How a Bacteria Colony Outwitted Computers By Evolving
Вставка
- Опубліковано 25 сер 2022
- How in the world can living things become computers that outperform real computers?
Article referenced in this video:
Solving a Hamiltonian Path Problem with a bacterial computerjbioleng.biomedcentral.com › articles
Learn more about biology’s Turing completeness:
• The cell as a computer...
Turing machine photo:
en.wikipedia.org/wiki/Turing_...
Music:
City Life - Artificial.Music (No Copyright Music)
Link: • City Life - Artificial...
Pure Water by Meydän
Link: • Meydän - Pure Water [C...
Forever Sunrise - by Jonny Easton
Link: • Forever Sunrise - Soft...
I wouldn't say this is really is the bacteria comming up with some smart strategy so much as it's a demonstration of using bacteria to do parallel processing, which honestly is still quite interesting
I'll even add that bacteria allow for MORE parallel processing than computers, because concurrency and how they behave is just less formally structured. Algorithms play by different rules completely under this case.
It is like a GPU, the checking part of guess and check is very simple, so instead of using buff CPU cores that take too much space and care, you can have
a billion tiny cores that each do something simple but at the same time. It's a real good idea actually
@@magic8ball237 most of the biochemistry the multicellular organisms depend on was already invented (in some form) by the bacteria.
The foundations of cell differentiation and 3D organisation, too....
Once the basic body plan for radially symmetric organisms developed: a whole cascade of variations inevitably followed.
Gene duplication and modifications of the redundant genetic makeup broadened the recombinational "space". Cambrian whole genome duplication in the animal kingdom (repeated twice) gave the evolution whole new dimensions to experiment with...
The paper you are referencing is just trying to solve the issue showcased by Jack Parker for Leonard Alderman's solution to the Hamiltonian Graph problem using DNA by a different method. The problem the paper tries to solve is the speed required to find the correct solution. In Alderman's procedure, one would have to spend many days chemically "sorting" the already produced paths in order to find the one pertaining to the correct solution to the problem at hand, while here that chemical processes is replaced by the introduction of the bacteria. The underling problem of why this doesn't work didn't get fixed. The time complexity got traded off for space complexity, if the graph was more complex, had more nodes and/or more edges, then the amount of DNA (bacteria) is what increases at an extreme rate. This is what Jack Parker pointed out and while searching for a change in a bacterium is significantly faster, when there could be millions (or worse) of them and that search method is not going to be all that helpful. All of these "biological computations" basically just offload a large portion of the work onto the programmer and still have the massive downside of being horrendously difficult to construct as chemical systems do not lend themselves to re-programmability as physical systems do. Also, computers do this just fine, like I can pull out my phone and get google maps to heuristically solve the Traveling Salesman Problem and it does it pretty well. No sitting in a lab for days, no editing of DNA, and no searching for a glowing bacterium. Just a pretty accurate guess that will be more than enough for the vast majority of situations.
Of course that tradeoff (space for time is obvious).... And the complexity of selecting/scanning for dozens of targets may be unmanageable.
But there's still an option to split one gene into dozens of pieces to see if it can be correctly recombined.
The numbers of bacteria cultured in batches containing many liters of media can be enormous... And if the selective advantage of the target is great enough there has to be an achievable solution for some complexity. Although I don't know how precise the bacterial recombination can be: precision would be highly limiting factor for too little "slices" of a gene. Mutations in active parts of the protein can be easily deleterious....
Still I wonder how far this idea can be stretched.
you forget that many decades ago many people had VERY similar sentiments towards using technology versus one's navigation skills (such as map and compass) as the technology during that era was equally as unreliable assuming it still wasn't theoretical only. any new developed system or technology starts off as rough or even nearly non functional but as our methods and techniques develop as well as our understanding said technology will vastly improve until it eventually becomes viable. what i'm trying to say is that bringing up the fact that technology does the job just fine right now is really not very meaningful in the discussion. we should be talking about how this could be better or worse if it theoretically doesn't have the problems it does in it's current state versus just saying that it's horribly inefficient........we know.
Great video, and superb music for it!
Amazing. Thanks for sharing
It takes 15.15 seconds to calculate ok, but if you add the time to sequence back the DNA to read the path you get to probably longer time (But I agree that this would be only a large constant overhead)
I think he was saying that it takes on average 15.15 operations to calculate a correct path rather than the 1000 odd that the computer did not seconds
He mentioned seconds per operation in order to emphasize the difference.
Your channel has an incredible quality, what is your studies background?
The more I think about it the less I feel that I understand it.
Shouldn't traits have some kind of evolutionary advantage? If there's like a hundred traits that do nothing but indicate, then the bacteria that has all of them would occupy 1/100! of given space. It's 10^-157.
How do we find it? How do we differentiate it from all the bacteria that have 80/100 traits if they all are mixed together? They look practically the same and there's 20! times more bacteria that have 80 traits than those that have all traits.
If there is an evolutionary advantage in having all the traits, then after some time those bacteria that solved the problem will be in the majority. But in order to do this, wouldn't we need to already know the answer to the problem that we are trying to solve?
To answer that, some traits are actually very selectable ie. you can encode antibiotic resistance or even the ability to break down some specific molecule (see the paper in desc.). There are a lot of biological assays out there for a lot of traits (tests that screen for something biological)
You don’t need to know the solution for it all to work. This method of solving is basically shuffling around the paths until you get the right answer. How do you know if you have the right answer?
1. You’ve visited every node of the graph = every trait that you want is present and you can screen for all of them
2. You’ve only visited each node only once => Sometimes even if all the traits light up there is a possibility of it passing the same point twice. Once you’ve extracted the path from the bacteria via sequencing, you can just search for the path that has no duplicate nodes and there’s the solution.
You can screen for those traits one after the other when there were enough duplications (generations) to assume that the number of possible (re)combination events is enough to reshuffle the segments of interest in the correct manner.
How many generations of bacterial replication would you need is probably highly dependent on the traits: so this should be established experimentally.
You can pre-select for a lot of rather innocent proteins. A dozen different targets should not be very problematic. But as an ex-molecular biologist I would be curious how many could be managed practically.
I think a more viable option would be to clone (get the DNA material into bacteria) combinations of those gene segments randomly to get genetic libraries.
If those segments are analogous to paths between places, then there are lots of unnecessary ones that would make the number of possible combinations unnecessarily enormous.
Of course there are different ways to make genetic libraries, and they are easier said than done 🙃😐🙂
Can you do a booklist video so that we can go in depth
Need better volume; more! To hard to hear.
kylo: more..........MORE!!!!!!!!!
This Video is very well made!
but I have one problem with this, the bacteria arent really outwitting the computer, its just that there are way more bacteria than computers
if you had lets say a 100 computers each trying to find a solution on both the cpu and the gpu, the gpu being used for parallelization of the task similar to the bacteria, and the cpu just being used because why the hell not, it would easily outperform even a billion bacteria
saying the bacteria outperformed the computer is biased
regardless of my criticism of this, it was an excellent video, cheers
I did say the whole colony collectively outwitted the computer though haha
Well, but its not just that its a bit more, bacteria only need a little bit of power, and very few size, computers need more
To add to that, in labs, you’d just plate the bacteria on essentially sugar jello. The next day you come back, you’d be presented with a billion of these bacteria on the very same plate. Cost and execution wise, this is much more effective than setting up a hundred computers to do the same thing.
Regardless, to be a little bit more apprehensive, using bacteria like this comes with it’s own set of problems, mainly coming from the fact that living things are constantly evolving, hard to predict, and there are so many interacting parts that we don’t really know about yet. It’s kinda like trying to throw your own code in without knowing how everything else works.
But anyways, thanks for the comment, it did help me think more deeply about this topic!
Yea but i can just plomp some bacteria in my local pond and completly evicerate your gpubarray in time with trillions of bacteria.
@@randomsnow6510 Modifying a bacteria to do this is much simpler than building a computer from scratch. Manufacturing modern CPUs is EXTREMELY high tech, while making these bacteria is basically just following an elaborate cooking recipe.
This may also indicate that it is possible to reduce the time complexity of the code.
Not really, the bacteria aren't doing anything special, they are just brute forcing the problem, and it works because there is such a huge number of them.
The time complexity of doing that is much much higher than that of the current algorithms, its just more efficient when the bacteria do it because they can do it so many times in parallel.
Set up is confusing. It is not clear what is it about and why to follow the story. Hook with application.
This is a confused, obscure and not particularly helpful video. I suggest you look up slime moulds and suchlike.