Artificial intelligence (AI) continues to fascinate. The most recent machine learning models, such as ChatGPT or Dall-e are particularly impressive. These tools manage to answer complex questions in detail, to create images, music and even videos. They seem, in appearance at least, to come closer and closer to human intelligence. Investments in the field are substantial and many researchers intend to further improve these tools.
An international team led by scientists from the American Johns Hopkins University (JHU), which published a study on February 28 in the review Frontiers in Science, proposes a revolutionary method: “organoid intelligence (IO)”. Behind this misunderstood term hides a field of biological computing research aimed at creating living computers, made up of neurons made from stem cells. Scientists point out that human brains are capable of phenomenal computing power, while consuming very little energy. By at least partially copying our brains, they hope to create even more powerful artificial intelligences than those we have today.
Human skin and blood for computers
“The starting point of our work was to recover skin or blood cells from a donor, explains Thomas Hartung, researcher at JHU and co-author of the study. Then, we reprogrammed these cells in order to transform them into stem cells, which make it possible to create various human tissues, including brain cells”. They placed these neurons in Petri dishes where they could communicate with each other. “We called these sets ‘brain organoids’, because they exhibit some of the architecture and functionality of a human brain: our neurons communicated with each other through electrical signals, as in the human central nervous system” , explains Thomas Hartung.
To verify this, they installed micro-electrodes in their boxes to record the activity of brain organoids. A system similar to an electroencephalogram: “By connecting these electrodes at the input, it becomes possible to ask them to perform certain actions – such as calculations – and then to retrieve the information at the output”, summarizes Claude Touzet, researcher in artificial intelligence , a specialist in artificial neural networks, who did not take part in this work. Some of the researchers who signed the study published in Borders had moreover succeeded, during a previous experience, teach brain cells to play the popular video game Pong. This previous project involved creating a brain-computer interface that provided neurons with simple electrical sensory input and feedback that allowed them to “learn” the game.
More efficient than algorithms and computers?
It remains to be seen whether organoid intelligence could one day prove more efficient than current computers. The study authors explain that while their brain organoids are only half a millimeter in size and have as many brain cells as a fly, they are working on larger organoids that will need pumps equivalent to blood vessels to deliver oxygen and nutrients necessary for their functioning, which should therefore improve further in the future. They also hope to develop better brain-computer interfaces and electronic sensory external sensors, such as radars, or even biological sensors, with, for example, retinal cells that would reproduce the eyes.
Progress to be compared with the physical limits of silicon computers which, according to them, will soon be reached. “The brain is wired in a completely different way from chips and transistors. It has around 100 billion neurons connected to each other. This leads to a huge difference in power compared to our current technology, says Thomas Hartung. But our memory at long term, estimated at 2,500 terabytes, is formed by these connections. A real feat for a brain that weighs 1.4 kilos and requires a power of only 10 watts”.
With such savings in weight and power, researchers believe that organoid intelligence could replace computers in many applications, including the creation of artificial intelligence, but also in medicine, for the exploration of neurodevelopmental and neurodegenerative disorders, or to revolutionize drug testing research.
“Organic intelligence is only the tip of the iceberg”
Claude Touzet, on the other hand, is more cautious: “Their study is interesting, because research always makes it possible to learn new things, and one of the interests of their organoids could be the reduction in electricity consumption. But the objectives that they quote realistic? Not all of them”. According to him, it does not seem credible that an organoid intelligence could constitute a tool allowing to implement a real AI and to make intelligent calculation. The organization of the neurons of the brain, in particular that of the cortex (at the base of human cognition) is in fact not found in organoids. This specialist believes that very soon there will be electronic circuits capable of doing the same thing better, by artificially emulating neural networks. “Not to mention that the biological has slow reaction times, that the neurons tire and do not support either hot or cold. It is difficult to compete with electronics in this way. Organoids are therefore not the solution”, slice-t -he.
Michael Levindirector of Biology Center of Tufts, is more optimistic. This expert was the first to create “xenobots”, living robots, with Josh Bongard, computer expert in robotics at the University of Vermont (United States). Together, they had simulated a robot prototype using a supercomputer, then had sculpted frog cells to transform them into “living machines”. Detailed work in a study published in January 2020 in the journal PNAS. “Organic intelligence is a good idea. But it is only the tip of the iceberg, he begins. There is an emerging field, that of diversified intelligence, which aims to seek or create intelligence through many tissues or cells, not just neural cells.”
In this cutting-edge field, researchers have recently discovered that most of the tissues and cells in our body possess actionable forms of intelligence. “Organoid intelligence doesn’t actually go far enough, because it’s just focused on the brain. The future is artificial intelligence, but diverse intelligence in a wide range of living, non-living media. and hybrids”, predicts the researcher.
So what form should artificial intelligence take? His colleague Josh Bongard reminds us that while organoids are interesting, they cannot move, any more than algorithms, unlike robots and… xenobots. “Gold a debate rages between AI and robotics researchers over whether a body is necessary for intelligenceAI models like ChatGPT, moreover, only have indirect access to the real world, through the humans they chat with. ‘they may or may not learn,’ adds Josh Bongard.
A point of view shared by Claude Touzet, who developed the theory of “developmental robotics” according to which artificial intelligence will go through the creation of robots capable of touching, hearing and interacting with the outside world. With the aim of making them acquire skills – carrying, walking, smelling, etc. – and to understand them. Robots with “positronic brains”, imagined by science fiction writer Isaac Asimov, do not seem so far away.