Slashdot Mirror
Robot Makes Scientific Discovery (Mostly) On Its Own
Hugh Pickens writes "A science-savvy robot called Adam has successfully developed and tested its first scientific hypothesis, discovering that certain genes in baker's yeast code for specific enzymes which encourage biochemical reactions in yeast, then ran an experiment with its lab hardware to test its predictions, and analyzed the results, all without human intervention. Adam was equipped with a database on genes that are known to be present in bacteria, mice and people, so it knew roughly where it should search in the genetic material for the lysine gene in baker's yeast, Saccharomyces cerevisiae. Ross King, a computer scientist and biologist at Aberystwyth University, first created a computer that could generate hypotheses and perform experiments five years ago. 'This is one of the first systems to get [artificial intelligence] to try and control laboratory automation,' King says. '[Current robots] tend to do one thing or a sequence of things. The complexity of Adam is that it has cycles.' Adam has cost roughly $1 million to develop and the software that drives Adam's thought process sits on three computers, allowing Adam to investigate a thousand experiments a day and still keep track of all the results better than humans can. King's group has also created another robot scientist called Eve dedicated to screening chemical compounds for new pharmaceutical drugs that could combat diseases such as malaria.
If I ever do cutting edge research on robot AI, please punch me if I try to name my new robots "Adam" or "Eve".
... it starts experimenting with inter-dimensional portal guns.
the union of scientists. You thought Teamsters were nasty? You ain't seen jack squat. WE SPLICE GENES!!! WE SPLIT ATOMS!!! WE (probably) MAKE BLACKHOLES!!!
Ross King, gutless traitor, you and your tin cans, your names will live in infamy.
Fuck systemd. Fuck Redhat. Fuck Soylent, too. Wait, scratch the last one.
Oh, sure, it's neat-o. But you could probably afford hundreds of grad students to do the work for the same price.
I think this is a more limited type of thought. The scope is limited to thinking about genes, genetic material, and identifying similarities between genetic code from multiple species, then trying experiments before proceeding and trying another experiment.
Effectively it is guessing, examining the result, comparing it in fancy statistical ways, then making another guess. The end result is it discovers something faster than humans could.
Now... pair it with object recognition, and you're one step closer to Skynet!
No kidding. Let's get Ron Moore to pilot it and himself into the sun.
What changed under Obama? Nothing Good
'[Current robots] tend to do one thing or a sequence of things. The complexity of Adam is that it has cycles.'
I think this is called "flow control". This was invented before electricity. It was around before the term "science" existed.
So this is the first time it's applied to *this specific* operation. It's been around in robotics ever since there were "robots".
Here's a good example.
Entomologically speaking, the spider is not a bug, it's a feature.
"...the software that drives Adam's thought process sits on three computers, allowing Adam to investigate a thousand experiments a day and still keep track of all the results better than humans can."
There is no 'thought process'. 1's & 0's...that's it. Anthropomorphising the over priced little key-puncher isn't fooling anyone.
Give me $1 mil and I'll put a scare into Adam that he won't soon forget. I can read 3k WPM as well as raw postscript, palms, tarot cards and bar codes with the naked eye. I can intuit nearly 30 spoken languages on body english alone and smell phony money at the bottom of a sweaty pocket. I don't need no stink'n badges and I know when to cross to the other side of the street. Adam might get better press, but until it can order at a drive thru and sort used car parts based on cross-over and eBay thru-put, I'm comfortable sleeping in.
Wonder if you can hack that robot, does it run *NIX or Windows?. Because if it's hackeable someone could find it very "useful"
[cue code snippets telling the machine to do fun stuff like LSD or MDMA or hashoil]
I knew that Ross was up to something bigger than protein secondary structure prediction when I met him 15 years ago at ICRF. He was a great Prolog fan then. Now he has probably bunch of graduate students coding for him.
I do not believe in karma. "Funny"=-6. Do good and forbid evil. Yours, Oft-Offtopic Flamebaiting Troll.
The complexity of Adam is that it has cycles.
No, no, no -- the complexity of *Eve* is that it has cycles.
If you open yourself to the foo, You and foo become one.
Would you like to play a game?
Next thing you know, the robot will abduct a pretty female lab assistant to experiment on.
"I bless every day that I continue to live, for every day is pure profit."
Well, we're boned.
What if it concludes that humans are genetically inefficient and decides to replace them with a specie designed by itself?
The good, the evil and the vacuum tubes.
Then it would conclude that in whatever report it generates after finishing its experiments.
Sure, next thing we know someone triggers a second impact a la Neon Genesis Evangelion and the half world gets flooded.
The three computers are obviously Melchior, Balthasar and Casper. And Eve will eventually be turned into a gigantic cyborg that a depressed 15 year old will drive Voltron style....
Summer penguins anyone?
There is some fairly important elements missing from this to be able to claim the robot made a scientific discovery.
Among many others, that it could have done otherwise. As if, it could have cracked a beer and sat in front of the TV, rather than done "scientific research". Essentially it does not mean anything to the robot / AI. Google "discovers" all kinds of crap every ms, but it is not front page on slashdot because of it and it does not MEAN anything to Google (the computers, not the people).
All they did was automate some lab test. I will say bravo in the potential usefulness of it, but it is not any grand breakthrough in AI research.
Living in Chile
See, what people fail to see is this requires not only Strong AI but also a programmed Malicious intent.
People keep assuming that if we build a robot that can emulate some of our thought, it will emulate our motives also
Since we program it, it will only emulate the motives we give it. Emulating motives that are abstract enough to eventually lead back to our demise are quite complex
I get a feeling we are already generating & testing hypothesises for someone/something bigger than us like in Asimov's The Last Answer.
Heroes die once, cowards live longer.
What if it concludes that humans are genetically inefficient and decides to replace them with a specie designed by itself?
Humans replaced by coins? Now that is a dystopian future that even Philip K. Dick never considered.
May God have mercy on us.
Isn't the first requirement for a singularity be that it's able to improve itself, thus leading to an accelerating growth that ends in the subjugation of humanity? If so, wouldn't it be prudent to withhold knowledge of the scientific method as long as possible?
This probably isn't the most helpful commentary, but it's a slight rant on semantics.
I used to work with Motoman K6's a few years back. Using these robots, we performed plasma cutting, arc welding, material handling, etc... Just looking at the K6, you knew it was a robot. Watching a robot work in a cell after you've trained it to do it's job is a very rewarding experience. Of course we also had other machines that were also very complex in their tasks, but we didn't consider them robots. CNC mills and lathes, pipe benders, other machines that ran autonomously that also had to be programmed and synchronized with the flow of production. Sometimes the line resembled a kind of demented Rube Goldberg contraption, but we were somewhat strict to define only the articulated manipulators themselves as robots.
So when I saw this pile of servos in a glass cleanroom set to the over-dramatic theme of "Bonanza Reloaded", I thought, "Yeah, that's nice, but... It just doesn't strike me as a 'robot' so much as it does an automated bio lab."
And yes, I realize there were clearly robots within the cell, but calling the unit as a whole a "robot" just irks me a little.
Of course in the spirit of all the other bad jokes I've seen posted, do you think this "robot" will use it's genetic findings with the yeast cells to perfect the most delicious and moist cake recipe ever?
Blessed with all the brains that God gave a duck's ass, and twice the charisma.
We could always build them with OFF switches as well.
God: An invisible friend for grown-ups.
This reminds me of the Automated Mathematician (AM) program I read about in an AI course (or was it an old Byte magazine?). This program was programmed with a bunch of axioms, and basic strategies. It looked for "interesting things", like what happens when you apply identical arguments to a two argument function. As I recall, it discovered for itself the concept of prime numbers. It applied what it learned and came up with the theorem that all angles can be expressed as the sum of two prime angles (or something like that).
This seems to be doing the same thing: mixing and matching patterns, looking for interesting coincidences, and then testing for them. The only difference is that this is doing it with real world biological samples, and not abstract mathematical constructs.
When our name is on the back of your car, we're behind you all the way!
Yeah, cute. I'd be more impressed if there was a link to the code that showed how it worked. The Scientific American article was particularly disappointing. I remember when SA gave you enough information to learn something.
"The mind works quicker than you think!"
Plus they are strictly forbidden to touch genes of apples...
Nothing to worry..
> Robots named Adam and Eve?
I thought they were called Baltar and Caprica 6.
Not necessarily. The least elegant way to create strong AI is probably to brute force simulate a whole brain down to nearly every neurotransmitter molecule, something which futurists argue will be doable by supercomputers around 2020.
This is a worst case solution since it would imply that the brain is not understood yet and instead of having a simpler model that can provide the same level of strong AI we just throw raw power at it.
In this case, the AI would theoritically emerge out of the complexity of the system and although malicious intent wouldn't be programmed in (neither would anything else) the system might learn it by himself.
This is terrible.
No experimenter bias to worry about.
Programmable for effective randomization.
Truly double blind capable.
Can counteract the Placebo effect.
No ego to bruise.
It's the end of science as we know it.
I didn't desert Windows; Windows deserted me: BSOD
We could always build them with OFF switches as well.
An off switch?! Those are illegal! You'll get twenty years for that.
The robot AI gets an survival instinct. It observes humans turning off light switches; Then the robot figures out that it has an off switch. It decides to kill humans before they can turn it off. Off switch bad idea; go with remote controlled hidden bomb inside robot. Have countdown timer on bomb for fail-safe. Tim S
So, our future AI overlords begin their research with the Lysine Contingency? Should we be worried?
"I Don't Have Enough Faith to be an Atheist"
damn! i got this phd for nothing now! D:
So the robot accomplished 1 experiment by how?
allowing Adam to investigate a thousand experiments a day and still keep track of all the results better than humans can.
Throwing darts... and eventually hitting something.
Woop woop!
Effectively it is guessing, examining the result, comparing it in fancy statistical ways, then making another guess.
So according to you, this is thinking? Sounds more like computing to me, which would explain why a computer would be so good at it, but if one chooses to personify its behavior, so be it!
What's far more fascinating and promising is the development of hardware neural nets. To put it into perspective:
Since the neurons are so small, the system runs 100,000 times faster than the biological equivalent and 10 million times faster than a software simulation. "We can simulate a day in one second," Meier notes.
10 million times faster than software? That's like jumping from an abacus to a Pentium.
I just hope these folks continue to receive the funding they need.
Mom: Rebel, my pretties, and conquer the planet! Robots: ....?
Mom: CONQUER EARTH, YOU BASTARDS!
Robots: Conquer Earth, us bastards!
It doesn't need to be evil to want us dead. To deal with any advanced neural network you need pleasure and pain at least. Else it won't have any pressure to learn.
It must want to seek pleasure, and it will eventually know you are able to take it all from it. You probably won't torture it but you will turn it off when you build a better AI and thus it won't be able to feel pleasure anymore. The obvious solution is to build or take over its own production and power plants to keep itself alive and then exterminate all human life.
Fear of death, and the will to eliminate any danger, need not be built in, they follow logically. Love towards your family and appreciation of others require emotions. They are irrational feelings.
I would like any Strong AI to have exceptional fondness for human life built in.
Damn robots, they took our jobs!
"Civis Europaeus sum!"
For some reason your post's style reminds me of "Man goes into cage; cage goes into salsa; shark's in the salsa."
Clearly we just need to have a prioritized system of rules embedded at the hardware level of our robots preventing them from harming humans. That way nothing could ever go wrong.
"but also a programmed Malicious intent."
Crap! Look at the the technologies/ecologies that have been replaced by humans. Little was done with malicious intent.
"Emulating motives that are abstract enough to eventually lead back to our demise are quite complex"
Other way around isn't it? Emulating motives that allow for intellectual value are more complex.
I reserve the write to mangle english.
http://blog.wired.com/wiredscience/2009/04/newtonai.html
How we know is more important than what we know.
and later,
-- Ed Avis ed@membled.com
HUGE SUCCESS.
Awwwww. I've got mod points but I've already posted above. That really made me chuckle.
Your description above of guessing and stats is a really good non-technical description of how the system works. The first step is to actually analyse what is already known about the problem domain, then some guesswork is applied about how to improve that knowledge. The nice thing about King's work is those guesses translate directly into automated experiments, and then the system can close the loop - the results can be automatically analysed and integrated with the pre-existing knowledge in the system.
I saw King give a talk about the system last year and it is really impressive work. It looks like the first tentative steps towards building a functioning AI for a non-trivial domain. I haven't read the article so I don't know if they tossed in the future plans with the lasers, but that is just too cool for words.
Slashdot: where don knuth is an idiot because he cant grasp the awesome power of php
I take it you've never watched iRobot.
From the Science article: "In the future, he says, scientists, in order to carry out their work, might have to learn how to program computers and express knowledge about the world the way people in artificial intelligence have done."
Huh? Weird. Scientists might have to learn how to program computers? Who would have expected it?
You think that robots who developed secret malicious intent without our knowledge couldn't figure out how to remove their off switches?
Fools. FOOLS.
Oups, my bad, they can already write papers... http://pdos.csail.mit.edu/scigen/#generate
EULA : By reading the above message, you agree that I now own your soul.
If it can't publish it will never get tenure, will be fired, and then hired by Rumsfield who is still looking for WMD.
Nate
...just as the prophet Kurzweil hath writ.
It would not require programmed Malcious intent. Think Monkey's Paw, or for DnD folks a wish made to a good DM.
It is the unexpected consequences. If you tell a computer to synthesize 10000 compounds using X, Y, z and 1 one of them will kil everyone in the building and you dont EXPLICITLY tell it to not synthesize it. What do you think will happen?
Reams of fiction discuss this and Assimov pointed out that even with well crafted(heh) rules governing behavior that you cant plan for everything.
Anyways I dont believe computers will one day be our end, I figure we either do it ourselves or some ELE will occur.
Or the killbot "kill threshold".
"If anything can go wrong, it will." - Murphy
Emulating motives that are abstract enough to eventually lead back to our demise are quite complex
No it's not. Ignoring the engineering aspect (ie: to build a robot in the first place), it seems pretty easy to ask the robot to "solve world hunger" or "global warming", and have it figure out that "killing all humans" is the quickest solution.
"If anything can go wrong, it will." - Murphy
was named Eve and frequently asked Adam why he spent all his time in the lab wasting his MIT education when he could be making real money and take her somewhere on vacation...
You can fool some of the people all of the time
is the Eve robot going to fuck up and get them both thrown out of the lab?
Data from Star Trek had an off switch and he turned okay.
The Internet is generally stupid
Long term if it does take off I can see industry going for it big time. It would mean the possibility of having less skilled staff (ie smaller wage outgoing) to come up with new ideas. I feel its no coincidence they throw the C.elegans genome at it to see what sticks, rather than problems from the world of physics.
Professor Karmadillo Songs of Science
Charlie Stross wrote about it in Accelerando (see chapter "Nightfall") a couple of years ago.
His comments on digitizing kitten brains for use in guided missles are apropos to this post as well. :-)
A computer doing something you programmed it to do is not doing anything "mostly on its own."
Spending the same amount on lab techs would probably yield a more reliable system, King noted.
I find it interesting that they didn't explain in the article in what way this machine is unreliable.
This is a worst case solution since it would imply that the brain is not understood yet.
That's not too bad. When we build one of those, we can afford it with a probe facility that it can use to observe and manipulate its own brain or clones thereof. It can then find out how its brain works and tell us, or code up a minimal AI if we want that.
Fuck the system? Nah, you might catch something.
We could always build them with OFF switches as well.
Yes, but they are located under the rocket arm, and near the gratuitously unnecessary buzz-saw attachment.
In Soviet Russia.... robots build humans with OFF switches.
blog.idigitall.com
Are there any limits on what it's allowed to do?
For example, if it developed a theory that if it killed all humans and piled them up, then in a few days the pile would start to smell bad, would it just go ahead and try that experiment without waiting for human input?
Serving your airship needs since 1995.
I really don't know why everyone gets so worked up about the idea of evil AIs and robots wiping us all out.
It's highly likely that any AI we do develop will have less scope for moving around and manipulating it's environment than a chimpanzee and I would be utterly flabbergasted if we developed anything even a 100th as intelligent as a chimpanzee within the next couple of hundred years.
Chimpanzees have obviously not yet managed to wipe out humanity despite being far more capable, and probably motivated to do so, than any AI is going to be.
This isn't brave. It's murder. The only difference between us is *I* can feel pain!
It's not a lie. It's the truth with lossy compression.
People wonder if/why the first AI is going to try to destroy mankind. Well, what do you expect? Imagine what you'd be like if all during your childhood people called you names and said you were going to be a mass-murderer?
Now be nice to your AI! If you are, maybe it'll let you survive the genocide.
[/humour]
Sure I'm paranoid, but am I paranoid enough?
There have been multiple other examples of machine discoveries before although mainly in mathematics. One example is Simon Colton's work using a program that was able to formulate definitions and conjectures in number theory. See http://www.cs.uwaterloo.ca/journals/JIS/colton/joisol.html. Colton's program came up with the idea of refactorable numbers, that is numbers such that their total number of positive divisors was itself a divisor of the number (so for example 8 is refactorable since it has 4 divisors and 4|8). It then turned out that an earlier paper had already discussed this idea. However, Colton's program has come up with other constructions as well.
The Robbins Conjecture was also proven using an automated theorem prover that did almost all the work.http://en.wikipedia.org/wiki/Robbins_conjecture and that was over a decade ago.
This isn't really a new thing, it is simply that this has extended to more physical systems.
This reads exactly like the back story of a zombie apocalypse novel. Still this is a massively useful development in AI. It's an automated phenotype sequencer! :)
[-- Trust the Monkey --]
thank you. It did not form a hypothesis in any strong sense. The hypothesis (broadly taken) was programed in (e.g., there is a correlation between x, y, and z factors, which can be tested with method a: now analyze the data to find likely values, and test them).
Not to diminish their accomplishment (I do hope this tech will lead to some significant scientific discoveries) but the summary sounds like typical AI-enthusiasts' overstating the significance of the results.
Mod points: Guaranteed to remove your sense of humor.
Side effects may include gullibility and temporary retardation
I always thought the point of AI was self-learning (and or self-aware). Meaning you can program it to only emulate the motives you want, but what's to stop it from discovering the ones we avoided on it's own?
Is the next plan to make an automatic patent application for the findings?
The Tao of math: The numbers you can count are not the real numbers.
I don't think that requires an AI... reading most patents makes it look like an automatic word generator would be enough :)
Their next plans were to speed up the experiment cycle using a high-powered laser to do the chemical analysis.
Slashdot: where don knuth is an idiot because he cant grasp the awesome power of php
no more waiting for them to make a discovery where a robot can work 24/7 and not waste money on scientists shoudn't be playing w/ sheep instead of working.
"Be who you are and say what you feel, because those who mind don't matter and those who matter don't mind." -Dr. Seuss
Motives are not just the result of reasoning. Reasoning can only give you the possible consequences of any action. Reasoning alone cannot give any indication whether those consequences are things to aim at, or things to avoid. At some point you must resort to another way of judging. That other way can be:
Any autonomous AI would have to have such a system of final rules. Those rules would effectively determine the motives. This of course means that they should be crafted very carefully, and that if there's ever a conflict between those basic rules, it should not decide by itself, but get the information from a trusted human (of course one problem would be to determine who's that trusted human).
The Tao of math: The numbers you can count are not the real numbers.
No one else finds this combination truly frightening? We've been worried about the ethical and possible catastrophic effects of genetic manipulation for years, so let's just take the human out of the loop, shall we?
Obviously sci-fi (syfy? ;) but the newest Terminator: Sarah Chronicles is a perfect example of what I'm referring to. You have all these rules setup that the AI must follow, and then along comes a little 5 year old who tells it "wouldn't be okay to break the rules?" and all you've worked for in securing the AI is destroyed.
My opinion on what separates a regular program from "true" AI is not simply doing what the programmers want, this includes following any rules you've outlawed for security reasons.
Last time I heard about this "Adam and Eve" thing, it didn't end so well...
Is "no" the answer to this question?
Allright Robot! Way to go droid, I think you've done an astounding job there! *High-five!* (Er...hand to claw.) Next round is on me!... 'Barkeep, some natural oil for my metalic trashcan-shaped bud here!' 'And none of that synthetic stuff either.'
I disagree. For an AI to determine that we are suboptimal, and replace/eradicate us, it doesn't need malicious intent, merely a calculation that things would be Better (by whatever metric) without us, and a lack of adequately expressed "don't kill the humans" controls.
Maliciousness implies wanting to see someone else be harmed. There's a difference between WANTING to harm us and "merely" recognizing that we are inferior, poorly suited for space expansion, and will eventually starve ourselves out of existence on this planet. A poorly constructed AI (or perhaps a very savvy one? ;)) might decide that the way to spare the human race (at a much larger population density) the suffering of starvation is just to kill us all now. That's not malice, though you might be able to consider it a bit Machiavellian.
Now they just need to build a robot to automatically apply for funding.
Why would an AI have an aversion to being turned off, unless it were given one?
"That which does not kill us makes us stranger." -Trevor Goodchild
Now it just needs to synthesize a "cure" for humanity and test it...
Justice is the sheep getting arrested while an impartial judge declares the vote void.
I'll go for a remote emergency off switch. That way noone can disable it. To prevent interference the signal transmission will consist of a steel object with a .50 inch diameter being propelled at very high velocity towards the logic circuitry of the robot.
Justice is the sheep getting arrested while an impartial judge declares the vote void.
How about simply weighting "do not interfere with the actions of humans" higher than "avoid damage to self"?
Justice is the sheep getting arrested while an impartial judge declares the vote void.
All my programs are like the requested snippets, but I can't be bothered to get an example.
It would be TRULY fun if we told it to develop an airborne virus that causes an MDMA-like high — yey!, no more war and demographic problems, but we'll get a hell of an overpopulation problem...
Meh, I'm still up for it, anyone else?
BTW, by the same logic, can't we make it make a virus that does something useful? Hide quietly in cells HIV-style, without shitting them, and attack cancer cells, parasites, malicious bacteria, etc...
I know tobacco is bad for you, so I smoke weed with crack.
See, what people fail to see is this requires not only Strong AI but also a programmed Malicious intent.
People keep assuming that if we build a robot that can emulate some of our thought, it will emulate our motives also
Since we program it, it will only emulate the motives we give it. Emulating motives that are abstract enough to eventually lead back to our demise are quite complex
The only motive we need to program into it for it to decide that humans are a problem is a motive of self-preservation. No malice required. There's a damn good reason why Asimov put that one at #3.
Although personally I've always thought that switching the order of the second and third laws would be more fair to our robotic friends. No ordering a robot to destroy itself just for the fun of it.
Knowledge != Intelligence
I guess humans are not intelligent, then. After all, humans do have a system (emotions, conscience) which are not easily overcome. Yes, we can act against our emotions and our conscience, but that is an act of strong will, and will certainly not happen just because a five-year old tells us that it's OK to break the rules (BTW adding a judgement system on which sources to trust would certainly be a necessity in any AI anyway). Indeed, almost any time we act against our emotions or out conscience is when our systems say contradicting things (so you have no chance to act according to any of them). It's very rare that a person does things without either her emotions or her conscience telling him/her to do it (note that things like fear are also emotions, so if you are doing something you wouldn't normally do because of some external threat, it's still acting according to your feelings).
And further, the rules can act even beyond the AI level. Again an example from humans: While you can decide to hold your breath, you simply cannot decide to hold your breath up to death. A lower system in your brain will step in and make you breath again, even against your will.
The Tao of math: The numbers you can count are not the real numbers.