Hacker News new | comments | show | ask | jobs | submit login
The Beautiful Intelligence of Bacteria and Other Microbes (quantamagazine.org)
150 points by nature24 on Nov 13, 2017 | hide | past | web | favorite | 15 comments

To me, the most interesting part about those biofilms are why they're so wrinkly: Those folds form channels and Bacillus moves water and nutrients through them (http://www.pnas.org/content/110/3/848/F1.expansion.html).

After binge watching Stranger Things 2, that seems a little bit more like science fiction.

The word "intelligence" is thrown around without anyone bothering to define it. Why are stigmergic/reactive fluctuations of these biofilms considered intelligence?

I prefer the Jeff Hawkins definition: Intelligence is the ability to make good predictions. Biofilms and other stigmergic systems such as ant colonies do not seem to fit this definition.

Intelligence is the ability to take decisions that lead to maximum utility, not just "good decisions". Utility is the main currency of decisions.

I found this article very interesting because it showcases the exploration/exploitation dilemma of reinforcement learning. Even a bacteria colony is a RL agent, reacting to external events in a useful way.

"Intelligence is the ability to take decisions that lead to maximum utility"

Source? There are many definitions of intelligence, e.g. at https://en.wikipedia.org/wiki/Intelligence

For example, a random process can find a configuration that maximizes utility. But that is not considered intelligence.

So... there is more to it.

agreed, if "good decisions" were the bar much of humanity would be struggling to qualify as intelligent :)

Utility being propagation of the species?

But by that definition, every not-yet-extinct species is intelligent. Or at least, every species that responds in any way to external environment in order to enhance survival and reproduction... which is pretty much every species.

You're assuming a binary extant/extinct dichotomy. That may work on the level of an individual species, but not on the colony level. The colony is simply attempting to propagate itself. So the utility at a given time could be given by something that measures the current size of the colony, the size of its descendants, and some estimate of how much each colony will further propagate.

Fine, but it's still a bogus use of the word "intelligence".

It's easy to get lost in the hype of statistical methods and neural networks, and in doing so forget that these do not make up the field of AI as a whole. Making predictions is, in my opinion, a subset of what AI generally addresses. In AI you often want to behave intelligently in a given system; that is, you want to observe a system and not only characterize it in a computationally amenable format, but use that characterization to decide on an action. You may make "predictions" in determining this characterization but what you're really looking for is the action, or rather a policy on your problem instance.

For example, in AlphaGo, the "predictions" are some heuristic evaluation of a Go game-state. The actual desired output of the AlphaGo system is not how good a state of the game is, but how to choose the best move at a given stage of the game. This is done through Monte Carlo Tree Search, an AI algorithm separate from the choice of algorithm used to make the prediction of "goodness" of a state, which does not in itself make predictions but uses the predicted heuristic values in order to determine the best move. In fact, MCTS can be used in the absence of a learned heuristic prediction altogether if you can derive a good heuristic to evaluate a state given the problem domain.

I think a better way to consider AI is as a set of tools for solving general problems. There is of course a bit of ambiguity here in what qualifies as general and what does not. Making a prediction given some input vector or data (classification/regression/dimensionality reduction) is a pretty general problem, but so is the problem of making optimal moves in a game. There are also general problems such as how to extrapolate from ground truths to prove a theorem (automated problem solving), or how to plan actions to arrive at some final state given some initial state and a set of possible transitions (planning), which are generally considered as part of AI.

Similarly, these biofilms are solving a complex problem with many different components. The dendritic swarming is essentially a method for solving the multi-armed bandit problem. The film needs to budget food consumption when it finds new sources so that it does not colonize that food source too heavily (expending too much of the food itself) so that there is enough remaining food to give to the dendrites, so there is a notion of planning with a temporal component (e.g. the metabolic rate of the colony). And the film's ultimate goal can be considered to maximize the total food consumed by the film and its member's descendants. To me, that makes the "intelligence" capabilities of the film pretty clear

> To me, that makes the "intelligence" capabilities of the film pretty clear

Maybe it's better to call it toilet-bowl-level-intelligence. The "intelligence" which arises purely from the system construction, not from the system conceiving different plans of actions given internal representations of various tasks.

Toilet bowl cleverly (and usually successfully) tries to keep water level constant in a wide range of inlet water pressures, water evaporation speeds, flush patterns, and even in the presence of leaks.

>The "intelligence" which arises purely from the system construction, not from the system conceiving different plans of actions given internal representations of various tasks.

There may be no difference between these besides the size of the system.

While the difference between a piece of flintstone and a piece of monocrystalline silicon doped in the specific manner or between bacterial colony and brain tissue could be hard to define (or impossible to define in terms of physics), it exists.

Slime molds again. Probably nature's most fascinating... thing(?). IIRC the name slime mold doesn't have much to do with actual mold. They're a seperate thing and behave pretty abnormal compared to most other lifeforms like virii, fungi or bacteria.

The pictures here are amazing.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact