A sample text widget

Etiam pulvinar consectetur dolor sed malesuada. Ut convallis euismod dolor nec pretium. Nunc ut tristique massa.

Nam sodales mi vitae dolor ullamcorper et vulputate enim accumsan. Morbi orci magna, tincidunt vitae molestie nec, molestie at mi. Nulla nulla lorem, suscipit in posuere in, interdum non magna.

Automating Science

This past week saw two related news items that I think could lead to game changing tools for scientific discovery.

The first is a computer program that examined devices like a double pendulum (pivot point is in the middle) and in a day deduced laws of motion that took scientists centuries to determine.

The program was created by Cornell University computational researcher Hod Lipson and computational biologist Michael Schmidt and was described in a paper published recently in Science.

Hod Lipson said:
“One of the biggest problems in science today is moving forward and finding the underlying principles in areas where there is lots and lots of data, but there’s a theoretical gap. We don’t know how things work…I think this is going to be an important tool.”

What this tool isn’t is some sort of expert system using an a priori model. That would be too easy . It starts with only some very simple building blocks like near-random groups of processes like addition, subtraction, division, and some algebraic operators. These then interact with the dataset itself.

The output consists of equations that attempt to explain the data. Initially these are all usually totally wrong but some are less wrong than others. Then, in typical genetic algorithm style, these best equations are modified and tested. This cycle repeats over and over evolving the equations until they can duplicate the dataset or even predict future states.

Lipson said that…
“If you just look at the data plainly, it’s difficult to see if there’s anything systematic going on there…But despite that fact, when the algorithm analyzed that data, it could see laws that we know are correct.”

The scientists then did something very important. They fed the algorithm random numbers…what do you think happened?…The computer correctly found nothing. How awesome a test is that? That’s one of the distinguishing characteristics of real scientists. Try to shoot down your own shit. Do you think any of those bible code people ran their programs against random letter combinations?

What future does this have?

I think we’ll see scientists increasingly partnering with programs like this to evaluate datasets that are simply too complex for humans to deal with.

University of Michigan computer scientist Martha Pollack, said that there’s a potential “to apply [this] to any type of dynamical system.” This includes environmental systems, weather patterns, population genetics, cosmology and oceanography etc etc.

“Just about any natural science has the type of structure that would be amenable,” she said.

On the other hand…University of Minnesota cognitive neuroscientist Michael Atherton says, “the creativity, expertise, and the recognition of importance is still dependent on human judgment. The main problem remains the same: how to codify a complex frame of reference.”

Lipson expressed the same sentiment when he said that:
“In the end, we still need a scientist to look at this and say, this is interesting,”

So, yes; Humans are still important. The real question is…For how long?

There was another similar news item recently. Imagine if you gave the program I just described arms and the ability to actually design and run different experiments.

That’s pretty much what lead scientist Ross King and other scientists at the United Kingdom’s Cambridge and Aberystwyth universities have done.

They claim that their robotic system, called Adam, is the very first to make a unique scientific discovery on its own without and real human intellectual input.
Specifically, this means that Adam is a little microcosm of the scientific process. It (or is it he) can actually:
•    Formulate hypotheses
•    Design and run experiments
•    Analyze the data
•    Determine which experiments to run next
How cool is that?
Elements of Adam have been automated before like genome sequencers and drug screeners. These still needed people to make a hypothesis, design the experiment, and draw conclusions. The software program I talked about above could hypothesize and make conclusions but it could not setup the pendulum and make it move. It can’t interact with its environment. Adam is the first to close the loop, to hypothesize, experiment, and re-hypothesize without people.

So what did Adam do exactly?
They gave him a model of yeast metabolism and a database of genes and proteins that other organisms use for their metabolism. They then unleashed him to see what he could find out about the 600 or 700 yeast genes that we have no clue about.

He came up with a hypothesis and designed experiments to test it using his fully automated lab of centrifuges, incubators, growth analyzers etc. The only thing humans did was refill some containers and remove some waste.  Adam then performed some modified experiments as follow-up and eventually discovered three yeast genes that encoded for a novel enzyme. This was then confirmed by hand by the human lab slaves…I mean scientists.

For the near future, I think we’ll see these incredibly productive partnerships of scientists and robots.

According to King:
“Robots will be doing more and more of actual experimental work and simple cycles of hypothesis generation. Humans would migrate to more strategic and creative positions. How can we waste trained post-docs by making them pipette things in labs? It’s crazy.”

His next project’s name is…..you guessed it…Eve.
Eve will apply artificial intelligence to automated drug design and testing. Currently this is a brute force game. Just test as many compounds on your list as possible. Eve will select which compounds to make and test which could greatly speed up the discovery of useful drugs.

What about deeper in future?
If history is any guide, I think these automated research bots will increasingly take over areas once considered the domain of humans only.

“There isn’t any intrinsic reason why that wouldn’t happen,” says King. “I think there’s a continuum between the really basic types of science that you’d get from Adam, and the things I can do, and then Einstein-type science. A computer can make beautiful chess moves, but it’s not doing anything special. It’s just doing more of the same thing. In my view that’s what’s going to happen in science.”

3 comments to Automating Science

  • larry coon

    Bob, you said they fed random inputs to the system and (expectedly) found nothing. Do the people at Cornell see any potential application to cryptanalysis? You have data that appear to be -nearly- random (degree of entropy of 99.99-plus percent), but aren’t random at all. I’d imagine that this sort of application simply isn’t amenable to a heuristic algorithm, but what do they think?

  • John Powell

    The Singularity draws nearer…

  • Very cool, Bob. I had previously done a piece on Adam, but your post adds some interesting new details. I wonder what implications this could have for pseudoscience as well as science: an automated paranormal researcher with no belief or skepticism about psi…There’s no way that anyone could invoke the experimenter effect special plead. I’ll admit it wouldn’t be too popular for long with its organic colleagues, as it wouldn’t try to cover its ass over negative test results by employing flawed statistics and logical fallacies as the humans would, likely making its use in that field short-lived. Sigh…

Leave a Reply