Biological organisms have some useful features that synthetic robots do not, such as the ability to heal, adapt to new conditions and nurture. However, it is difficult to turn biological tissues into robots or tools: experimental techniques, such as generating a gene to perform a specific task, are difficult to control and scalable.
Now, a team of scientists from the University of Vermont and Tufts in Massachusetts uses a supercomputer to create novel life forms with specific tasks, from living organisms to frog cells.
New, AI-designed organic bots crawl around the petri dish and cure themselves. Unsurprisingly, biobots clean up a small garbage can.
“This is a clear choice in our evolutionary algorithm,” says Josh Bongard, a robotist at the University of Vermont who co-authored the research published this week in the Proceedings of the National Academy of Sciences.
The idea of AI-designed biobots came from the DARPA funding call for autonomous machines adapting and developing in the environment. Bongard and biologist Michael Levine at Tufts University wanted to take advantage of Mother Nature’s work and build an already capable machine: living cells.
Researchers spent several days at the University of Vermont running an evolutionary algorithm on a supercomputer. Inspired by natural selection, the algorithm used biological building blocks to create a random population of new life-form candidates. The algorithm is then dominated by Internet design with a fitness function that scores on each candidate’s ability to perform a specific task – in this case, the ability to move.
The most promising designs became the basis for spreading a set of new designs and re-selecting the best of them. After a 100-run rinse and repeat, and bounce algorithm, billions of possible designs, the team had five finalist-AI-built designs that went well in silico.
Bongard’s team sent the final designs to Levin’s Laboratory in Tufts, where micro-surgeon Douglas Blackiston found that four of the five designs were too difficult or impossible to manufacture. But the fifth design seemed remarkable. Blackstone used small forceps and a small electrode under a microscope to closely approximate computer design of heart and skin cells from the African frog Xenopus laevis. When halved, the cells can fold back into themselves – today’s robots and computers clearly do nothing.
Once built, the heart cells shrink, and the millimeter-wide biobots move around the petri dish. When the team placed small pellets in the dish, the cells worked together to clean the pellets in an unexpected way.
Bongard is the future of using such biobots to clean microplastics in the ocean, especially when biobots are 100 percent biodegradable and in salt water. “This may lead to the unique attractiveness of these organisms for environmental protection,” says Bongard.
At the moment, miniscule robots are among the best in locomotives, but Bongard is focused on other things. The next step, they are developing a “cage boat” – an empty cube for carrying and carrying a payload. In that capacity, bots can be built from one’s own cells and then used to deliver drugs deep into the body without an immune response, the authors suggest.
Without the nervous system to understand the digestive system or the surrounding environment to digest food, organisms only live for days. In the future, the accumulation of different types of cells may change: “If we want them to last longer, we want them to be able to find and eat food sources,” Bonger said. “We should be able to include sensory organs in these biobots.” The collaborators are now building AI-designed biobots with mammalian cells.
The team is well aware that some of their new creatures may be left without understanding and they may slip into an unknown valley.
Additionally, as they create new lifestyles, tell the digestive, nervous, and reproductive systems – this team works with bioethicists and adheres to strict animal welfare laws.