On the Brains of Machines

This picture is kind of like an infrared camera but for algorithms.

It's a heat map for the eyeballs of a computer; what is it looking at, what are its clues?

In this case, it's looking at the water, not at the ship, in order to identify the image as a ship. (We're also assigning agency to this thing, in case anyone's keeping track.)

Neural nets are a big deal these days, but they come with a new problem. We don't know what they're doing, because the thing that makes them so special is that they figure out their own algorithm. (Agency again.) Computer programmers are not writing the programs; the networks write the programs using trial and error. Machine Learning is another name for this idea of iterative development.

There's a lot of people who would like to know what's going on in there, mostly to see how these things are getting their answers, and to make sure that the algorithms don't cheat to get their answers. Some learn bad habits, like detecting "ships" in pictures with water (which means they're good at detecting water, not ships), or by skimming metadata, which means they're good at classifying metadata, not pictures of stuff. These heat maps, and more importantly the forensics-like algorithms that inform them, are very helpful. They let us see inside the brains of the machine.

***

Speaking of disembodied brains, here's the artificial synapse. It uses a new type of hardware memory system that works more like a brain does, in an array, where they can do their computing business simultaneously. Neuromorphic computing.

And if you want to grow those artificial synapses in a 3-D tissue culture (brains in a dish), call these guys.

Cerebral organoids -- they're more for studying how the brain works than they are about making artificial brains. At least they're not using human brains, right?

Wrong; there are ethical concerns that these organoids might develop consciousness, or have already developed consciousness. 

Notes:

What is it like being a brain in a computer?
Clarifying how artificial intelligence systems make choices
Mar 2019, phys.org

Sebastian Lapuschkin et al, Unmasking Clever Hans predictors and assessing what machines really learn, Nature Communications (2019). DOI: 10.1038/s41467-019-08987-4

Fast, efficient and durable artificial synapse developed
Apr 2019, phys.org

Elliot J. Fuller et al. Parallel programming of an ionic floating-gate memory array for scalable neuromorphic computing, Science (2019). DOI: 10.1126/science.aaw5581

Researchers grow active mini-brain-networks
Jun 2019, phys.org

Stem Cell Reports, Sakaguchi et al.: "Self-organized synchronous calcium transients in a cultured human neural network derived from cerebral organoids"
https://www.cell.com/stem-cell-reports/fulltext/S2213-6711(19)30197-3
DOI: 10.1016/j.stemcr.2019.05.029