up to 1.5 books a week. These are books which I summarized on my personal blog for later reference. All the learnings developed my mental model of the world and greatly aided in the way I tackled the project.

Edmodo’s VP of engineering hammered in the importance of not boiling the ocean when attempting to solve a technical problem, of always being judicious with the use of resource during my time working as a tech-lead under his wing.  Another key lesson learned from him is that in some circumstances being liked and being effective do not go hand in hand. As the key decision maker, it is important to steadfastly practice the discipline of being effective.

Head of Design, Tim and Lukas helped me appreciate the significance of UX during my time working with them and how it ties to user psychology.

Edmodo’s CEO introduced us to mindfulness meditation late-2016 to help us weather through the turbulent times that was happening within the company then. It was rough. The practice which I have adopted till to date has helped keep my mind balance while navigating the uncertainties of the path I am treading.

Edmodo’s VP of product sent me for a course late-2017 which helped consolidate all the knowledge I have acquired till then into a coherent whole. The knowledge gained has helped greatly accelerated the progress of GetData.IO. During the same period, I was also introduced by him the Vipasanna mediation practice which coincidentally a large percentage of the management team practices.

One very significant paradigm shift I observed in myself during this period of continued education is the observed relationship between myself and the project. It has changed from an attitude of urgently needing to succeed at all cost to an attitude of open curiosity and fascination as one would an open ended science project.

Moving forward

To date, I have started working full time on the project again. GetData.IO has the support of more than 1,500 community members worldwide. Our mission is to turn the Web into the fully functional Giant Graph Database of Human Knowledge. Financially, with the help of our community members, the project is now self-sustaining. I feel grateful for all the support and lessons gained during this 6 year journey. I look forward to the journey ahead as I continue along my path.