Skip to Content

Artificial Intelligence For Biology?

A new paper in PLoS Computational Biology is getting a lot of attention (which event, while not trying to be snarky about it, is not something that happens every day). Here’s the press release, which I can guarantee that most of the articles written about this work will be based on. That’s because the paper itself becomes heavy going after a bit – the authors (from Tufts) have applied machine learning to the various biochemical pathways involved in flatworm regeneration.
That in itself sounds somewhat interesting, but not likely to attract the attention of the newspapers. But here’s the claim being made for it:

An artificial intelligence system has for the first time reverse-engineered the regeneration mechanism of planaria–the small worms whose extraordinary power to regrow body parts has made them a research model in human regenerative medicine.
The discovery by Tufts University biologists presents the first model of regeneration discovered by a non-human intelligence and the first comprehensive model of planarian regeneration, which had eluded human scientists for over 100 years.

The “100 years” part is hyperbole, because it’s not like people have been doing a detailed mechanistic search for that amount of time. Biology wasn’t up to the job, as the earlier biologists well knew. But is the artificial intelligence part hyperbole, or not? As the many enzymes and other proteins involved in planarians have been worked out, it has definitely been a challenge to figure out what’s doing what to what else for which reasons, and when. (That’s the shortest description of pathway elucidation that I can come up with!) The questions about this work are (1) is the model proposed correct (or at least plausibly correct)? (2) Was it truly worked out by a computational process? And (3) does this process rise to the level of “artificial intelligence”?
We’ll take those in order. I’m actually willing to stipulate the first point, pending the planarian people. There are a lot of researchers in the regeneration field who will be able to render a more meaningful opinion than mine, and I’ll wait for them to weigh in. I can look at the proposed pathways and say things like “Yeah, beta-catenin would probably have to be involved, damn thing is everywhere. . .yeah, don’t see how you can leave Wnt out of it. . .” and other such useful comments, but that doesn’t help us much.
What about the second point? What the authors have done is apply evolutionary algorithms to a modeled version of the various pathways involved, and let it rip, rearranging and tweaking the orders and relationships until it recapitulates the experimental data. It is interesting that this process didn’t spit out a wooly Ptolemaic scheme full of epicycles and special pleading, but rather a reasonably streamlined account of what could be going on. The former is always what you have to guard against with machine-learning systems – overfitting. You can make any model work if you’re willing to accept sufficient wheels within wheels, but at some point you have to wonder if you’re optimizing towards reality.
How close is the proposed scheme to what people already might have been thinking (or might have already proposed themselves?) In other words, did we need a ghost come from the grave to tell us this? I am not up on the planarian stem-cell literature, but my impression is that this new model really is more comprehensive than anything that’s been proposed before. It provides testable hypotheses. For example, it interprets the results of some experiments as inferring the existence of (yet unknown) regulatory molecules and genes. (The authors present candidates for two of these, and I would guess that experimental evidence in this area will be coming soon).
It’s also important to note, as the authors do, that this model is not comprehensive. It only takes into account 2-D morphology, and has nothing to say about (for example) the arrangement of planarian internal organs. This, though, seems to be a matter of degree, only – if you’re willing to collect more data, code it up, and run the model for longer after doing some more coding on it, its successor should presumably be able to deal with this sort of thing.
And that brings us to point three: is this a discovery made via artificial intelligence? Here we get into the sticky swamp of defining intelligence, there to recognize the artificial variety. The arguments here have not ceased, and probably won’t cease until an AI hosts its own late-night talk show. Is the Siri software artificial intelligence? Are the directions you get from Google Maps? A search done through the chemical literature on SciFinder or the like? An earlier age would have probably answered “yes” (and an even earlier age would have fled in terror) but we’ve become more used to this sort of thing.
I think that one big problem in this area is that the word “intelligence” is often taken (consciously or not) to mean “human intelligence”. That doesn’t have to be true, but it does move the argument to whether border collies or African grey parrots demonstrate intelligence. (Personally, I think they do, just at a lower level and in different ways than humans). Is Google Maps as smart, in its own field, as a border collie? As a hamster? As a fire ant, or a planarian? Tough question, and part of the toughness is that we expect intelligence to be able to handle more than one particular problem. Ants are very good at what they do, but they seem to me clearly to be bundles of algorithms, and is a computer program any different, fundamentally? (Is a border collie merely a larger bundle of more complex algorithms? Are we? I will defer discussion of this disturbing question, because I see no way to answer it).
One of the hardest parts of the work in this current paper, I think, was the formalization step, where the existing phenomena from the experimental literature were coded into a computable framework. Now that took intelligence. Designing all the experiments (decades worth) that went into this hopper took quite a bit of it, too. Banging through it all, though, to come up with a model that fit the data, tweaking and prodding and adjusting and starting all over when it didn’t work – which is what the evolutionary algorithms did – takes something else: inhuman patience and focus. That’s what computers are really good at, relentless grinding. I can’t call it intelligence, and I can call it artificial intelligence only in the sense that an inflatable palm is an artificial tree. I realize that we do have to call it something, though, but the term “artificial intelligence” probably confuses more than it illuminates.

30 comments on “Artificial Intelligence For Biology?”

  1. John M says:

    I agree. This was accomplished with computational methods that are among those we refer to as “artificial intelligence”, but that’s is not the same thing as true artificial “intelligence”. Use of the term “non-human intelligence” in the press release is uncalled for; they are disingenuously implying they have advanced the capabilities of AI in some significant way.

  2. steve says:

    What they’ve done is possibly worked out the timing of the gene regulation involved in the process. That is not the same as working out the biology. It’s equivalent to saying that logging all of the times that workers come and leave will allow you to build a Boeing 747. The genes much be transcribed, modified post-transcriptionally, translated, modified post-translationally, secreted, form morphogenic gradients, have cells respond in a concentration-dependent manner, undergo morphogenic movements, etc, etc. Not to say that it’s not an accomplishment but gene regulation is far from the whole story.

  3. Frank Adrian says:

    There was a similar debate in the mathematics field when the Four Color Theorem was proved. Here, brute force was used to enumerate all possible solutions to the FCT and proved that all “reduced” configurations could be colored with four colors (if my fuzzy memory serves right). A similar search process was run here, using evolutionary search (due to the size of the space) rather than brute-force enumerative search.

    I would assume this algorithm found reasonable results as long as the evaluation function for the space was close to correct. It looks like a great paper to read (I’m interested in evolutionary algorithms) and, if it holds up biologically, another cap in the feathers of the algorithmists.

  4. Anonymous says:

    Agree w/ #2. When will we finally stop equating gene expression with things like protein expression and real time cellular physiology? Gene expression in many cases may not correlate with actual protein expression at all. Just do a PCR experiment followed by a western blot for many types of proteins. Proteins can contain multiple different sites for phosphorylation, and on top of that, each of those same phosphorylation sites can even be modified by another post-translational modifications with a sugar like O-GlcNAc, and proteins may have all sorts of other PTMs (300 of which are known) such as ubiquitination, hydroxylation, acetylation etc. etc. When you include the amount of PTMs that can occur on a single protein the sheer number of combinatorial possibilities, each of which may produce a different set of biology in theory, is absolutely mind blowing. Unfortunately none of the post-translationalome is coded within DNA. There’s a massive set of information outside of the genetic code that is controlled by an even more complex metabolic machine that is even harder to decipher. Gene expression alone =/= life.

  5. once and future academic says:

    What proportion of scientific papers now come with a press release? Has it become essential? I have published a few academic papers but never once put out a press release. Who is primarily responsible for the breathless press release that goes with this paper? PLoS, Tufts, or the authors of the paper?

  6. Molecular Geek says:

    This isn’t quite my specialty, but it’s not that far from what I’ve trained in. A cursory glance at it looks plausible, though I think I’m going to spend part of tomorrow afternoon poking at it more completely. The good news is that the length is partially due to some really good diagrams explaining their workflow. In my opinion, the biggest problem with these kinds of projects is right up front in figure 1: section c where they reference a functional ontology describing biological processes. There are several floating around out there, and none of them are particularly good at describing relationships between nodes in terms of underlying mechanisms. It doesn’t invalidate this paper, but it does point to where its weakness may be. To be fair, Tufts has some really good people in applied CS and Biology (or at least they did when I worked there), so this could well be the real deal. And the press release is a product of the PLOS publicity machine.

  7. steve says:

    Of course, if it weren’t for the PR I doubt very much that we’d be discussing it.

  8. Matt says:

    In the popular imagination AI is supposed to be some amazing, transcendent thing imagined but never realized. So in 1960 educated people generally “knew” that a world champion chess computer was really AI and by 2010 they generally “knew” that it wasn’t. Deep Blue never In the field’s own literature you see that search is a classic AI problem, and indeed that’s what made Google a $300 billion AI company. But hardly anyone actually calls Google an AI company because most people now “know” that AI is supposed to be a little person or demon in a box who teaches us valuable lessons about morality, prejudice, or Faustian bargains. It’s not supposed to be a program that uncomplainingly and tirelessly works to get you to click on ads. That’s just too mundane to be believed. Everyone will keep staring AI in the face and asking “where is it?” long after I’m dead.

  9. Anonymous says:

    It seems a little pointless to take issue with the term “Artificial Intelligence” as it is a recognised term for a branch of technology. It always seemed to be a bit old fashioned to me, with terms like “machine learning” being preferred for a while, but it seems to have come back around again.
    I agree that “non-human intelligence” is a bit disingenuous.
    Regardless, looks like an interesting piece of work.

  10. John Wayne says:

    “… most people now “know” that AI is supposed to be a little person or demon in a box who teaches us valuable lessons about morality, prejudice, or Faustian bargains. It’s not supposed to be a program that uncomplainingly and tirelessly works to get you to click on ads.”
    Well, this made me laugh out loud for real. Matt gets a cookie.

  11. ChristianR says:

    “Artificial intelligence has the same relation to intelligence as artificial flowers have to flowers.” — David Parnas

  12. Kshitij says:

    What they have is a combinatorial search algorithm on a defined search space, which is hardly intelligence, artificial or otherwise.
    An alternative to an evolutionary algorithm is: try every single arrangement of the pathway components, and pick the one that matches experiment best. That would give you the best pathway given the search parameters they have defined.
    An evolutionary algorithm is simply a heuristic for finding an approximate answer to the search problem they have defined. That’s an implementation detail, an algorithmic choice, and one that comes with compromises.
    If you had a computer fast enough you would want to run the full, brute-force search, because it’s guaranteed to give you the optimal answer, given the search problem they have defined.
    Doesn’t sound like artificial intelligence any more, does it?
    Like Derek says, ALL the intelligence here is in defining the specific search space and objective function. And they have provided no automated mechanism to define search spaces and objective functions on arbitrary scientific problems. THAT would be artificial intelligence.

  13. Anon says:

    I wonder how good google would be at chess, if you framed the search so it would return “next move after e2-e4”, etc. It should have essentially every game every played and recorded in its index somewhere…

  14. Anon says:

    I wonder how good google would be at chess, if you framed the search so it would return “next move after e2-e4”, etc. It should have essentially every game every played and recorded in its index somewhere…

  15. Matt says:

    Those “researcher” charlatans in are using boring old algorithms again?* Everyone knows you get real artificial intelligence by sticking glowing alien artifacts in a robot chassis. Or when lightning strikes your computer. I might accept an algorithm if it was written by a scientist bitten by a radioactive owl.
    The important thing is that AI should be appear to be an effect without causes, ineffable, as limpid as the luminiferous ether and sublime as mine own élan vital. AI may answer some questions faster and more accurately than any human, but it lacks the holistic biodynamics of analog thought. It never works well. When it does work well it’s cheating. The machines can take our jobs but they can never take our overconfidence.
    *In the future I intend to hold pharmaceutical research to the standards of comic books also. Where’s my all purpose regeneration serum, you layabouts?

  16. Anonymous says:

    “the first comprehensive model of planarian regeneration, which had eluded human scientists for over 100 years”.
    That’s nothing. The fact that the human brain is connected to the lymphatic system has escaped discovery since … well, since the human brain has been connected to the lymphatic system, about 120 million years ago.
    The complete hyperbole of the PR for something so trivial, combined with its basic ignorance, pisses me off no end.
    Just keep PR out of science!

  17. Morten G says:

    @16 Matt
    However, sticking a glowing alien artifact in a robot chassis highlights some interesting problems about self-aware computers. The “singularity”. How do you prevent them from going mad? We have had millions of years of evolution to add biological methods and distractions to our programming. How do you keep a self-aware AI from going mad? Even our current AIs go off the rails again and again.

  18. Charlie Kilian says:

    Others have made similar points, but I want to reiterate that “artificial intelligence” is the name for a particular category of algorithms in the field of computer science. It’s a name that I’ll freely admit is not very enlightening, and it certainly lends itself to sensational press releases such as this. Historically, they were actually early attempts at computers doing human-like intelligence, but that is not what real CS practitioners think of anymore when they talk about AI algorithms, and it hasn’t been for decades.
    As Phil Karlton supposedly said, there are only two hard problems in computer science: cache invalidation and naming things. If I’ve learned anything reading this blog, it’s that the latter problem isn’t unique to CS.

  19. steve says:

    Say what you will about AI but has anyone actually seen Derek in person? My understanding is that he is really an AI program designed to provoke interactions among chemists who rarely get out of the lab.

  20. Anonymous BMS Researcher says:

    @20 Steve: I have met Derek in person; he looks pretty much like the pictures on his blog.

  21. Matt says:

    @Morten G: I am drifting ever further from the original topic, but I don’t think the Singularity is a thing that is actually worth worrying about. People don’t know how to begin making a computer that apparently has volition independent of its makers, the sort of fictional AI that yearns for freedom and resents its human masters. Not only is it very difficult, there’s very little upside to a self-aware opinionated machine for most applications. People want e.g. machines that can transcribe speech to text, win poker tournaments, drive between points in chaotic urban environments, match passport photos against security camera recordings. There’s no reason to try to build in self awareness for most AI goals and I don’t think it will spontaneously emerge as a side effect either.

  22. steve says:

    @21 – That’s assuming that you’re not just part of the same program… 😉

  23. Matthew K says:

    For all current real-world applications, :”AI” = search. Massively powerful, such large scale search in such large parameter spaces that the results can be surprising, but search.

  24. steve says:

    I know next to nothing about AI but always thought that the idea was that a computer program would learn one set of rules and then apply them to an unrelated situation, learn from the results of that situation and come up with new rules for a new situation. To me that’s the sign of intelligence; has any program been able to achieve that goal?

  25. Matt says:

    I know next to nothing about AI but always thought that the idea was that a computer program would learn one set of rules and then apply them to an unrelated situation, learn from the results of that situation and come up with new rules for a new situation. To me that’s the sign of intelligence; has any program been able to achieve that goal?
    That’s one kind of AI approach. As to whether it’s been achieved, the answer depends on how much unrelatedness you require. On the Imagenet Large Scale Visual Recognition Challenge, Google, Microsoft, and Baidu all have systems that are nearly equal to or slightly beyond human ability for labeling unknown images after having trained from a set of known images:
    I think that’s a pretty good example, personally. You may think it doesn’t show enough versatility. The same program can’t find puns in text or the next best move in a chess game.
    Over the last 20 years a pattern I’ve observed is people saying that they’ll believe AI works when it does something beyond the current state of the art. When AI reaches the next level that some people previously endorsed as their threshold of acceptance, the challenge gets revised to say they’ll believe AI really works when it does the next thing beyond the current state of the art. Or the challenge gets revised to clarify that the problem is technically solved but the solution wasn’t exciting enough. “Ugh – don’t you ever get tired of solving problems via searching??” They’re yearning for magic, and categorically won’t find it in any thing made tangible.

  26. Fred the Fourth says:

    Charlie Kilian:
    “Invalidation and naming”
    Too right. I once told a product manager at Adobe (back in ’94) that the system I was working on had so many interacting caches that it had achieved intelligence: namely, the inability to completely forget anything.

  27. steve says:

    @26 – Maybe they’re all looking for a successful Turing test.

  28. metaphysician says:

    Personally, I figure there are two good benchmarks for spotting whether someone actually succeeded in creating a “true” AI:
    1. The computer comes forth and engages you in a debate about the issue of whether it counts as an AI, possibly as related to demands for better working conditions or higher pay.
    2. The computer takes over the world.

  29. Danny Sichel says:

    The debates about this on rec.arts.sf.written were so much fun.
    “By 2000 we were supposed to have computers bright enough to argue with us, but that doesn’t mean the way Word does it.” — Jo Walton

Comments are closed.