Jeff Bezos: "Kingpin" AI assistant

43  2017-11-25 by bezos_kingpin

I happened across a character at Amazon who was working on machine learning for the executive branch. Specifically, for Jeff Bezos himself. This AI "daily world briefing at a glance" was called 'Kingpin'.

Amazon executives traditionally received daily briefings compiled in the form of "[x]-pagers" (ie one-pagers, five-pagers, etc.). Bezos has a strong tendency to promote the concept of writing ideas down in stone, from his time with books. Amazon decided to put it's machine learning knowledge to use by creating a dedicated machine learning/neural network assistant to crawl the entire web, stock markets, news releases, internal company reports, calendars etc. and to summarize these points into the desired briefing length for the day.

Bezos would then meet (or teleconference) with other high level executives and discuss the daily points of this briefing. This provides an extremely valuable level of intelligence because Kingpin can crawl multiple data sources (the only technical details I know were about normalizing these different data streams) at a mind-bogglingly fast rate compared to a normal assistant (or team of assistants).

Bezos has recently passed $100B net worth, and his acquisitions of Washington Post, Whole Foods, and their secure service for US government uses have seemed like extreme successes (whether morally agreeable or not). Kingpin clearly seems to be working. What does this mean about what other business executives might be working on? Why would Zuckerberg (who talks about creating his own "Jarvis") or Sergey Brin or Bill Gates not have the same?

To clarify, this AI is not smart. It is a statistical whiz-kid which evaluates billions of possible discussion points and creates the most concise summary which touches on every important event for the day. But that speed and breadth is extremely, extremely valuable to someone with so much scale under their control.

Is this the future? A ('dumb') AI selecting which articles are of importance in our daily news-feed? Using AI for information compression is a very important thing as we already know from comment bots which summarize articles.

Jeff Bezos is ready for the future. How will you outpace him?

14 comments

1  __G_A_R_D_E_N_E_R__ 2017-11-25

Scooping up all actionable data from human computation.

1  bezos_kingpin 2017-11-25

Yes, I was curious about how they ranked which items were most likely to be picked by a human. The labelled data must have used statistical trends developed from things like Amazon's daily product trends (you can infer what people are interested in at the moment based on what they buy) or something like the WaPo's front page traffic (which news stories get the most clicks? Which big stories get less clicks than they should?).

I wish I spent my time getting more detail. Unfortunately, technical information was aimed at a very basic level of understanding. Mostly learned about normalizing multiple data streams and creating a generic "crawler/encoder" that could create brief sentences from a variety of interesting data points (ie. whether an anomalous stock price, news story, or upcoming event, the output is the same natural language sentence format).

It's amazing how powerful such an assistant would be. Knowledge is power.

1  bezos_kingpin 2017-11-25

Yes, I was curious about how they ranked which items were most likely to be picked by a human. The labelled data must have used statistical trends developed from things like Amazon's daily product trends (you can infer what people are interested in at the moment based on what they buy) or something like the WaPo's front page traffic (which news stories get the most clicks? Which big stories get less clicks than they should?).

I wish I spent my time getting more detail. Unfortunately, technical information was aimed at a very basic level of understanding. Mostly learned about normalizing multiple data streams and creating a generic "crawler/encoder" that could create brief sentences from a variety of interesting data points (ie. whether an anomalous stock price, news story, or upcoming event, the output is the same natural language sentence format).

It's amazing how powerful such an assistant would be. Knowledge is power.

1  bezos_kingpin 2017-11-25

Yes, I was curious about how they ranked which items were most likely to be picked by a human. The labelled data must have used statistical trends developed from things like Amazon's daily product trends (you can infer what people are interested in at the moment based on what they buy) or something like the WaPo's front page traffic (which news stories get the most clicks? Which big stories get less clicks than they should?).

I wish I spent my time getting more detail. Unfortunately, technical information was aimed at a very basic level of understanding. Mostly learned about normalizing multiple data streams and creating a generic "crawler/encoder" that could create brief sentences from a variety of interesting data points (ie. whether an anomalous stock price, news story, or upcoming event, the output is the same natural language sentence format).

It's amazing how powerful such an assistant would be. Knowledge is power.

1  __G_A_R_D_E_N_E_R__ 2017-11-25

Power corrupts.

1  -AVENTUS- 2017-11-25

I'm still on the fence regarding Bezos. Even knowing this I have not been able to make a judgement on his motivations yet.

1  bezos_kingpin 2017-11-25

I think he is the corporate viral death of the Earth, personally. Amazon treats employees like trash when they can (warehouse workers) and keeps their tech workers extremely compartmentalized. They have no goal other than to expand at all costs.

The recent "bio-dome" in Seattle was actually an experiment testing a possible configuration for living quarters on Mars. They were monitoring employee patterns and trying to determine the optimal configuration. I don't think the employees inside the dome knew about that before-hand. Bezos is also purely using this for his own benefit, as Amazon doesn't have anything to do with Blue Origin... allegedly.

1  -AVENTUS- 2017-11-25

I'm still not sure. I have to watch old interviews of him before he had money, and current interviews. And learn more about amazon.

Obviously the warehouse and delivery employees are paid lowly, and this is not the most honorable thing. But I would have to balance pros/cons compared to previous market leaders like wall mart to assess overall damage as it relates to consumer benefit, and insane product diversity.

1  Irapotato 2017-11-25

I work at amazon, and $13.25 starting pay for warehouse workers is certainly not trash lmao. And in terms of keeping tech workers compartmentalized...you have inside knowledge of how amazon treats tech workers? Lol. Expanding is kind of the goal of every corporation.

1  cheesuscripes 2017-11-25

Are you an Amazon employee who works in the warehouse, or a temp?

The "temps" in Nevada for instance were treated much worse than the regular employees, and laid off like clockwork, in a rotating schedule so as to not trigger unemployment insurance.

I could go on...

1  HussellWilson 2017-11-25

$13.25 definitely isn't good, especially for a warehouse worker, what are you talking about?

1  Richie209 2017-11-25

When minimum wage is $11 in January, working 4 10 hour shifts a week for what a supervisor at a fast food joint makes isn't much better. I'm located near a few Amazon warehouses and many former employees who all have had nothing but negative things to say about this employer. Miss me with the "but they have a cool cafeteria and heath benefits"

1  martini-meow 2017-11-25

What is Blue Origin? (steps away to web research,too)

1  HussellWilson 2017-11-25

You talking about the glass sphere thing? I work next to it, it isn't even finished yet and it will be open to the public.

1  Compl3t3lyInnocent 2017-11-25

ELI5?

1  bezos_kingpin 2017-11-25

Jeff Bezos has an Amazon AI project called "Kingpin" that searches multiple data-bases of current and past information to provide a multi-page daily briefing of the most important things to know/decide about that day.

This AI is not "smart" (ie. it's only statistical) but it provides him with a massive advantage of being able to control something with the scale of Amazon. The name of the project is slightly unnerving.