GOOG AI directs me to interview with Ari Ben-Menashe on Jeffrey Epstein

People talk about a future cybernetic era in which human intelligence will be fused in some way with machine intelligence (AI). To a degree, that era has already arrived. The GOOG AI watches almost everything I do -- not just my search queries, but pages I access via Chrome, seminars and interviews I watch on YouTube, my meetings on Google Calendar, what topics I discuss over gmail, where I travel, etc. I can now depend on it to make useful recommendations. (I hope the AI remains friendly to me in the future...)

This morning it suggested the interview below with Ari Ben-Menashe. Probably because it knows I have been interested in Jeffrey Epstein (see post Epstein and the Big Lie from Aug 2019), the activities of intelligence services (see, e.g., Twilight Struggles in a Wilderness of Mirrors: Admiral Mike Rogers, the NSA, and Obama-era Political Spying), and also nuclear weapons history.

Ben-Menashe was an Israeli intelligence operative, best known for his role in Iran-Contra in the 1980s. He was also one of the main sources for the book The Samson Option, by Sy Hersh (the journalist who uncovered both My Lai and Abu Ghraib). The Samson Option describes how the world became aware of the Israeli nuclear program, thanks to whistle-blower Mordechai Vanunu. After revealing the secret program to the British Sunday Times, Vanunu was kidnapped by Israeli intelligence agents, stood trial in Israel, and spent almost 20 years in prison. Ben-Menashe worked with publisher Robert Maxwell (Ghislaine Maxwell's father) to locate Vanunu in London and to capture him using a honey trap (female agent).

Ben-Menashe knew Jeffrey Epstein and Ghislaine Maxwell through Robert Maxwell. He states on the record that Epstein was involved in a honeypot operation for Israeli intelligence.

Ben-Menashe also comments on topics such as:

Epstein's "suicide" in MCC (where, by coincidence, Ben-Menashe was also held in the aftermath of Iran-Contra).

Ghislaine Maxwell's current location.

Robert Maxwell's mysterious death.

How Epstein could live and operate as if he had a 10-11 figure net worth when his actual wealth was one or two orders of magnitude less.

I do not know whether any of this is true, but I found the interview interesting.




Warning: in the comments I will censor anti-Jewish remarks.

COVID-19: CFR ~1% estimated in large random sample (Austria)

CV19 antigen test of a random representative sample in Austria. CFR (or IFR, to be very precise) is close to 1%.

WSJ: More twice as many people have been infected by the new coronavirus in Austria than official figures showed, according to a new survey, with a fatality rate of 0.77%.

The nationwide survey, which the Austrian government described as the first of a country with a sizable population, showed that lockdown measures, which are particularly strict in Austria, were necessary to avoid mass casualties and overwhelming the health-care system, said Heinz Fassmann, the country’s education minister, who presented the study in Vienna Friday.

The study, conducted by polling firm SORA in cooperation with the government the Red Cross, tested a random, representative sample of 1,544 people aged 0 to 94 from across the country in their homes or in drive-in testing stations. It indicated that 28,500 people, or around 0.33% of Austria’s 8.9 million population, were infected with the virus by April 6, sharply higher than the 12,467 infections recorded by that date, with 220 people dying of Covid-19, the disease the virus causes.

The findings suggest that while the death rate implied from the study, 0.77%, is lower than the World Health Organization’s estimate for reported cases, which is over 3%, it would still mean that the virus could kill many millions of people before a vaccine is available.

95% confidence interval for infection rate is 0.12% to 0.76%, so CFR range of ~0.3% to ~2%. The "standard model" of CV-19 epidemiology seems to be correct.

Note Added: First sign of Google / Apple action to bring the full power of geolocation to bear on contact tracing and isolation! These capabilities have been available in China for some time now.

Bloomberg: Apple Inc. and Google unveiled a rare partnership to add technology to their smartphone platforms that will alert users if they have come into contact with a person with Covid-19. People must opt in to the system, but it has the potential to monitor about a third of the world’s population.

The technology, known as contact-tracing, is designed to curb the spread of the novel coronavirus by telling users they should quarantine or isolate themselves after contact with an infected individual.

The Silicon Valley rivals said on Friday that they are building the technology into their iOS and Android operating systems in two steps. In mid-May, the companies will add the ability for iPhones and Android phones to wirelessly exchange anonymous information via apps run by public health authorities. The companies will also release frameworks for public health apps to manage the functionality.

This means that if a user tests positive for Covid-19, and adds that data to their public health app, users who they came into close proximity with over the previous several days will be notified of their contact. This period could be 14 days, but health agencies can set the time range.

The second step takes longer. In the coming months, both companies will add the technology directly into their operating systems so this contact-tracing software works without having to download an app. Users must opt in, but this approach means many more people can be included. Apple’s iOS and Google’s Android have about 3 billion users between them, over a third of the world’s population. ...

COVID-19: Exiting Lockdown and Geolocation

Pressure will mount around the end of this month (assuming we are past the peak death rate and virus spread is under control) for the US to exit lockdown. This needs to be done in a smart way, which includes:

1. Required use of facemasks
2. Cocooning of vulnerable populations
3. Contact tracing and forced isolation of cases, perhaps using geolocation technology

See related posts

COVID-19: Smart Technologies and Exit from Lockdown (Singapore)
COVID-19: CBA, CFR, Open Borders
COVID-19: Cocoon the vulnerable, save the economy?
COVID-19 Notes

WSJ: Western governments aiming to relax restrictions on movement are turning to unprecedented surveillance to track people infected with the new coronavirus and identify those with whom they have been in contact.

Governments in China, Singapore, Israel and South Korea that are already using such data credit the practice with helping slow the spread of the virus. The U.S. and European nations, which have often been more protective of citizens’ data than those countries, are now looking at a similar approach, using apps and cellphone data.

“I think that everything is gravitating towards proximity tracking,” said Chris Boos, a member of Pan-European Privacy-Preserving Proximity Tracing, a project that is working to create a shared system that could take uploads from apps in different countries. “If somebody gets sick, we know who could be infected, and instead of quarantining millions, we’re quarantining 10.” ...

Some European countries are going further, creating programs to help track individuals—with their permission—who have been exposed and must be quarantined. The Czech Republic and Iceland have introduced such programs and larger countries including the U.K., Germany and Spain are studying similar efforts. Hundreds of new location-tracking apps are being developed and pitched to those governments, Mr. Boos said.

U.S. authorities are able to glean data on broad population movements from the mobile-marketing industry, which has geographic data points on hundreds of millions of U.S. mobile devices, mainly taken from apps that users have installed on their phones.

Europe’s leap to collecting personal data marks a shift for the continent, where companies face more legal restrictions on what data they may collect. Authorities say they have found workarounds that don’t violate the European Union’s General Data Protection Regulation, or GDPR, which restricts how personal information can be shared. ...

Google, Apple, Facebook, etc. are reluctant to draw attention to their already formidable geolocation capabilites. But this crisis may focus public awareness on their ability to track almost all Americans throughout the day.

WSJ: Google will help public health officials use its vast storage of data to track people’s movements amid the coronavirus pandemic, in what the company called an effort to assist in “unprecedented times.”

The initiative, announced by the company late Thursday, uses a portion of the information that the search giant has collected on users, including through Google Maps, to create reports on the degree to which locales are abiding by social-distancing measures. The “mobility reports” will be posted publicly and show, for instance, whether particular localities, states or countries are seeing more or less people flow into shops, grocery stores, pharmacies and parks. ... 

This is just a hint at what Google is capable of. Check out Google Timeline! Of course, users have to opt in to create their Google Timeline. But it should be immediately obvious that Google already HAS the information necessary to populate a detailed geolocation history of every individual...




Added from the comments:

There are really two separate issues here:

1. What is the basic epidemiology of CV19? i.e., R0, CFR, age distribution of vulnerability, comorbidities, mechanism of spread, utility of masks, etc.

2. What is the cost benefit analysis for various strategies (e.g., lockdown vs permissive sweep with cocooning)

While we have not reached full convergence on #1 I think reasonable people agree that the "mainstream" consensus has a decent chance of being correct: e.g., CFR ~ 1% or so, possibility of wide sweep in any population, overload of ICUs means much higher CFR, warmer weather might not save the day, etc. Once this scenario for #1 has, say, >50% chance of being right you are forced to at least take it seriously and then you are on to #2. (It is not required to believe that the scenario above is true at 95% or 99% confidence level...)

#2 is a question of trade-offs and two reasonable people can easily disagree until the end of time... I've already posted very simple CBA that show the answer can go either way depending on how you "price" QALYs, what you think long term effects on economy are from lockdown -- i.e., how fragile you think financial, supply chain, psychological systems are in various places; is it a ~$trillion cost, or could it go nonlinear?

Re: Physicists (and addressing gmachine comment below which has a lot of truth in it), we have no trouble understanding modeling done by other people (whether in finance, climate, or epidemiology), and we are also trained to deal with very uncertain data / statistical situations. We can "take apart" the model in our head to see where the dependencies are and how the uncertainties propagate through the model. I am amazed often to meet people who built a very complex model (e.g., thousands of lines of code, lots of input parameters), but they lack the chops to develop good intuition for how their model works, to make qualitative estimates for uncertainty quantification, etc. I have seen this in economics, finance, biology, and climate contexts many times. "There are levels to this thing..." Understanding the model can be more g-loaded than building it!

Finally, we are trained to think from first principles -- which assumptions are crucial to reach the conclusions, which are not? What are the key uncertainties in the analysis? Do we really need very specific assumptions about, e.g., social interaction rates as in the Imperial models? Or can I do a quick Fermi estimate which gets me a more robust answer at the cost of a factor of 2 uncertainty that does not really affect the main conclusion -- e.g., will ICU overload happen?

Enrico Fermi at the Trinity test: "I tried to estimate its strength by dropping from about six feet small pieces of paper before, during, and after the passage of the blast wave. Since, at the time, there was no wind I could observe very distinctly and actually measure the displacement of the pieces of paper that were in the process of falling while the blast was passing. The shift was about 2 1/2 meters, which, at the time, I estimated to correspond to the blast that would be produced by ten thousand tons of T.N.T." The actual yield was about 20 kt. Sometimes a smart guy can get to within a factor of two, and with much greater clarity, than a huge team of modelers...

PanOpticon in my Pocket: 0.35GB/month of surveillance, no charge!

Your location is monitored roughly every 10 minutes, if not more often, thanks to your phone. There are multiple methods: GPS or wifi connections or cell-tower pings, or even Bluetooth. This data is stored forever and is available to certain people for analysis. Technically the data is anonymous, but it is easy to connect your geolocation data to your real world identity -- the data shows where you sleep at night (home address) and work during the day. It can be cross-referenced with cookies placed on your browser by ad networks, so your online activities (purchases, web browsing, social media) can be linked to your spatial-temporal movements.

Some quantities which can be easily calculated using this data: How many people visited a specific Toyota dealership last month? How many times did someone test drive a car? Who were those people who test drove a car? How many people stopped / started a typical 9-5 job commute pattern? (BLS only dreams of knowing this number.) What was the occupancy of a specific hotel or rental property last month? How many people were on the 1:30 PM flight from LAX to Laguardia last Friday? Who were they? ...

Of course, absolute numbers may be noisy, but diffs from month to month or year to year, with reasonable normalization / averaging, can yield insights at the micro, macro, and individual firm level.

If your quant team is not looking at this data, it should be ;-)

Google Data Collection
Professor Douglas C. Schmidt, Vanderbilt University
August 15, 2018

... Both Android and Chrome send data to Google even in the absence of any user interaction. Our experiments show that a dormant, stationary Android phone (with Chrome active in the background) communicated location information to Google 340 times during a 24-hour period, or at an average of 14 data communications per hour. In fact, location information constituted 35% of all the data samples sent to Google. In contrast, a similar experiment showed that on an iOS Apple device with Safari (where neither Android nor Chrome were used), Google could not collect any appreciable data (location or otherwise) in the absence of a user interaction with the device.

e. After a user starts interacting with an Android phone (e.g. moves around, visits webpages, uses apps), passive communications to Google server domains increase significantly, even in cases where the user did not use any prominent Google applications (i.e. no Google Search, no YouTube, no Gmail, and no Google Maps). This increase is driven largely by data activity from Google’s publisher and advertiser products (e.g. Google Analytics, DoubleClick, AdWords)11. Such data constituted 46% of all requests to Google servers from the Android phone. Google collected location at a 1.4x higher rate compared to the stationary phone experiment with no user interaction. Magnitude wise, Google’s servers communicated 11.6 MB of data per day (or 0.35 GB/month) with the Android device. This experiment suggests that even if a user does not interact with any key Google applications, Google is still able to collect considerable information through its advertiser and publisher products.

f. While using an iOS device, if a user decides to forgo the use of any Google product (i.e. no Android, no Chrome, no Google applications), and visits only non-Google webpages, the number of times data is communicated to Google servers still remains surprisingly high. This communication is driven purely by advertiser/publisher services. The number of times such Google services are called from an iOS device is similar to an Android device. In this experiment, the total magnitude of data communicated to Google servers from an iOS device is found to be approximately half of that from the Android device.

g. Advertising identifiers (which are purportedly “user anonymous” and collect activity data on apps and 3rd-party webpage visits) can get connected with a user’s Google identity. This happens via passing of device-level identification information to Google servers by an Android device. Likewise, the DoubleClick cookie ID (which tracks a user’s activity on the 3rd-party webpages) is another purportedly “user anonymous” identifier that Google can connect to a user’s Google Account if a user accesses a Google application in the same browser in which a 3rd-party webpage was previously accessed. Overall, our findings indicate that Google has the ability to connect the anonymous data collected through passive means with the personal information of the user.

Google Duplex and the (short) Turing Test

Click this link and listen to the brief conversation. No cheating! Which speaker is human and which is a robot?

I wrote about a "strong" version of the Turing Test in this old post from 2004:

When I first read about the Turing test as a kid, I thought it was pretty superficial. I even wrote some silly programs which would respond to inputs, mimicking conversation. Over short periods of time, with an undiscerning tester, computers can now pass a weak version of the Turing test. However, one can define the strong version as taking place over a long period of time, and with a sophisticated tester. Were I administering the test, I would try to teach the second party something (such as quantum mechanics) and watch carefully to see whether it could learn the subject and eventually contribute something interesting or original. Any machine that could do so would, in my opinion, have to be considered intelligent.

AI isn't ready to pass the strong Turing Test, yet. But humans will become increasing unsure about the machine intelligences proliferating in the world around them.

The key to all AI advances is to narrow the scope of the problem so that the machine can deal with it. Optimization/Learning in lower dimensional spaces is much easier than in high dimensional spaces. In sufficiently narrow situations (specific tasks, abstract games of strategy, etc.), machines are already better than humans.

Google AI Blog:

Google Duplex: An AI System for Accomplishing Real-World Tasks Over the Phone

...Today we announce Google Duplex, a new technology for conducting natural conversations to carry out “real world” tasks over the phone. The technology is directed towards completing specific tasks, such as scheduling certain types of appointments. For such tasks, the system makes the conversational experience as natural as possible, allowing people to speak normally, like they would to another person, without having to adapt to a machine.

One of the key research insights was to constrain Duplex to closed domains, which are narrow enough to explore extensively. Duplex can only carry out natural conversations after being deeply trained in such domains. It cannot carry out general conversations.

Here are examples of Duplex making phone calls (using different voices)...

I switched from iOS to Android in the last year because I could see that Google Assistant was much better than Siri and was starting to have very intriguing capabilities!

Big Tech compensation in 2018

I don't work in Big Tech so I don't know whether his numbers are realistic. If they are realistic, then I'd say careers in Big Tech (for someone with the ability to do high level software work) dominate all the other (risk-adjusted) options right now. This includes finance, startups, etc.

No wonder the cost of living in the bay area is starting to rival Manhattan!

Anyone care to comment?

Meanwhile, in the low-skill part of the economy:

The Economics of Ride-Hailing: Driver Revenue, Expenses and Taxes

MIT Center for Energy and Environmental Policy Research

We perform a detailed analysis of Uber and Lyft ride-hailing driver economics by pairing results from a survey of over 1100 drivers with detailed vehicle cost information. Results show that per hour worked, median profit from driving is $3.37/hour before taxes, and 74% of drivers earn less than the minimum wage in their state. 30% of drivers are actually losing money once vehicle expenses are included. On a per-mile basis, median gross driver revenue is $0.59/mile but vehicle operating expenses reduce real driver profit to a median of $0.29/mile. For tax purposes the $0.54/mile standard mileage deduction in 2016 means that nearly half of drivers can declare a loss on their taxes. If drivers are fully able to capitalize on these losses for tax purposes, 73.5% of an estimated U.S. market $4.8B in annual ride-hailing driver profit is untaxed.

Note Uber disputes this result and claims the low hourly result is due in part to the researchers misinterpreting one of the survey questions. Uber's analysis puts the hourly compensation at ~$15.

Pocket AI from Beijing and Smartphones

I need to replace my old iPhone 6, and, predictably, this led me down the rabbit hole of learning about mobile phones, the mobile industry, and even mobile technologies. Some quick remarks: from the least to most expensive phones, Chinese companies are now competitive with industry leaders like Samsung and Apple. The Chinese market is hyper-competitive: small innovative startups (Oppo, OnePlus, etc.) compete with medium sized entities (e.g., Xiaomi, only recently a small startup itself) and giants like Huawei and Lenovo (Motorola). To gauge the landscape, watch phone reviews by Indian techies (or this guy in Germany), who tend to be very focused on cost performance and have access to handsets not sold in the US.

Here's a short video about OnePlus which also explains a bit about the Shenzhen hardware ecosystem:




Huawei's Kirin 970 chipset includes a dedicated "Neural Processor Unit" (NPU), optimized for the matrix operations used in machine learning. An NPU allows the phone to execute ML code for tasks such as image and voice recognition, language translation, etc. without relying on cloud connectivity. At the moment it is mostly a marketing gimmick, but one can imagine in a few years (perhaps earlier!) the NPU could be as important to the phone experience as the GPU.

Here's a review of the Mate 10 Pro, Huawei's $1k flagship phone, with a brief demo of some of the AI features:



The NPU appears to be based on technology licensed from a small Beijing startup, Cambricon. The founder is an alumnus of the Special Class for Gifted Young, University of Science and Technology of China. I've reviewed many Physics PhD applications from 19 year old graduates of this program. There is an SV bidding war over chip designers in this area, ever since the advent of Google's proprietary TPU (and software package Tensorflow), which accounts for most of its computation at data centers around the world.

Here's a quick demo of text recognition and machine translation from Chinese to English:




Some marketing video about the AI processor:




From cat recognition to Her or Joi? How long? I was recently offered the opportunity to be a beta tester for a startup that is building a smartphone AI assistant. I was intrigued but didn't want to give them access to all of my information...




PS One of the reasons I am leaving iOS for Android is that Google Assistant is getting very good, whereas in my experience Siri is terrible!

Meanwhile, down on the Farm

Note Added in response to 2020 Twitter mob attack which attempts to misrepresent my views: This blog post discusses the firing of James Damore by Google. It was a sensation at the time in Silicon Valley and made national news. This post is primarily about the scientific content of Damore's memo. Initial media reports describing his memo were very misleading and few people made the effort to read what Damore actually wrote before attacking him. I happened to notice that the Stanford Medical School magazine had (by coincidence) just featured an article on some of the issues discussed by Damore. Whether (below) the Stanford neuroscientist Nirao Shah or the former President of the American Psychological Association Diane Halpern are correct or not about the science, it seems unfair to call Damore a crank if he is simply referencing (in good faith) results in the published scientific literature. The same kinds of results are presented in the article below, written for the alumni of Stanford Medical School.

In the second part of the post below I describe some recent survey results on individual preferences among mathematically gifted men and women who are part of a ~50 year longitudinal study -- they have been studied since childhood. I note specifically that differences in preferences between men and women are not necessarily biological in origin (we simply don't know): they could be the result of sexism in child rearing, schooling, postdoc training, etc.

However, the point is that the survey results are likely descriptive of how actual adult men and women think and feel, and may have implications for labor markets. This is NOT a discussion about ability differences between men and women (all the individuals in the study are mathematically gifted), but rather about preferences concerning life fulfillment, lifestyle, work-life balance, etc. And again, no causation is assumed -- the situation may be entirely due to sexism in society, with zero biological basis.

The Spring 2017 issue of the Stanford Medical School magazine has a special theme: Sex, Gender, and Medicine. I recommend the article excerpted below to journalists covering the Google Manifesto / James Damore firing. After reading it, they can decide for themselves whether his memo is based on established neuroscience or bro-pseudoscience.

Perhaps top Google executives will want to head down the road to Stanford for a refresher course in reality.

Stanford Neuroscience Professor Nirao Shah and Diane Halpern, past president of the American Psychological Association, would both make excellent expert witnesses in the Trial of the Century.

Two minds: The cognitive differences between men and women

... Nirao Shah decided in 1998 to study sex-based differences in the brain ... “I wanted to find and explore neural circuits that regulate specific behaviors,” says Shah, then a newly minted Caltech PhD who was beginning a postdoctoral fellowship at Columbia. So, he zeroed in on sex-associated behavioral differences in mating, parenting and aggression.

“These behaviors are essential for survival and propagation,” says Shah, MD, PhD, now a Stanford professor of psychiatry and behavioral sciences and of neurobiology. “They’re innate rather than learned — at least in animals — so the circuitry involved ought to be developmentally hard-wired into the brain. These circuits should differ depending on which sex you’re looking at.”

His plan was to learn what he could about the activity of genes tied to behaviors that differ between the sexes, then use that knowledge to help identify the neuronal circuits — clusters of nerve cells in close communication with one another — underlying those behaviors.

At the time, this was not a universally popular idea. The neuroscience community had largely considered any observed sex-associated differences in cognition and behavior in humans to be due to the effects of cultural influences. Animal researchers, for their part, seldom even bothered to use female rodents in their experiments, figuring that the cyclical variations in their reproductive hormones would introduce confounding variability into the search for fundamental neurological insights.

But over the past 15 years or so, there’s been a sea change as new technologies have generated a growing pile of evidence that there are inherent differences in how men’s and women’s brains are wired and how they work.

... There was too much data pointing to the biological basis of sex-based cognitive differences to ignore, Halpern says. For one thing, the animal-research findings resonated with sex-based differences ascribed to people. These findings continue to accrue. In a study of 34 rhesus monkeys, for example, males strongly preferred toys with wheels over plush toys, whereas females found plush toys likable. It would be tough to argue that the monkeys’ parents bought them sex-typed toys or that simian society encourages its male offspring to play more with trucks. A much more recent study established that boys and girls 9 to 17 months old — an age when children show few if any signs of recognizing either their own or other children’s sex — nonetheless show marked differences in their preference for stereotypically male versus stereotypically female toys.

Halpern and others have cataloged plenty of human behavioral differences. “These findings have all been replicated,” she says.

... “You see sex differences in spatial-visualization ability in 2- and 3-month-old infants,” Halpern says. Infant girls respond more readily to faces and begin talking earlier. Boys react earlier in infancy to experimentally induced perceptual discrepancies in their visual environment. In adulthood, women remain more oriented to faces, men to things.

All these measured differences are averages derived from pooling widely varying individual results. While statistically significant, the differences tend not to be gigantic. They are most noticeable at the extremes of a bell curve, rather than in the middle, where most people cluster. ...

See also Gender differences in preferences, choices, and outcomes: SMPY longitudinal study. These preference asymmetries are not necessarily determined by biology. They could be entirely due to societal influences. But nevertheless, they characterize the pool of human capital from which Google is trying to hire.

The recent SMPY paper below describes a group of mathematically gifted (top 1% ability) individuals who have been followed for 40 years. This is precisely the pool from which one would hope to draw STEM and technological leadership talent. There are 1037 men and 613 women in the study.

The figures show significant gender differences in life and career preferences, which affect choices and outcomes even after ability is controlled for. (Click for larger versions.) According to the results, SMPY men are more concerned with money, prestige, success, creating or inventing something with impact, etc. SMPY women prefer time and work flexibility, want to give back to the community, and are less comfortable advocating unpopular ideas. Some of these asymmetries are at the 0.5 SD level or greater. Here are three survey items with a ~ 0.4 SD or more asymmetry:

# Society should invest in my ideas because they are more important than those of other people.

# Discomforting others does not deter me from stating the facts.

# Receiving criticism from others does not inhibit me from expressing my thoughts.

I would guess that Silicon Valley entrepreneurs and leading technologists are typically about +2 SD on each of these items! One can directly estimate M/F ratios from these parameters ...

For example, if a typical male SV entrepreneur / tech leader is roughly +2SD on these traits whereas a female is +2.5SD, the population fraction would be 3:1 or 4:1 larger for males. This doesn't mean that the females who are > +2.5SD (in the female population) are ill-suited to the role (they may be as good as the men), just that there are fewer of them in the general population. I was shocked to see that even top Google leadership didn't understand this point that Damore tried to make in his memo.

A 6ft3 Asian-American guard (Jeremy Lin) might be just as good as other guards in the NBA, but the fraction of Asian-American males who are 6ft3 is smaller than for other groups, like African-Americans. Even if there were no discrimination against Asian players, you'd expect to see fewer (relative to base population) in the NBA due to the average height difference.

Behold the Brogrammer: James Damore (Bloomberg video)

Watch a few minutes of this Bloomberg interview and I think you'll agree he's both sincere and well-meaning, if a bit naive about the buzzsaw he has stepped into. Definitely not a brogrammer.

He reminds me of Richard Hendricks of the HBO show Silicon Valley.

See also Damore vs Google: Trial of the Century? and In the matter of James Damore, ex-Googler

Damore vs Google: Trial of the Century?

Note Added in response to 2020 Twitter mob attack which attempts to misrepresent my views: This blog post discusses the firing of James Damore by Google. It was a sensation at the time in Silicon Valley and made national news. This post is primarily about the legal status of the firing, and the legal status of the claims made by Damore in his memo. Initial media reports describing his memo were very misleading and few people made the effort to read what Damore actually wrote before attacking him.

In his memo, James Damore asserts that Google is engaged in illegal discriminatory practices as part of its efforts to increase diversity. (See earlier post, In the matter of James Damore, ex-Googler.)

The image below is from the actual memo. Does Damore sound like a sexist brogrammer Neanderthal?

OKRs = Objectives and Key Results. Damore is pointing out that pro-diversity objectives may incentivize managers to discriminate by gender or race in hiring and promotion.

According to Margot Cleveland (attorney who teaches labor law at Notre Dame):

The Federalist: ... Damore wrote “Google has created several discriminatory practices.” This reads of a classic case of opposition to an unlawful employment practice. (Under the case law, the practice need not actually be illegal if the employee reasonably believed it discriminatory.)

This passage may well be Google’s undoing. Damore can present a prima facie case of illegal retaliation: he engaged in protected activity by opposing several discriminatory practices, and was fired from his job. The close temporal nexus creates an inference that Google fired him because of his opposition to illegal discrimination.

... Google will counter that it fired him not because of his opposition but because of the gender stereotypes he included in the memo.

But of course the Google Brain was simultaneously using these "stereotypes" = correlations as its core revenue driver:

Professor Cleveland concludes:

... Once before a jury, Google will be hard-pressed to justify Damore’s firing because the jury will be force-fed the actual words Damore wrote, not the press’ hysterical gloss. In this regard, Google was in a no-win situation: Once the Neanderthal narrative formed, Google had no real choice but to fire Damore—which doesn’t make it right or, as Google is likely to find out soon, legal. In the meantime, the rest of the country will be treated to a nice civics refresher course and a deep-dive into federal employment and labor law.

Not to mention a deep-dive into the science of statistical / distributional group differences!

Bloomberg video interview with Damore.

In the matter of James Damore, ex-Googler

James Damore, Harvard PhD* in Systems Biology, and (until last week) an engineer at Google, was fired for writing this memo: Google’s Ideological Echo Chamber, which dares to display the figure above.

Here is Damore's brief summary of his memo (which contains many citations to original scientific research), and the conclusion:

Google’s political bias has equated the freedom from offense with psychological safety, but shaming into silence is the antithesis of psychological safety.
● This silencing has created an ideological echo chamber where some ideas are too sacred to be honestly discussed.
● The lack of discussion fosters the most extreme and authoritarian elements of this ideology.
○ Extreme: all disparities in representation are due to oppression
○ Authoritarian: we should discriminate to correct for this oppression
● Differences in distributions of traits between men and women may in part explain why we don't have 50% representation of women in tech and leadership.
● Discrimination to reach equal representation is unfair, divisive, and bad for business.


I hope it’s clear that I’m not saying that diversity is bad, that Google or society is 100% fair, that we shouldn’t try to correct for existing biases, or that minorities have the same experience of those in the majority. My larger point is that we have an intolerance for ideas and evidence that don’t fit a certain ideology. I’m also not saying that we should restrict people to certain gender roles; I’m advocating for quite the opposite: treat people as individuals, not as just another member of their group (tribalism).

This actual excerpt is of course very different from the heavily biased (mendacious) characterizations of the memo in the (lying) media. Perhaps that should make you wonder about the reliability of mainstream accounts concerning this matter.

Damore correctly anticipated his own demise! CEO Sundar Pichai's company-wide message seems to ban almost all scientific discussion of statistical or distributional group differences, on threat of termination:

This has been a very difficult time. I wanted to provide an update on the memo that was circulated over this past week.

First, let me say that we strongly support the right of Googlers to express themselves, and much of what was in that memo is fair to debate, regardless of whether a vast majority of Googlers disagree with it. However, portions of the memo violate our Code of Conduct and cross the line by advancing harmful gender stereotypes in our workplace. Our job is to build great products for users that make a difference in their lives. To suggest a group of our colleagues have traits that make them less biologically suited to that work is offensive and not OK. It is contrary to our basic values and our Code of Conduct, which expects “each Googler to do their utmost to create a workplace culture that is free of harassment, intimidation, bias and unlawful discrimination.”

The memo has clearly impacted our co-workers, some of whom are hurting and feel judged based on their gender. Our co-workers shouldn’t have to worry that each time they open their mouths to speak in a meeting, they have to prove that they are not like the memo states, being “agreeable” rather than “assertive,” showing a “lower stress tolerance,” or being “neurotic.”

At the same time, there are co-workers who are questioning whether they can safely express their views in the workplace (especially those with a minority viewpoint). They too feel under threat, and that is also not OK. People must feel free to express dissent. So to be clear again, many points raised in the memo—such as the portions criticizing Google’s trainings, questioning the role of ideology in the workplace, and debating whether programs for women and underserved groups are sufficiently open to all—are important topics. The author had a right to express their views on those topics—we encourage an environment in which people can do this and it remains our policy to not take action against anyone for prompting these discussions. ...

Larry Summers was fired from the Harvard presidency (at least in part) for pointing out (correctly, it seems) that males exhibit higher variance in performance on cognitive tests (more very low- and high-scoring men than women per capita). His detractors justified the termination due to his highly public and symbolic role as the leader of the institution. In contrast, Damore was simply an engineer (with a background in computational biology) expressing his opinion on some basic scientific questions still under active investigation by researchers all over the world. His firing has to be regarded as scary authoritarian policing of thought.

See also Bounded Cognition, Gender differences in preferences, choices, and outcomes, 2:1 faculty preference for women on STEM tenure track (PNAS), and Gender trouble in the valley.

A literature review at Slate Star Codex.

If I worked at Google would this blog post get me fired?


Note Added:

* Damore may be ABD (left Harvard before completing his dissertation) rather than a PhD.

Damore is going to fight Google in court (NYTimes):

Mr. Damore, who worked on infrastructure for Google’s search product, said he believed that the company’s actions were illegal and that he would “likely be pursuing legal action.”

“I have a legal right to express my concerns about the terms and conditions of my working environment and to bring up potentially illegal behavior, which is what my document does,” Mr. Damore said.

Before being fired, Mr. Damore said, he had submitted a complaint to the National Labor Relations Board claiming that Google’s upper management was “misrepresenting and shaming me in order to silence my complaints.” He added that it was “illegal to retaliate” against an N.L.R.B. charge.

According to The Federalist, Damore has a case. This trial of the century might expose large numbers of people to the ideas in his memo...

Bloomberg video interview with Damore.

Thought vectors and the dimensionality of the space of concepts

This NYTimes Magazine article describes the implementation of a new deep neural net version of Google Translate. The previous version used statistical methods that had reached a plateau in effectiveness, due to limitations of short-range correlations in conditional probabilities. I've found the new version to be much better than the old one (this is quantified a bit in the article).

These are some of the relevant papers. Recent Google implementation, and new advances:
https://arxiv.org/abs/1609.08144, https://arxiv.org/abs/1611.04558.

Le 2014, Baidu 2015, Lipton et al. review article 2015.

More deep learning.

NYTimes: ... There was, however, another option: just design, mass-produce and install in dispersed data centers a new kind of chip to make everything faster. These chips would be called T.P.U.s, or “tensor processing units,” ... “Normally,” Dean said, “special-purpose hardware is a bad idea. It usually works to speed up one thing. But because of the generality of neural networks, you can leverage this special-purpose hardware for a lot of other things.” [ Nvidia currently has the lead in GPUs used in neural network applications, but perhaps TPUs will become a sideline business for Google if their TensorFlow software becomes widely used ... ]

Just as the chip-design process was nearly complete, Le and two colleagues finally demonstrated that neural networks might be configured to handle the structure of language. He drew upon an idea, called “word embeddings,” that had been around for more than 10 years. When you summarize images, you can divine a picture of what each stage of the summary looks like — an edge, a circle, etc. When you summarize language in a similar way, you essentially produce multidimensional maps of the distances, based on common usage, between one word and every single other word in the language. The machine is not “analyzing” the data the way that we might, with linguistic rules that identify some of them as nouns and others as verbs. Instead, it is shifting and twisting and warping the words around in the map. In two dimensions, you cannot make this map useful. You want, for example, “cat” to be in the rough vicinity of “dog,” but you also want “cat” to be near “tail” and near “supercilious” and near “meme,” because you want to try to capture all of the different relationships — both strong and weak — that the word “cat” has to other words. It can be related to all these other words simultaneously only if it is related to each of them in a different dimension. You can’t easily make a 160,000-dimensional map, but it turns out you can represent a language pretty well in a mere thousand or so dimensions — in other words, a universe in which each word is designated by a list of a thousand numbers. Le gave me a good-natured hard time for my continual requests for a mental picture of these maps. “Gideon,” he would say, with the blunt regular demurral of Bartleby, “I do not generally like trying to visualize thousand-dimensional vectors in three-dimensional space.”

Still, certain dimensions in the space, it turned out, did seem to represent legible human categories, like gender or relative size. If you took the thousand numbers that meant “king” and literally just subtracted the thousand numbers that meant “queen,” you got the same numerical result as if you subtracted the numbers for “woman” from the numbers for “man.” And if you took the entire space of the English language and the entire space of French, you could, at least in theory, train a network to learn how to take a sentence in one space and propose an equivalent in the other. You just had to give it millions and millions of English sentences as inputs on one side and their desired French outputs on the other, and over time it would recognize the relevant patterns in words the way that an image classifier recognized the relevant patterns in pixels. You could then give it a sentence in English and ask it to predict the best French analogue.

That the conceptual vocabulary of human language (and hence, of the human mind) has dimensionality of order 1000 is kind of obvious*** if you are familiar with Chinese ideograms. (Ideogram = a written character symbolizing an idea or concept.) One can read the newspaper with mastery of roughly 2-3k characters. Of course, some minds operate in higher dimensions than others ;-)

The major difference between words and pixels, however, is that all of the pixels in an image are there at once, whereas words appear in a progression over time. You needed a way for the network to “hold in mind” the progression of a chronological sequence — the complete pathway from the first word to the last. In a period of about a week, in September 2014, three papers came out — one by Le and two others by academics in Canada and Germany — that at last provided all the theoretical tools necessary to do this sort of thing. That research allowed for open-ended projects like Brain’s Magenta, an investigation into how machines might generate art and music. It also cleared the way toward an instrumental task like machine translation. Hinton told me he thought at the time that this follow-up work would take at least five more years.

The entire article is worth reading (there's even a bit near the end which addresses Searle's Chinese Room confusion). However, the author underestimates the importance of machine translation. The "thought vector" structure of human language encodes the key primitives used in human intelligence. Efficient methods for working with these structures (e.g., for reading and learning from vast quantities of existing text) will greatly accelerate AGI.

*** Some further explanation, from the comments:

The average person has a vocabulary of perhaps 10-20k words. But if you eliminate redundancy (synonyms + see below) you are probably only left with a few thousand words. With these words one could express most concepts (e.g., those required for newspaper articles). Some ideas might require concatenations of multiple words: "cougar" = "big mountain cat" , etc.

But the ~1k figure gives you some idea of how many distinct "primitives" (= "big", "mountain", "cat") are found in human thinking. It's not the number of distinct concepts, but rather the rough number of primitives out of which we build everything else.

Of course, truly deep areas of science discover / invent new concepts which are almost new primitives (fundamental, but didn't exist before!), such as "entropy", "quantum field", "gauge boson", "black hole", "natural selection", "convex optimization", "spontaneous symmetry breaking", "phase transition" etc.

If we trained a deep net to translate sentences about Physics from Martian to English, we could (roughly) estimate the "conceptual depth" of the subject. We could even compare two different subjects, such as Physics versus Art History.

Scifoo 2016

Photos from Palo Alto and Scifoo 2016. We weren't allowed to take photos inside the Googleplex.

Productive Bubbles

These slides are from one of the best sessions I attended at scifoo. Bill Janeway's perspective was both theoretical and historical, but in addition we had Sam Altman of Y Combinator to discuss Airbnb and other examples of 2 way market platforms (Uber, etc.) that may be enjoying speculative bubbles at the moment.

See also Andrew Odlyzko (Caltech '71 ;-) on British railway manias for specific cases of speculative funding of useful infrastructure: here, here and here.

Sci Foo 2015

I'm in Palo Alto for this annual meeting of scientists and entrepreneurs at Google. If you read this blog, come over and say hello!

Action photos! Note most of the sessions were in smaller conference rooms, but we weren't allowed to take photographs there.

SCI FOO 2014: photos

The day before SCI FOO I visited Complete Genomics, which is very close to the Googleplex.

Self-driving cars:

SCI FOO festivities:

I did an interview with O'Reilly. It should appear in podcast form at some point and I'll post a link.

Obligatory selfie:

@ SCI FOO 2014

Sorry for the lack of blog activity. I just returned from Asia and am in Palo Alto for SCI FOO 2014. Hopefully I'll post some cool photos from the event, which starts tomorrow evening. If you are there and read this blog then come over and say hello. If I had free t-shirts I'd give you one, but don't get your hopes up!

Earlier SCI FOO posts.

Larry's rules

I recommend this long article about Larry Page and the evolution of his role at Google.

Here are Larry's management rules (more suitable, perhaps, for a startup than to a larger, mature organization):

Don't delegate: Do everything you can yourself to make things go faster.

Don't get in the way if you're not adding value. Let the people actually doing the work talk to each other while you go do something else.

Don't be a bureaucrat.

Ideas are more important than age. Just because someone is junior doesn't mean they don't deserve respect and cooperation.

The worst thing you can do is stop someone from doing something by saying, “No. Period.” If you say no, you have to help them find a better way to get it done.

The Will to Power

Great bio of former Googler and now Yahoo CEO Marissa Mayer. I got sucked in and read the whole thing. Lots of color on life at Google, Yahoo and in SV. Philosophy 160A at Stanford is intro to mathematical logic.

... Mayer credits her teachers for helping her become less shy.

They did this by showing Mayer that she could “organize” more than just her backpack, desk, and homework — that she could organize people, as their leader.

Mayer’s childhood piano teacher, Joanne Beckman, remembers Mayer being very different from other children in that she was someone who “watched people” in order to “figure out why they were doing what they were doing.”

“A lot of kids that age are very interested in themselves,” Beckman says, “She was looking at other people.”

By “looking” at her teachers, figuring out why they were doing what they were doing, Mayer overcame her “painful” shyness with peers by taking on the teacher’s role.

Even when she was in fifth grade, Mr. Flanagan could see the pedagogical side of Mayer developing. He thought she would become a teacher someday.

... In 1993, Mayer applied to, and was accepted into, 10 schools, including Harvard, Yale, Duke, and Northwestern.

To decide which one she would go to, Mayer created a spreadsheet, weighing variables for each.

She picked Stanford. Her plan was to become a brain doctor — a profession that doesn’t draw much on the leadership traits Mayer was quickly developing.

... That summer, Mayer attended the National Youth Science Camp in West Virginia. It was nerd heaven. Picture science labs housed in wooden cabins shaded by trees. Mayer especially loved one experiment where they mixed water and corn starch to make a sloppy goo-like substance that seemed to defy gravity.

One day, a post-doctoral student from Yale named Zune Nguyen spoke to the campers as a guest lecturer. He stunned all the smart kids in the room with puzzles and brainteasers. For days, the campers couldn’t stop talking about his talk.

Finally, one of Mayer’s counselors had enough.

“You know, you have it all wrong,” the counselor said to Mayer and the campers. “It’s not what Zune knows, it’s how Zune thinks.”

The counselor said that what made Nguyen so amazing wasn’t the facts that he knew, but rather how he approached the world and how he thought about problems. The counselor said the most remarkable thing about Nguyen was that you could put him in an entirely new environment or present him with an entirely new problem, and within a matter of minutes he would be asking the right questions and making the right observations.

From that moment on, the phrase: “It’s not what Zune knows, but how Zune thinks,” stuck with Mayer as a sort of personal guiding proverb.

In the fall, Mayer went to Stanford and began taking pre-med classes. She planned to become a doctor. But by the end of her freshman year, she was sick of it.

“I was just doing too many flashcards,” she says. “They were easy for me, but it was just a lot of memorization.”

She says she wanted to find a major “that really made me think” — that would train her to “think critically, and become a great problem-solver.” She also wanted to “study how people think, how they reason, how they express themselves.”

“I had this nagging voice in my head saying ‘It’s not what Zune knows, but how Zune thinks.’”

... So that semester at Stanford was full of all-nighters for Mayer and her Philosophy 160A group.

Mayer ended up in a group that included Josh Elman, now a venture capitalist. Looking back on those study sessions, Elman remembers “times when people in the group were bouncing off the walls.”

He says, “Marissa was always like, ‘OK, back to work. Let’s get this done.’ She was focused on making sure we got the right answer quickly.”

“It felt like she was the smartest student in the room — and the most serious. You always knew those two things about her. Very smart. Very serious.”

The social dynamic of the group was typical for Mayer. As usual, she commanded the room — organized the group’s work in an all-business fashion — but was otherwise shy, and somewhat reclusive.

In the years ahead, this combination — Mayer’s willingness to be authoritative and demanding the way a teacher would, with a “painful” fear or reluctance of being personal — would cause problems for Mayer.

One Stanford classmate interpreted Mayer’s shyness as being “kind of stuck up.”

“She would do her work and then leave. When other people would stay and hang out and have pizza, she’d just be out of there because the work is done.”

Indeed, Mayer doesn’t seem to have had a very active social life in college.

One person who lived in her dorm said she appeared to always be “down to business” and “not much for socializing.”

“She wasn’t one of those people into making new friends around the dorm. She was always doing something more important than just chilling.”

The simplest explanation for Mayer’s social behavior at Stanford remains that Mayer was, as she has said many times, “painfully shy.” ...

Note the geeky laugh and the number of times she says "really smart people" ;-) @24min she talks about her personal strengths and decision strategies.

Google Glass

What's it like to try Google Glass?

I WALKED AWAY CONVINCED THIS WASN’T JUST ONE OF GOOGLE’S WEIRD FLIGHTS OF FANCY

Is it ready for everyone right now? Not really. Does the Glass team still have huge distance to cover in making the experience work just the way it should every time you use it? Definitely.

But I walked away convinced that this wasn’t just one of Google’s weird flights of fancy. The more I used Glass the more it made sense to me; the more I wanted it. If the team had told me I could sign up to have my current glasses augmented with Glass technology, I would have put pen to paper (and money in their hands) right then and there. And it’s that kind of stuff that will make the difference between this being a niche device for geeks and a product that everyone wants to experience.

After a few hours with Glass, I’ve decided that the question is no longer ‘if,’ but ‘when?’