Google Your Search Is Over (Part 2)

All Your Data Are Belong To Us

We ended off yesterday with a brief note about Google’s strategy of freely giving away all kinds of great stuff...

• Maps
• YouTube
• Search
• Android
• Gmail
• GSuite
• Cloud Storage
• Gcal
• Chrome
• Language Translation
• etc.

...often with no apparent monetization model on the horizon.

So again, why would they do this?

I’m With Stupid

Some people get rich studying artificial intelligence. Me, I make money studying natural stupidity. - Carl Icahn

For those not in the know, a “computer” is really a universal machine - a computer can do anything you tell it to do.

But computers are also universally stupid; a computer will only do EXACTLY what you tell it to do.

If you’ve never programmed a computer before, imagine this scenario:

Your Imagination:

You are now in charge of a child who will do anything you tell them, but they cannot think for themselves, and they come with no stored knowledge.

It’s getting late, so you tell the child: Go to bed

Instead of going to bed, they stare back at you blankly and say: ERROR, Undefined

You now must explain to them what a bed is, which you do.

Let’s try this again.

You: Go to bed.

Now the child goes TO bed, but does not get in.

You: Go to bed THEN get in bed THEN go to sleep.

Now the child goes to sleep without brushing their teeth, changing, etc...

Let’s take it up a notch.

You’re on vacation in cottage country, a 3 hour drive from home, and it’s bedtime, so you tell the child: Go to your room THEN get changed THEN go to bed THEN get in bed THEN go to sleep.

The child now goes back to THEIR bed, at your home, 150 miles away.

Now try to imagine hard coding something like driving a car, or worse, something nebulous like “humour...”

That said, once a computer knows how to do something, it will do it exactly the same way, every single time. (Ingredient 1: Perfect Memory)

Enter Machine Learning

Not So Stupid After All

DeepMind cofounder Shane Legg’s formal definition of intelligence).

Can anyone here define intelligence?

It’s deceptively difficult.

Scott Guthrie - probably the 2nd most powerful MSFT executive after Satya Nadella - once defined (artificial) intelligence as “**just trained statistical analytics based on past data.”*

Controversial futurist Ray Kurzweil would shorten that to “pattern recognition...”

And interestingly, he’s not the only one.

• Jim Simons, the most successful hedge fund manager ever, did his seminal work in, and earned his fortune by applying, pattern recognition.

• Jeff Bezos said: “the human brain is an incredible pattern-matching machine”

• Warren Buffett is noted within the investment community for his pattern recognition abilities

The photo below is a sample question from Raven’s Progressive Matrices, the most popular IQ test:
Notice any - ahem - patterns?

Now that we have basic neural networks, computers don’t need to be hard coded: users can define a goal - a “utility function” - and computers can pattern match by practicing, or by watching humans.

Unfortunately, current AI systems are slow learners: they need an insane amount of practice before seeing any useful patterns (there’s a saying in the machine learning community that “life begins at a billion moments”).

For example, back in 2011 - a lifetime ago in tech - IBM’s Watson trounced the world’s best Jeopardy players, but it needed 200mm pages of Wikipedia to do it.

(I’m sure Ken Jennings is a voracious reader, but I doubt he has read 200 million pages of anything)

All that said, once an AI system achieves human level competency, it will rapidly zoom right past... (Ingredient 2: Super Smart)

And when I say rapidly, I mean RAPIDLY.

I Feel The Need, The Need for Speed!

In real life, the hare wins

Grossly oversimplified, the upper bound of nerve conduction velocity (let’s call it “the speed of natural intelligence”) is 120 meters per second.

Under unfavorable conditions, electricity (“the speed of artificial intelligence”) travels at half the speed of light, or roughly 150,000,000 meters per second.

So worst case scenario, a computer “thinks” 1 million times faster than a human…

That’s actually hard to conceptualize, but let’s give it a shot.

Usain Bolt - probably the fastest person ever - runs the 100m dash roughly twice as fast as the average person… so basically a useless comparison.

In terms of distance over time, a reasonable approximation would be the delta between the Apollo 10 rocket ship (~40,000 km/h, and the fastest crewed vehicle ever recorded) and a snail (~0.040 km/h).

In 20 hours, the Apollo would take you to the moon and back, while the snail won’t even get you from Goldman Sachs world headquarters at 200 West to Stuyvesant High School for a look at future bankers.

Recruiting starts earlier this year

Better, but I never liked those “to the moon” comparisons.

In “thinking” terms?

One week for us is 10,000 years for a digital brain. (Ingredient 3: Amazingly Fast)

When the 3 key ingredients are mixed together:

• Perfect Memory
• Super Smart
• Amazingly Fast

We get some interesting - and potentially terrifying - results...

Hit The Earth Like a Comet, Invasion!

The development of AI would be as momentous as the landing of extraterrestrials on this planet. If aliens landed, the first question would not be about the economy! - Peter Thiel

The above video (It’s only 5 minutes, check it out) shows the evolution of a computer program from a terrible Mario Bros player to a Super Mario Bros player (get it?).

In just 24 hours, the program went from worst in the world, to the best there is, the best there was, and the best there ever will be.

This is an example of “narrow AI” - It is amazingly fast, with a perfect memory, and is super smart within its purview...

But it doesn't "look" human.

That’s not necessarily a bad thing. For example, this antenna..

...was designed for NASA by an evolutionary computer design program, and it outperformed every human variant.

This is great!

Where we might run into problems is with bad actors or ill-defined utility functions.

For example, this sort of shape shifting technology is fun...

https://www.youtube.com/watch?v=CO5ydSJzxLk
The flesh & blood terminator…

… but it’s not hard to imagine evil applications.

A business looking to “maximize the present value of future cash flows” might work out very well for society!

But who knows what could happen when an AI first hedge fund wants to blowout a quarter by buying and selling pound Sterling while staying inside some volatility limits (paging George Soros!)

A military algo that wants to protect the United States from election tampering by Russian nationals might decide to simply eliminate all Russians, or seal off all borders.

Or it could be something seemingly benign, like a music recommendation engine that forces the entire world to listen to Justin Bieber 24/7 365 (that’s the real summoning the demon, amirite?)...

So this is all very interesting, but very narrow in scope, and still this does not answer the question:

Why is Google hoovering up every single piece of data it can get its greedy “don’t be evil” hands on?

One Algo To Rule Them All

"The question of whether a computer can think is no more interesting than the question of whether a submarine can swim." - Edsger Dijkstra

There is an open research problem in AI known as “Transfer Learning,” which involves using knowledge gained from solving one problem and using it to solve a different but related problem i.e. using a model trained to recognize cars to also recognize trucks.

(Andrew Ng believes Transfer Learning will drive massive commercial success in artificial intelligence )

If you believe the premise that “intelligence” is at its core generalized pattern recognition - and some of the smartest minds in the world, including former Google CEO Eric Schmidt, who has gone on record saying that the problems Google solves are all pattern matching problems, believe it is - and that problem solving transfers across domains, then Google’s strategy of solving real world problems at scale makes perfect sense.

If GOOG can “solve intelligence” for one problem, then the company can solve intelligence for EVERY problem, just give them the data...

What value would you place on a company that employs not simply the best writer, mathematician, engineer, poet, quant, scientist, musician, comedian, physicist, etc. but the best who is also 1,000,000 times better than the second best?

This Time is Different

"I believe that probably the five largest American companies by market cap...they have a market value of over two-and-a-half trillion dollars...and if you take those five companies, essentially you could run them with no equity capital at all. None… Now that was not the case in the past." - Warren Buffett

This is getting a little disconnected, but what does commerce look like when every business is an information business, and growth no longer requires reinvestment?

Will there be an explosion of heretofore unimaginably profitable corporations, with one company per human need/desire?

• Google to find
• Amazon to make
• Uber to move
• Nike to perform
• abc to xyz...

Will it be one company that owns ALL energy, or is it one company per energy source? One company per geography?

Should one company own ALL transportation, or is it one company per transportation type? One per land mass?

Maybe the monopolies of tomorrow will be much more durable than anything of yesteryear.

Would this be a good thing? Peter Thiel believes that a monopoly in a dynamic environment is actually very positive for society...

Or maybe not.

Maybe the monopolies of the future will be even more fleeting, with each successive generation of AI surpassing its predecessor more quickly than the last, until we reach some physical bound of intelligence.

What does economics look like when the cost of producing new goods is driven down to the energy expenditure required to transform base elements? Are we going to have permanent negative interest rates?

Is the future “means of production” just dueling algos?

God From The Machine

I don’t look to jump over 7 foot bars, I look around for 1 foot bars that I can step over - Warren Buffett

Quick show of hands: how many of you think the internet is going to become LESS important over the next 5 years?

Who here uses a search engine other than Google for day-to-day?

Who thinks regulators will have any effect on big tech as a business?

I’m actually surprised Buffett hasn’t picked up any GOOG shares. It’s one of the three “toll road” internet businesses (In his 2018 letter to investors, Chamath Palahapatiya says SV startups spend almost 40 cents of every VC dollar on Google, Facebook, and Amazon) and has a ton of downside protection.

Even if we assume:

• The cloud computing business goes nowhere (possible)
• All of their overseas money gets stolen by foreign governments (unlikely)
• None of Alphabet’s “other bets” deliver (some already have)
• Ad revenue only grows with the overall market (not as a properly defined function of desktop, mobile, digital, etc.)

Then, relative to interest rates, investors are giving up an extra 4% return per year and a hedge against runaway artificial intelligence.

Looks like a 1 foot bar to me.

What am I missing?

CC