Osmosis Weekly #0002: A.I., Kanye & Vat-Grown Meat

Hey Osmotics,

Inside today's Osmosis Weekly...

• Featured Story: Artificial Intelligence
• Worth Checking Out: Launch Fast and Break Stuff
• Dumb Things to Share: BBQ Sauce & Fake Fish

In the Ted Chiang novella, Story of Your Life (1998) – later adapted by Denis Villeneuve into the hauntingly beautiful film Arrival (2016)  – aliens come to Earth and the human population decide to send... not a diplomat, not a world leader... but a linguist to open communications.

Only in science fiction would humanity be smart enough to send scientists to fix a global situation. (This is not, I repeat, not a commentary on how we've handled COVID-19).

What I do want to draw your attention to however – is the idea that one day we may have interspecies communication (say that three-times-fast)... and that day may actually be sooner than you think.

Alien Lifeform In Our Oceans?!

There's a Canadian marine-biologist right now –  who, for the last 13 years – has tracked and followed sperm whales in the Caribbean... recording their communications, unravelling their social network, and hoping... that one day... we might actually understand what all that clicking means.

Last Monday, CETI (that's Cetacean Translation Initiative, and not the other CETI, which stands for Communication with Extraterrestrial Intelligence. Which is really fucking hilarious if you think about it) partnered up with Dr. Shane Gero (the aforementioned biologist) and as a team of cryptographers, linguists, and machine learning experts... they've set a bold and audacious five-year goal of cracking the whale coda.

Which brings us to this week's feature topic...

Machine Learning: or How I Learned to Stop Worrying and Love Robots

In my summary of Deep Work, I included CGP Gray's short film Humans Need Not Apply (2014), which talks about how machines will eventually replace a shit ton of human jobs.

But how scared should we actually be of this future? Are we talking like The Matrix here... where there's a robot uprising... or a utopian future where everything is automated by artificial intelligence and nobody has to work?

The answer is probably somewhere in the middle, a lot more boring than science fiction, with a bunch of logical tradeoffs and unexpected externalities.

Let's start with some basic definitions first though.

What Exactly Is Artificial Intelligence?

First, here's what A.I. isn't: IBM's Deep Blue beating Garry Kasparov at chess in 1997. Why? Because Deep Blue wasn't "thinking" so much as calculating faster. (200 million board positions per second, if you care.)

What does that mean? Well, a computer program can be broken down into three basic elements:

INPUT ➔ FUNCTIONS ➔ OUTPUT

"Functions", as you may recall from math, is what you want numbers to do, like addition, subtraction, etc. For example, if the function is "multiplication" and you input 3 and 8, the output would be 24.

Here's the key: the function can never output anything other than what it was programmed to do. Sure, it can do it faster than us puny humans, but it will never do anything other that what we told it to do.

So what is artificial intelligence then? That's when the computer learns by writing its own functions. Machine learning, a subfield of A.I., is doing exactly just that.

The most common example given to explain machine learning involve cats. Yes, 😽😹🙀. Machine learning uses something called a "neural network" and it looks something like this:

INPUT ➔ BLACKBOX ➔ OUTPUT

For input, you feed it a bunch of pictures of things that are cats 😺 and things that are not cats 🐶🍆🪘. The neural network (often accused of being a blackbox) will run through a bunch of criteria like size, shape, color... and spit out outputs declaring whether the pictures are cats (or not).

Here's what's important: At first, the neural network is going to get a lot of those answers wrong. A human on the other side needs to tell the machine when it's right or wrong, like a teacher marking a spelling test. In other words, the neural network requires human training.

And over time, the neural network rewrites those functions by itself until it gives the correct answer most of the time.

Here's Where It Gets Cyberpunk

Humans are really good at fucking things up. And as you can see already... neural networks require us humans to first, feed it data and secondly, train it to give us the "correct" answers.

You can see where this is going...

On the data side, you can have bad data, too little data, or even corrupt data. Like blurry pictures of cats 😼 for example. Or worse, biased data. Let's say the guy feeding pictures of cats to the neural network only likes Siamese cats. So he feeds more Siamese cat pictures into the machine than other types of cats.

Well, now, that neural network will share the same biases as its hoo-man owner... and not recognize shorthairs, Bengals and Coons very well... because it doesn't have enough data.

On the training side, we run into the same problems. So, OK, a cat is a cat is a cat. But what if we're training the neural network to do something more nuanced and subtle... like say, recognizing hate speech, sexism or racism?

See where this can get very murky very quickly? If the people feeding data and training neural networks are all say, rich white dudes, we're going to have problems.

And if you've been wondering why the tech world freaked out last December when Google fired Timnit Gebru, the co-lead of their "ethical AI team"... you now know one of the four problems Gebru brought up.

This is as sci-fi tropey as it gets. It doesn't matter how technologically advanced we get as a race... or how far in space we travel... we're still going to bring our human biases with us. #HumansBeHumaning <- Please make this a thing.

Of course - one can be hopeful and imagine a world where we're super-careful about training neural networks this coming decade, right?

I'll let your varying levels of optimism/pessimism be the judge of whether that will happen or not. If anything, I just gave you a great debate topic for parties once we all get vaxxed.

So let's talk about the cool, positive stuff that A.I. can and is already doing. And for that, I highly recommend HBR's webinar: Human Plus Machine: Reimagining Work in the Age of AI, an interview with H. James Wilson on his book of the same title.

Humans + Robots = 6.5X More Productivity

Wilson argues that A.I. will support how humans work, not completely replace us. In fact, companies who rely on automation  alone have see just a 1.5X in productivity increase... whereas companies who redesign jobs to be human+machine partnerships see as much as 6.5X growth.

Using human+machine partnerships...

• A team of radiologists and pathologist bumped their cancer predictions from 96% to 99.5%....
• Rio Tinto is now mining remotely with humans controlling robots from a control center...
• Adidas is leveraging A.I. to make shoes for local markets...

Wilson goes on to claim that two-thirds of our children will work in jobs that aren't even invented yet due to the massive change A.I. will bring.

What are some of these jobs that's already happening? Wilson says there are six A.I.-related job categories every company should prepare for.

On the human-help-machine side, we have trainers, explainers, and sustainers. On the machine-help-human side, we have amplify, interact, and embody.

• Trainers teach AI to be cooperative, empathetic and even funny.
• Explainers analyze why AI makes certain recommendations.
• Sustainers are safety engineers who handle unintended consequences... like robots harming humans.

In return, machines will...

• Amplify what humans do best by handling the low-level work like transactions, predicting data, and iterating quickly... while humans have a monopoly on raw creation, leading, judging and improvising.
• Interact with equipment... so humans can predict and reduce unplanned downtime. Routine checklist work also disappears.
• And allow humans to fully embody their highest calling. There are now labs where robots do the grunt work of testing... allowing scientists to complete 400 times more experiments each week.

In other words – collaborating allows scientists to make 100 years of discoveries in a single year. In fact, the speed in which Moderna rolled out their vaccines was due to... machine learning.

Put another way – if we're to believe Wilson's research, the future of work is bright. With low level work like admin, logistics and operations handled by an A.I., workers can be more human and less robotic. Take that, Karl Marx!

Of course, as we discussed earlier... along the way, we'll need to make sure A.I. and their neural networks are trained in an ethical and equitable way. Otherwise, we'll just transfer our current inequalities to the machines... and further entrench them. +

Worth Checking Out...

This week: Lessons on "getting to MVP" (or go-to-market GTM) from three different "lazy launches" that moved fast and broke things.

10,000 Visitors on Day One

This indie hacker launched a web app and got 11K visitors on day one, here's how they did it...

• First, he had a very clear concept: "I don't need 100,000 things to watch, I need one great show/movie."
• Second, the idea was executed as soon as lockdown began last year, when we were all trapped at home with Netflix, Prime, and Hulu. AKA trend surfing.
• Finally, he launched it to a relevant subreddit (the right audience).

Failing SaaS To $24k MRR Turnaround

How do you turnaround a bleeding business? Start over.

• GrowSurf also launched fast using low code tools like Firebase, but they quickly ran into problems. No growth. Why?
• The product was built fast and shoddy. It had no clear target market. And as such, it attracted bad users.
• Once the founders decided to focus on B2B SaaS owners as their target, redesigned the app to serve them, and wrote copy that addressed their specific needs... everything turned around.

Kanye West Dating App

This might be an oldie for some, but worth re-reading. It's the story behind yeezy.dating, a dating app Harry Dry decided (on a whim) to make... but instead of writing a single line of code, he goes out and gets a ton of press first... to the point where he gets Kanye West's CFO on the phone.

I'm not going to do bullets for this one, because the story is full of hijinks and it's hilarious. If you haven't read it, please do.

KEY LESSONS from all three stories? Yes, move fast, but also... intention matters. The Watch Something web app and Yeezy.Dating platform both had very clear concept USPs, target audiences, and timing... whereas GrowSurf tried to be for anyone and everyone (at first).

Once GrowSurf defined their ICP (Ideal Customer Profile), they were able to say yes to their target market, and say no to everyone else, providing superior service.

Now, of course, longevity for the other two apps is questionable. It will come down to growth strategy, execution and traction. But you gotta give it to them. They nailed market validation first, got to MVP quickly and worried about everything else later.

Dumb Things to Share With Loved Ones

Because it's easier than talking about your feelings...

• I got vaxxed yesterday. 💉 I didn't get the cool Pfizer one, with the mRNA tech (like my wife did). I had to settle for Astra-Zeneca with ole skool dead viruses. Regardless, I got me some vaccinated attitude à la Marc Rebillet.
  
  👇 TURN THIS MUTHER UP!!! 👇

• Let's go down the NFT rabbit hole together. So I dug into the Instagram page Clarke Read runs. (He's the guy who gave us a great rundown of NFT marketplaces last week). This IG basically tracks what's sold and for how much. It's kinda cool. Leave Britney Alone! got $45K. But money aside, here are some I liked just for the art: 1, 2, 3, 4 & 5. And this one is just disturbing. Don't click on it...

• Let's go full circle. I started this issue with whales, let's end it with lab-grown salmon. This company in San Francisco (where else?) has created Coho salmon sashimi from cells in a lab. I mean, sure it takes them three-and-a-half-weeks to make one pound... but we're getting closer and closer to the day when your girlfriend might chastise you for wasting real meat (the good stuff) versus "vat-grown" á la William Gibson's Neuromancer.

Case sawed shakily at his steak, reducing it to uneaten bite-sized fragments, which he pushed around in the rich sauce, finally abandoning the whole thing.

"Jesus," Molly said, her own plate empty, "gimme that. You know what this costs?" She took his plate. "They gotta raise a whole animal for years and then they kill it. This isn't vat-stuff." She forked a mouthful up and chewed.

p. 132-133 Neuromancer (1984) by William Gibson

• OK, ok. One more. (I can't help myself...) Did you know Mark Zuckerberg is a huge fan of Sweet Baby Rays BBQ Sauce? Like yuge fan...

Coming Soon(ish)...

• Breakthrough Advertising (1966) by Eugene Schwartz : 0% First Draft
• Trusted Advisor (1998) by David H. Maister : 0% Raw Notes
• The Hard Things About Hard Things (2014) by Ben Horowitz : 30% Read

To suggest books, write me at colin@osmosis.dev
Osmosis+ members can vote on what books I should prioritize here.

Are you new here? Liked what you read? Click on "Learn by Osmosis >>" below and get Osmosis Weekly delivered to your inbox: