Scientists Working on Merging AI With Human Brain Cells

Link to article

Nowadays, the term AI (artificial intelligence) is all over the news. Software programs like ChatGPT can write tests for teachers, essays for students, storybooks, and many more things. Users can download it for free and even chat with it like Siri to get answers to questions. But this sort of AI is just software code written by humans (and constantly upgraded for additional features and better efficiency). It doesn't really think on its own. It seeks out information on the Internet in a very sophisticated way to provide what the user wants. What would happen if we could actually connect the software to brain cells on silicon chips? Would that allow the end result to operate quickly and truly learn in the process?

Simulation image of brain and computer merged (Monash University)

ChatGPT makes mistakes, though. If you look at this article from Zapier, written by Harry Guiness, you can see that two versions of ChatGPT gave different answers to his open-ended question: "Harry Guiness is..." One version stated he wrote for some publications correctly, but also named ones he never wrote for. Guiness concluded:  It's not even copy/pasting from the internet and trusting the source of the information. Instead, it's simply predicting a string of words that will come next based on the billions of data points it has.

Another person testing out ChatGPT said, once I’ve pasted in about 5 pages of summaries (roughly 2,400 words) it starts failing to retrieve information that was pasted in earlier in the thread. So basically, it can only recall the information from the latest 2,400 words or so of the thread. (OpenAI)

Researchers at Monash University in Australia have just created what they call "programmable biological computing platforms" by growing brain cells and put them on silicon computer chips. 

Left, cell growth chamber; Right, enlarged image of cells and chip sensors (cell.com)

To get started, how did they grow brain cells? They can either get them from a cell bank repository or remove them from lab animals. A cell bank provided the Monash U researchers with hiPSCs, which are human-induced pluripotent stem cells. Let's break that down. They came originally from a human donation. They are stem cells, which means very young types of cells like those we have in an early embryo. "Stem" means they can "branch out" to form different types of cells as the embryo grows: liver, heart, muscle, brain, etc. Pluripotent means these are at a very immature stage of development and in fact, given the right circumstances in the body, they could become any number of tissues, not just one. The "induced" part means researchers have manipulated these stem cells chemically so they will become only one type: brain neurons.

From a standard breed of lab mouse, they removed 800,000 cells from a special location in the brain that had already formed in embryos. These were further along in development than the human stem cells. Both types of cells can be maintained with liquid nutrients, gases, and pH in separate plastic containers supported by incubators to keep them at body temperature. These are the tools from biology, and the researchers wanted to compare results from the two different sources.

Preparing the 2 types of brain cells (from cell.com, 2022)

If you look at the right side of the above diagram, you will see how a year ago, researchers used cells on a chip to control a computer game similar to Atari's Pong game. It had only one paddle, which the cells controlled to contact the digital ball rebounding around the screen. Here's a 1-minute video demonstrating that. 

Courtesy of YouTube

The cells were stimulated (rewarded) with a stream of organized electrical activity whenever they controlled the paddle correctly, and they were given a chaotic electrical burst when they failed. Human cells learned a little better than mouse cells. Below are photographs showing the cell chamber on a chip and an electron microscope's heavily magnified image of the cells.

Cell chamber on a chip (left); electron micrograph photo of cells (Cortical Labs)

Programmers often start out experimenting with games before they develop more sophisticated software, and AI is no different. The DishBrain project that the Australian researchers are investigating has just started with $600,000 in government funding.