Robot Scientist Adam makes discovery that has eluded human scientists for years. | Hacker News Reader

Robot Scientist Adam makes discovery that has eluded human scientists for years.

singularityhub.com

79 pointsbkudria35 comments

The headline is sexy, but I think it gets the main point wrong. The first sentence in the article that, I think, is the proper lead is: As artificial intelligence continues to evolve however, we are beginning to see the introduction of robotics in many “high-skill” fields such as research and medicine.

The point isn't that a robot was able to make a discovery that human researchers could not. The point is that biologists who do this kind of research will be able to work at a higher level of abstraction: something that used to require manual work on their part can now be automated. We don't replace the biology researchers, rather we've improved their tools.

If we assume the robots cost money , and research funds are limited , and the robot's research is important, there's no reason why it won't "replace" biologists.

I mean replace in the sense that we will still need biologists to do the science. The robot can not do science on its own, it's only a tool that scientists will use. It might replace pure lab-workers who just do grunt work. So it's possible undergrad bio students won't be able to use that as an "in" into a research group.

Also, biology labs are already expensive to equip. This development doesn't change that.

It designed what experiments to commence to solve the problem. No biology undergrad was able to do it so far for "orphan" yeast enzymes.

I think this is significant and that such automated experiment design and execution might replace some researchers.

The article did not say that it "designed" anything.

Actually it says that in the first few paragraphs; and the rest of the article heavily implies the same.

Reading more on the original sources and I think it's a bit of both worlds. The robot was given a set of initial data plus the problem it was facing and it iteratively created experiments - at each stage deciding which experiment to do next to improve it's "knowledge" and progress towards the goal.

From sciencemag.com [1]:

The basis of science is the hypothetico-deductive method and the recording of experiments in sufficient detail to enable reproducibility. We report the development of Robot Scientist "Adam," which advances the automation of both.

1. http://www.sciencemag.org/cgi/content/abstract/324/5923/85

I have access to the full article, and they do phrase it that the robot "designed" the experiments it carried about. But it didn't conjure these designs out of thin air. The specific experiments it performed are a result of the general rules given to it by software created by the human biologists. (Some of which was coded in Prolog, which I think is pretty darn nifty.)

The amount of information they had to provide the robot was significant - both in terms of raw data and in terms of the software controlling it. All of that requires humans. I doubt tenured professors will be the ones writing the software for such robots.

edit: This is what they cite as the software they used for the model: Philip G.K. Reiser, et al., Developing a Logical Model of Yeast Metabolism, http://www.ida.liu.se/ext/epa/cis/2001/024/tcover.html

> But it didn't conjure these designs out of thin air.

And human scientist obviously do?

Of course you had to give the robot required knowledge on the plate because it couldn't pull what it needed to know from library. After all, books are written in some incomprehensible human language.

The moment that robots will begin to understand human language people no longer will be able to call them just tools.

I think "design" is being used in a loose sense here. It's not designing an experiment the way that Rutherford designed experiments to probe the nature of an atom. What it's doing is more like selecting an independent variable gene X and a dependent variable enzyme Y, and selecting concentrations of X and reagent Z to mix together to maximize the expectation value of the statistical significance of the level of Y that is measured afterwards. That's using a combination of search algorithms and Bayesian learning to optimize the parameters within a well-defined framework. Then those parameters are hooked up to physical beakers to get more data for the next generation of data.

Don't get me wrong, this is an impressive technical achievement, and I think it will do a hell of a lot to speed up the advancement of science in the field of microbiology. My hat is off to these researchers because this little guy they build is totally awesome. I'm just quibbling about what the word "design" does and does not mean in this context. I agree that biologists will still have a role in research labs, and I'm glad that they (or at least a well-funded subset of them) will be freed from the more monotonous details of their trade so that they can work on the more creative, abstract, and "science-y" parts of it[1].

[1] I do not know enough about biology to know what exactly those parts are. I assume they exist, even though I do tend to have the "stamp-collecting" bias against biology[2].

[2] Quote from Ernest Rutherford, mentioned above: "All science is either physics or stamp collecting." I.e., biology is heavy on cataloguing and light on abstractions.

I think the way to determine if the word "design" should be used here , is to see if before this robot , the word design was used in this context. otherwise will be doing the same "it's not AI" fallacy all over again.

On the other hand this will make biologists more productive relative to other scientists. So maybe there will be more funds for biologists compared to other scientists.

The real point is that scientists (not just biologists) will be able to get more done period. We've been bringing computers and automation into science for a while now and it has made a big difference.

Sometimes you need to collect A LOT of data to be able to form a theory. Sometimes you just need to try all the possibilities.

A lot of science depends on that being done, and with automation now it can be done. When I say done, I don't mean it's just faster and cheaper, I mean it is FEASABLE when it wouldn't be with human labour.

This a great scientific tool, think microscope, it's in that ballpark.

The highest level of abstraction is lying on a beach while robots do all the work.

Unfortunately, I think the current socio-economic system can not allow that, even if it is the logical conclusion. Of course people would still do challenging tasks (or "work"), but not for material rewards.

I haven't read the paper, yet, but it seems as though the article really inflates the claims for the robot's cognitive achievement. There were a set of candidate genes to test as the cause of each enzyme. This was a matter of creating a set of deletion mutants and checking whether the mutants produced the enzymes, a laborious task which would be pretty easy to automate without artificial intelligence. It does seem to be a helpful automation, though.

Yeah from what the article describes it sounds like at is simplest form a nested for loop varing across a number of variables. I assume it is quite a bit more complex than that but artificial intelligence?

The perception of what is artificial intelligence is changing all the time. Playing chess is just looping (and not even quite a bit more complex), but it used to be artificial intelligence.

You're wrong. Don't assume, read the article and the papers it cites.

I think the key point is this: "Dr. King and colleagues gave Adam a database containing information on the enzymes, the chemicals and reagents to do the experiments, and access to the yeast cultures."

I imagine it would have taken a really, really smart robot about 300 years to figure that stuff out on its own. And only IF it had a reason to figure it out.

Not AI ... just another Expert System. Doing exactly what it's told to, efficiently.

Wrong. An expert system only draws conclusions from a knowledge base using deductive inference. This system also uses abductive inference and inductive inference. Read the paper.

Kudos to the team that properly defined the problem and employed the use of an expert system to solve it.

Not yet sentient as the title delicately alludes however. Were a Nobel Prize awarded it would still go to the team not the device.

Where are we on the Kurzweil timeline btw.

Where the nobel prize goes is an interesting question. let's say that some robot discovers an important discovery , that deserves the nobel. let's also say the methods of operation of the robot are standard practice at the time of building it. and the discovery came through luck and tons of robotic work. who should win the nobel? would the robot be considered by the nobel comity ?

That was the question posed by Hawking back in 2003(?). His assertion was that the device would have to be considered intelligent, albeit another form of intelligence, and possibly deserving of rights.

It does seem as though we're getting there.

The classic philosopher's version of this is the Chinese Room: http://en.wikipedia.org/wiki/Chinese_room

My naive understanding is that the intelligence is not in the room or the book, but in the person who wrote the book.

What? No. This has nothing to do with the Chinese Room argument. The Chinese Room argument is about refuting the claim that reasoning according to rules can ever be considered intelligence (i.e. refuting Strong AI).

My naive interpretation is that both of these situations involve assigning "intelligence" to a machine or system built by a human w/o said human's continued intervention.

I'm thinking about the systems, not the rules within the systems. If the "official" philosophy line is that it's all about those rules, well maybe that's why official philosophy is so damn confusing.

I just finished reading "On Intelligence" by Jeff Hawkins, let me recommend that as a really well written and lucid explanation of the meaning of intelligence. If Numenta can build systems that can replicate what a 2 year old can do given a 2 year old's memory and sensory capabilities, I'll be ready to say that Jeff's explanation wins, and that philosophers better find something else that's impossible to prove or disprove (like the existence of God) to go argue about.

With the caveat that opposing answers sound obvious to different people, this answer sounded obvious to me when I heard it: The entire system, as a whole, is intelligent; the separate components are not. Like Vizzini, Searle has made an error common even among the intelligent, and overlooked an important possible combination of location.

We seem to be less than 3 years behind schedule on the Kurzweil timeline. Everything he predicted for 2010 is either here right now or is just around the corner. One of major 'misses' was real-time translations of voice phonecalls, and both Microsoft and Google are launching this service very soon.

Neat. Once you have a very clear problem statement, you can write a program that can perform experiments and analyze the results. It's a step.

Very impressive, but as others have said: without AI at the helm, this is mostly a complex robot taking the place of a human technician. ( Which costs more? )

I suspect that much of the macro-scale manipulation performed by this robot (and humans) will be made obsolete in coming years by microfluidic analytic instruments.

While the robot is definitely impressive, without substantial AI in it, I dont think it qualifies to be called a 'scientist' just yet. It clearly cannot appreciate what its doing -- it might just as well have been cleaning the dishes.

As a former member of this research group, I can tell you are spouting nonsense and I think you had better run along and read the paper before making derogatory comments. It's just that people and media are far more interested in seeing robots than in seeing software.

FYI: I have spent the last 10 years working on algorithmic scientific method.

"And though some may be hesitant to accept the ever-increasing roles of robots in our world, I, for one, " welcome our new robot overlords.

(Sorry, but I couldn't resist that perfect segue.)

BTW The paper was published a year ago, so this is hardly news. And the previous significant article was in Nature in 2004.

Am I the only one who was more excited to see the link to Tron Legacy?

Reply on news.ycombinator.com