Machine learning and the end of science?
Deep learning and machine learning are going to be huge, right?
Everyone can see it coming now. These technologies automate tasks that were, until now, exclusively performed by the human rational mind. Any task that has a well defined, discrete set of actions is ripe for takeover by machine learning. Examples of these actions are numerous, and include:
- Move this game piece there.
- Recommend this product/media/advertisement to this user.
- Steer left 20 degrees and accelerate to 30 MPH.
- Sell 30% of shares at $60.
- Give this treatment to this patient.
- Answer this question.
- Investigate this individual for criminal activity.
So yeah, it’s going to affect every industry, and every person. The future belongs to our new robot overlords, and organizations will work more efficiently with less human effort. Well, once the systems are perfected…
I’ve been wondering a bit about what Automating all the Tasks via machine learning means to science and research, and it’s a bit worrisome — and not in a “They took our jobs!” sort of way, or a “They’re going to be racist and/or hurt someone” sort of way. Those are both worthy topics, of course.
Consider a simple bacterium. It has a genome that contains an exquisite set of instructions for helping that little guy:
1) Find energy, and
2) Use it to reproduce
(this is the Meaning of Life, by the way)
A bacterium doesn’t think about how it finds energy, or how it is going to reproduce, it just does. It identifies signals in its environment, and responds automatically (with perhaps some stochastic “randomness”).
We humans aren’t so different, fundamentally, but we do possess the extraordinary ability to perform rational thought and do mental abstraction. By thinking, we created civilization and machines which, in turn allow us to find more energy and use it to reproduce.
It’s now 2016 — or 1459691119 in machine time. The civilization and machines we’ve created for finding energy and using it to reproduce are so efficient — we don’t need to dedicate much of our brainpower for those tasks anymore. So we have civilization and machines to entertain us while we wait.
Now back to machine learning.
Deep learning and the like are the equivalent of digital bacteria, not digital humans. They don’t rationalize like we do — they just adapt to new challenges (with perhaps some stochastic “randomness”), given a set of inputs and potential actions.
And they aren’t organically evolved in a complex ecosystem like bacteria — machine learning algorithms are artificially selected in a sterile environment to perform only a specific task. So it’s hard to see how they will ever be able to truly think for themselves win a wider context.
How on Earth are machine learning algorithms going to discover new things in the universe? Aside from being really entertaining, the discovery of new things in the universe is very useful for helping us humans find energy and use it to reproduce.
For example — the algorithm that automatically identifies the best therapy for a single cancer patient isn’t capable of discovering a better therapy.
And a product recommendation engine isn’t capable of inventing a new, better product.
Machine learning algorithms can’t do science by themselves — and they can’t invent — but they will certainly help humans out by doing much of the heavy lifting. We should start talking about how humans and data-processing algorithms can work together.
