The Marketing Analytics Intersect
 

Today, an incredible development that will push your thinking super-hard and super far.

I am insanely excited about today's future-oriented topic, and some of you might be a little freaked out about: Super intelligence

Wikipedia defines superintelligence as: A hypothetical agent that possesses intelligence far surpassing that of the brightest and most gifted human minds.

Thus far, superintelligence has been considered in the context of problem-solving systems, however, my imagination of it is quite a bit more expansive.

I think of it as it is described in this wonderful New York Times article about a chess playing program: AlphaZero had the finesse of a virtuoso and the power of a machine. It was humankind’s first glimpse of an awesome new kind of intelligence.

An awesome and new kind because...

Thus far chess playing programs defended like iron. They are far stronger than any human player, but they have no real understanding of the game. They are tremendously fast and strong, but utterly lack insight.

AlphaZero is smarter, not just faster. It seems to express insight. It plays intuitively and beautifully, with a romantic, attacking style. It played gambits and took risks. In some games it paralyzed its opponents and toyed with them.

Let that sink in.

In an amazing development, AlphaZero did not learn by using classified data (i.e. learning from all the games played in the past by humans or machines).

The only input it was given were the rules of the game. It played against itself (millions of times) to become smarter than any human in just a matter of hours. Soon after that, it was the most intelligent entity on the planet at the game of chess.

This, by the way, happened a year ago.

Let that sink in.

***

Steven Strogatz, who wrote the NYT story, hypothesizes that a future evolution of AlphaZero, dubbed AlphaInfinity, could possess intelligence at a level that humans, try as they might, won't be able to fathom.

In my keynotes on Machine Learning I use a metaphor: It will be like a squirrel trying to understand how humans designed an aircraft engine to achieve flight.

A superintelligence like Alphainfinity is not in sight yet. Still, ponder the possibility that such a superintelligence could - I dare say easily - solve all our problems from healthcare to global warming to flight schedules. We would just sit back, with our acorns :), and marvel at this incomprehensible intelligence.

Mr. Strogatz's last paragraph is deeply worth internalizing, please read it.

***

I'm sure your immediate thought is: Wait. What happens to us? To work? To politics? To gossip? To building places of worship? To solving scientific problems?

In TMAI #100, I took a stab at predicting what happens to humanity in a hundred or so years.

14 months later, I remain optimistic and believe there are better than even odds that we will transition to a dramatically different species living a dramatically different existence for the essential reason we are here (joy).

***

Anything that is this emotionally and mentally challenging will have nay-sayers.

I welcome nay-sayers with a warm embrace. I believe nay-sayers are singularly responsible for all progress mankind makes - there is always someone who will take up the challenge of proving them wrong! :)

Among things they point out, here are three frequent ones:

1. God made humans, therefore machines will never be as intelligent.

"As intelligent" will certainly be proven untrue.

There is an interesting nuance in this assertion: The type of intelligent.

I am not claiming that machines will be human-like intelligent (based on my research I believe that being human-like bring big limitations).

Consider this though: If you want task x done, and the machine can do that task more efficiently and effectively than a human... Does it really matter what kind of intelligent it is?

Such a superintelligence is some ways away, but it will be able to complete tasks like running transportation systems, growing crops, telling you want you need before you need it, writing better stories, and even truly falling in love.

The type of intelligence used to get there will be different than our (human) intelligence - and far, far superior on a multitude of dimensions.

2. Are you on crack? There is not enough data for machines to get smart!

Two things.

First, you are right that in order for certain types of learning to occur we will need a lot more data. A billion trillion new sensors deployed strategically to collect all the data about weather. A billion of tiny nano-sensors floating in our bodies and minds will collect medical and psychological data about us - down to an atomic level. All the data in every transportation system - not just all the engine and speed data - but sensors embedded in the skin-structure of the vehicle will collect stress and other data. None of these three examples I shared exist today. But, do you honestly think that in 2019 these are not solvable problems? All three are currently being worked on. Your brain activity is just electrical impulses.

Second, the beauty of AlphaZero is that it does not need classified data. It just needs the rules. Life is not as simple as Chess - nor that structured. But, starting small and growing big this is simply a matter of focus. We don't know the rules for true love. I believe we will get there when the incentives are so big.

3. The human brain is a gazillion trillion computations at super low power and superintelligence will require chazimilion mamilian dapromilion amount of xyz.

True.

But consider the power in a single Tensor Processing Unit (perhaps the world's most powerful single purpose computing device). Then consider that between May 2016 and May 2018 the processors are a multitude times faster with a multitude more chips giving incredible performance boosts. This will continue at that scale for a while.

Then, there's the fast approaching Quantum Supremacy - the ability of quantum computing devices to solve problems classical computing practically cannot. An unimaginable explosion of computational power - and the that problems we think that power can solve.

Jürgen Schmidhuber predicts it will take 20 - 25 years to have a recurrent neural network comparable with the human brain. It is not that far away.

***

There are, of course, many other nays being said. So many of them are valid, and for a whole bunch others we don't have an imaginable answer.

The trends, though, all point in the direction of humanity having to co-exist with an intelligence (or many intelli, is that a word?) that will have what some people today call God-like, or omniscient, powers.

Bottom-line: I do not believe in wishing for things to not come true. I believe in planning. :) So here's one concrete thing I'm planning for myself: Constantly keep learning so that I can have different careers. Here's one thing I'm planning for my kids: Give them the broadest set of skills and a focused collection of values. For all of us, with the goal of maximizing survivability and thriveability.

What's your plan?

Avinash.

 
 
Powered by Mad Mimi®A GoDaddy® company