Is OI the Key to AI?

The recent intense spike in energy demand has generated a lot of attention since the generative artificial intelligence (AI) boom began in November 2022.

It takes an incredible amount of energy to both train and operate artificial intelligence software, as we explored last week in The Bleeding Edge – AI’s Thirst for Power.

OpenAI’s GPT-4 generative AI, which powers its ChatGPT, required about 10 megawatts (MW) of electricity to train. That’s roughly equivalent to the power requirements of 10,000 average homes.

It’s also about 833,000 times the electricity required to power the human brain.

And in the case of the Frontier supercomputer that we discussed in The Bleeding Edge – The Power of a Human Brain, it uses about 1.75 million times more electricity.

The reality is that this is just the beginning of a flood of demand for more. 

Another Way to Meet Demand

Artificial general intelligence (AGI) energy requirements will be measured in gigawatts (GW). 

And for artificial superintelligence (ASI), more than 100 GW will be required – more than 20% of current U.S. electricity production.

These aren’t small, incremental increases in electricity demand. New power plants will have to be constructed to meet demand.

Obviously, the industry would prefer that the sources of electricity have no carbon emissions. But the reality is that the majority of electricity will continue to come from coal and natural gas for the foreseeable future. After all, massive data centers cannot run on intermittent power.

These latest developments have reignited some interest in how to use technology, and more specifically biology, to create a computing system more like the human brain with extremely low power requirements.

In other words, a computing system more advanced than the human brain (which doesn’t exist yet) and requires less power to operate than a light bulb.

The nascent field of biological computing, which actually uses human tissue for computation, is what holds the potential of using a tiny fraction of the electricity requirements of today’s supercomputers.

In theory, a biological computer could be faster and more powerful (at massive scale) than silicon-based computing systems, and by far the most efficient in terms of electricity requirements.

The prevailing approach to biological computing is to create three-dimensional, simplified versions of brain cells, capable of replicating human brain functions like learning and memory. These manufactured brain cells are referred to as “organoids,” and they can be derived from induced pluripotent stem cells (iPSC).

Given the momentum in this approach, a group of scientists named the new technology last year as “organoid intelligence.” Or simply OI.

I’m sure some of us are thinking that this is just theory or a lab experiment…

So it might come as a surprise to learn that a Swiss company has recently made some impressive progress in this field.

Organoid Intelligence

Below is an actual manufactured brain organoid about 0.5 mm in size. It was derived from 10,000 living neurons. It was created from iPSC cells from human skin.

Source: Final Spark

The company, Final Spark, refers to these manufactured brain organoids as neurospheres.

Neurospheres are connected to electrical devices that allow electrical signals to be sent and received. Here is what a “biochip” looks like, incorporating both the neurospheres and the electrical connections – actual computing hardware.

Source: Final Spark

Better yet, in the short clip below, we can actually see the activity of the neurons from the neurosphere, represented by the electrophysiological signals, which are in a constant state of change as they process data.

Source: Final Spark

Even more interesting is that last month, Final Spark announced that it has created a cloud-based service, whereby researchers can access and run computation on 16 human brain organoids.

This biological computing system is called the Neuroplatform. And it’s a combination of wetware (liquids within which the neurospheres survive), electronics, and traditional computing. 

The neurospheres and electrode arrays are housed in a microfluidic “life” support system to keep the human brain organoids alive. Just like our own cells, the neurospheres need fluids to stay “alive.”

Source: Final Spark

A traditional computing system is used with custom software to send signals to the biocomputer through a digital-to-analog computer. And the opposite happens in reverse, the biocomputer sends signals back to the traditional computer using an analog-to-digital converter.

Final Spark’s system is impressive and a breakthrough in its own right. 

Can OI Solve AI’s Power Problem?

Are we onto something incredible? Will this change the landscape for supercomputers? Is OI the key to accelerating artificial intelligence with dramatically reduced energy consumption? Some think so.

In fact, one headline read, “Computer made out of human brains could solve the world’s energy crisis.”

That’s a big claim. And it’s also complete nonsense.

The reality is that:

• Final Spark’s Neuroplatform is in its infancy. It is only capable of very basic computation. In that way, it is an experiment – a proof of concept.

• It requires traditional computing systems and artificial intelligence to interpret the signals that are returned by the biocomputer.

• The lifespan of these neurospheres is believed to be around 100 days. After that, they need to be replaced.

Do we think that the artificial intelligence industry is going to slow down and wait for the organoid intelligence industry to catch up to the raw horsepower of silicon-based GPUs and AI-specific semiconductors?

No way, not a chance in hell.

And to state the obvious, the answer to reducing carbon emissions, if that’s the goal, is not through biological computing. The answer is to co-locate massive data centers with clean energy production capable of providing industrial-scale, 24/7 power supply to keep computing systems running around the clock.

But there is still value in organoid intelligence systems…

This technology can help us better understand how the brain functions and the mechanisms of conditions like dementia, Asperger’s syndrome, Alzheimer’s disease, autism, and other neurological disorders.

For this alone, it is worth pursuing.

And in the meantime, it’s full steam ahead with the torrid, exponential pace of semiconductor development that is fueling the imminent breakthrough in artificial general intelligence.

AGI will wait for nothing.