Humans evolved to interpret patterns in data – not for when data finds patterns in us.
At Bonsai, we improve business results by fixing poorly deployed technology. But for this post, I want to take a step back and discuss issues that arise even when tech is ‘working as intended’.
Are technologists driving towards a better future for everyone?
There’s been worry about feedback-loops and engagement algorithms for decades. I’ll never forget attending a forum at my old stomping grounds where product teams were demoing personalized Google search. A Nostradamus-like employee asked the question:
{Paraphrasing} “Personalized search sounds great, but are we worried about how much value Google search is providing if we tailor our results too tightly to what people want to find?”
Looking back, personalized search results are probably the least nefarious consequence of rapid ML-driven personalization. I’d argue the value far outweighs the social costs. The same cannot be said for much else. Social networks, content publishers, and every modern service with cookie-or-device tracked experiences takes user engagement data and turns those signals into systems that appeal to our subconscious and feed us instant gratification and information we can’t help but consume. The catch? It comes at the expense of our better judgment, which we’d otherwise exercise if not for readily available confirmation of our biases, emotions and beliefs.
The tragedy of this becomes more apparent every day. Where once only troubled minds could go, now healthy-minded, intelligent individuals have unwittingly followed. The pattern-finding nature of our brains mixes terribly with pattern-generating information systems . “How could someone believe this?” We often say when people embrace false claims, fabricated rhetoric and fake news. We blame capacity – we’ve all asked, “Are people just getting dumber?” But honestly, I don’t think it’s that at all. I fear the real issues are much darker.
Let me digress to my favorite domain – baseball.
If you take batting practice off of a pitching machine, you end up predicting the location and timing of the pitch. The dataset is homogenous; you get the same pitch, in the same spot, at the same speed. Eventually, even poor hitters start smacking line drives – maybe even home runs – with regularity. But every great hitter knows that hitting home runs off of a pitching machine doesn’t translate to hitting home runs during the game. In games, pitches come from all angles, speeds, and directions. Great hitters hone their skills with help from pitching machines, but mostly by practicing response to variation. No hitter has ever performed at a major league level without practice at-bats versus live pitching. Baseball fans know this as Spring Training.
Another feature of pitching machines is that you can tune their output to your specifications. Let’s say that you are really good at hitting an inside fastball – it’s in your “sweet spot”. Not hitting enough home runs in batting practice? No problem – you can tune the pitching machine to throw inside fastballs. Instantly you get the pitch in your “sweet spot”, and start hitting bombs.
What does this have to do with online personalization? Our current version optimizes everything around us like mini-pitching machines, constantly tweaked to our “sweet spot”. We all love “likes”, “shares”, and “engagement”. When the numbers go up, things must be getting better, right? And here’s our problem in a nutshell. We all love hitting home runs, but hitting them off of a pitching machine doesn’t make you good at baseball. Reading news tailored specifically to your interests won’t make you well informed. Seeing images you’ve ‘liked’ won’t allow you to see the full picture. And whether or not we know it, fulfillment and value comes from diverse experiences. When information comes to us through nature’s random lens, our brains free to use that innate human ability: diagnosing meaning and cause. Why aren’t we all smarter from today’s ML? We’re hitting home runs off of pitching machines – how many of us could hit one in an actual game?
Expand Your Worldview: Easy, Right?
I’ve got bad news for you here. I don’t think we are ever going back to a world where humans are the best at finding patterns. This machine learning stuff has lapped us and it’s never going back. So long as ML utilization continues unchecked by legal, ethical, or societal regulation, there’s simply too much money to be made capitalizing on the vulnerabilities of human desire to connect and find meaning behind information presented to us. Why is this direction so negative? ML has solved our patterns, but has no concept of meaning. Fantasia showed us the consequences of unchecked machine learning almost a hundred years ago: Mickey taught the broom to carry the water. The broom never knew it was supposed to simply fill the well.
Google’s ML application was a singular force for good because of it’s core design principle: “organize the world’s information and make it universally accessible and useful.”
That described Google in 2010. Think about what Google – and many other services – do for you now: “predict the answer you are looking for as quickly and accurately as possible.”
It used to be Google got everyone – no matter your age, gender, country of origin, income, means or mindset – the widest, best lens of information possible. From there, we found the answer. Now, before we even ask, Google just finds us the answer. And the more means you have – the more you can spend on internet speed, smartphone specs, or smart-home hardware, the more Google does for you. And they are not alone. This is true for Amazon. This is true for Facebook. This is true for eBay, Expedia, Apple, or any of the thousands of other applied ML companies dominating our economy.
The old version of ML democratized information, and enhanced our strengths as humans. I’d argue it made us exponentially smarter. This current version of ML let’s people of means pretend to be Babe Ruth. The problem is, they’re simply taking batting practice.
Should assistant ML be “rolled back”?
No. We must use and improve upon it. To do so, we need to stop working on speeding up tricks and start thinking a little deeper. We must acknowledge our systems aren’t yet smart. They certainly aren’t wise. Teaching the broom to carry water didn’t take much skill. It turns out even Mickey can predict clicks, likes, shares, or recognize your face in picture with enough data and a spell book. But Mickey was no sorcerer. Are our current tech giants evil companies, led by evil people? I’d argue absolutely not. I believe these leaders are simply apprentices: they really don’t understand the magic they’ve unlocked, and they certainly don’t grasp the purpose they should serve.
Are there good examples of what personalization could be? What we could rediscover? One of my favorites was Google Reader (RIP).
Why was Google Reader so amazing?
- It was curated to my settings, dependable, simple, but it didn’t constantly attempt to trap me in my own feedback-pattern-finding rabbit hole.
- It made life immensely easier! Instead of attempting to generate engagement or remove decisioning from my day, it eliminated clutter and allowed me to apply my brain without unnecessary distraction. I think it was a lot like Steve Jobs’ singular outfit – I didn’t have to worry about a dozen fonts, web styles, content formats or URL locations, I could simply read a lot, absorb, learn and explore.
Google Reader might have received slightly less pageviews per day than today’s “Google News”. It probably served a smaller pool of personalized ads per day. But the social capital Google Reader generated far made up for the short term dollar shortfall. A $50 billion dollar business could have been built with Google Reader. It might not have happened in 5 years, it might have taken 15 — but that concept could have been a success for 100 years or more. Did Google Reader have huge flaws? Absolutely! It was bad at allowing content creators to monetize their work. But if Google Reader were around today and solved content subscriptions in 8 years of iteration, I’d have at least 20 more subscriptions and a better perspective on the world than I do now.
I think solving responsible AI & personalization is the ultimate question facing the technology industry today.
In a world where anyone can access the world’s information, the vast majority of us have been drawn into worlds where we consume less truth, opting instead for the reality being built for us by machines skilled in confirming our biases. Once we begin the journey of applying wisdom to our magic, our tech will serve us so much better.
