The human brain has the storage capacity of 2.5 million gigabytes, or 2.5 petabytes, and consumes just enough power to light an LED. To put it in perspective, around 20 watts of power is sufficient for a memory that can store 3 million hours of video. Now, scientists are trying to harness a speck of the human CPU to kickstart the era of wetware, or biocomputers. In a town on the shores of Geneva, clumps of brain cells that can receive electrical signals and respond to them— just like computers— have become the talk of the town. Research teams around the world send this blob of brain cells a task, and hope for a return signal after processing the information. This might be the first steps into the uncharted territory of biocomputers or wetwares for humankind, a pivotal moment in the Information Age.
Academic laboratories and companies are growing human neurons and turning them into biological transistors. Scientists hope to design a supercomputer using these neurons that consumes only a fraction of the power current models consume in the coming years. The results so far have not been impressive, but it is drawing immense attention and funds. Scientists have already begun buying or gaining online access to these brain-cell processors.
The scientific community is divided over how to use biocomputers. Some suggest replacing traditional computers with wetware, while others advocate using biocomputers to study how the brain works.
“Trying to understand biological intelligence is a fascinating scientific problem,” says Benjamin Ward-Cherrier, a robotics researcher at the University of Bristol, UK, who spends time on the Swiss brain blobs. “And looking at it from the bottom up — with simple small versions of our brain and building those up — I think is a better way of doing it than top down.”
Biocomputing is associated with both optimism and pessimism due to its hype. While some believe that biocomputers can go toe-to-toe with A.I., others argue that the hype over the “brain-in-a-jar” concept is too far-fetched and a hindrance to research, as it can spark negative hysteria that stalls scientific progress.
The best supercomputers consume vast amounts of power, while the brain matches that capability with minimal power consumption. Currently, neuromorphic computing tries to mimic the way neurons fire and uses this to compute on silicon chips. Specifically, scientists want to emulate how neurons require a threshold of electrical activity for a signal to fire—also known as the “all-or-none” mechanism.
Biocomputing sticks to using the source. Using iPS (induced pluripotent stem) cells, which can be programmed to develop into different kinds of cells, including brain cells (basically neurons), is growing. These are supplied with nutrients for growth and placed on a bed of electrodes that relay information via electrical signals. Some neurons might respond by generating an electrical signal known as an action potential. The electric pulse can then be decoded to reveal information through various algorithms.
Currently, the system uses organoids, which are clusters of neurons and associated cells such as oligodendrocytes and astrocytes. Ward-Cherrier and colleagues used a 10,000-cell organoid made of neurons to read Braille. Using a tactile sensor mounted on a robot, the team collected the data, which was then translated into electric signals recorded by a set of electrodes attached to the setup.
The researchers wanted to examine the consistency of the signals and whether their responses differed across stimulation signals. To test this, the team collected the data from the electrodes and calculated averages for specific letters. They found that in 61% of cases, the response was similar across a single organoid for a given letter. When three organoids were used for the same task, the percentage rose to 83%, showing consistency and the ability to differentiate between two stimuli.
The next plan is to see if the response from the organoids can be used as instruction for the robot, or if it can reread the same letter, a kind of setup known as the closed-loop system, which is yet to be displayed using human organoids. Since the human organoid responds to simple electrical signals, it can be remotely accessed via the web. The organoids are currently kept in Vevey, Switzerland.
Researchers at the University of Michigan are experimenting with various forms of stimulation and how the organoids respond to them, while a team at the Free University of Berlin is using machine learning to establish patterns in neural firing. Companies ready to invest more bucks get exclusive access to the organoids, and companies that people would not put in the same sentence with biocomputing are using the facilities. Other labs, like the neural organoids in Alysson Muotri’s lab at the University of California, San Diego, are en route to using them to predict oil spills in the Amazon rainforest, and are being financed by an oil company.
Currently, the organoid systems don’t learn and are more like reflexes than voluntary decision-making. To enhance neural systems’ learning abilities, scientists can administer dopamine shots, which strengthen synapses and promote repeatability. The other option is to train the model using a pattern, which Cortical Labs employed in Melbourne, Australia. They grew neurons in a petri dish, connected them to computers via electrodes and wires, and used them to play Pong, a 1970s game. The program allowed the neurons to control a virtual paddle to hit a virtual ball. If it hit the ball, the neurons were provided a boost of energy, and when it moved in the wrong direction, white noise was played. Over time, the neurons learnt to recognize the pattern and moved the paddle to get a boost of energy. This shows that brain cells try to repeat the same actions to achieve predictable, familiar outcomes.
Cortical Labs has enabled the use of neural systems and has commercialized them, too. They are selling these neural cultures for $35,000, under the name of CL1, which they call the world’s first biological computer. Many labs are adding this to their inventory, using it to test plasticity and proficiency of wetware in AI. Some have taken a different approach, developing games and other forms of entertainment.
Brett Kagan, chief scientific officer at Cortical Labs, boldly claimed in a 2022 paper regarding the Pong-playing neurons, putting the word “sentience” in the title. What followed was backlash from the scientific community, which feared this word could raise ethical concerns and lead to the shutdown of research on these matters. The claims have taken a greater hit since a non-biological hydrogel learnt to play pong as well, showing that simple mechanisms like these do not translate into sentience.
Copyright @smorescience. All rights reserved. Do not copy, cite, publish, or distribute this content without permission.
