How can biology inspire artificial intelligence?

Humans solve complex problems with high efficiency and low energy consumption, while artificial intelligence today requires huge amounts of energy and data for training, and they lack explainability.

Stefano Nichele believes that we should develop AI systems that are less artificial and more true to biology.

Nichele is one of the founders and deputy head of the OsloMet & SimulaMet AI Lab. He is involved with projects like Socrates, DeepCA and Nordic Center for Sustainable and Trustworthy AI (NordSTAR), as well as running his own lab, the Living Technology Lab.

– Intelligence in humans and other living beings is strongly connected to the brain being capable of adapting through different sensor inputs and control the body.

Living beings also have other properties that are beneficial for the development of intelligence, like adapting to different environments, even when the environments are changing.

And they can reproduce offspring that perhaps fits the evolution better than the previous population.

– We see this for example with Covid-19 and the new mutations, sometimes there are new mutations that are better at infecting than others. Nature also has optimization processes that are beneficial to look at.

Biology survives on few resources

The artificial intelligence that we have today requires huge amounts of electricity.

– We ask ourselves if todays methods for artificial intelligence are sustainable. Can we scale them up and still use them without having the need to build a datacenter that consumes the same electricity as the whole city of Oslo?

The biological intelligence of humans has a high degree of cognitive skills, but the energy and resources that are used is very limited. The evolution has led to it being able to cope with a very limited amount of resources, food, energy and electricity.

– I think we need to look at how the information exchanges and transfers in these biological systems, how neurons communicate, how cells communicate, how they grow, how they die, how they multiply and follow a program that is encoded in the DNA. And, how do they change their state when the environment changes.

– If we learn about these mechanisms, then maybe we can try to improve the hardware that we use and build hardware that is more suitable to this. This is basic research, but it can contribute in the long term. 

Higher trust in humans than machines

Another challenge with artificial intelligence, is that it can be difficult to understand how decisions are made. We tend to distrust algorithms when we dont understand a decision or process that is done by them.

At NordSTAR, one of the main questions the researchers ask is how we can trust decisions made by an algorithm in artificial intelligence.

– Bioinspired artificial intelligence is very relevant here, because we tend to trust processes that are more familiar to the human way of thinking.

– When we get an expert opinion from a doctor we tend to percieve it better. Not because we better understand the process of the doctor, but because the doctor has intelligent features that are more familiar to us and we tend to trust them.

Learning from evolution

A biological organism can reproduce offspring that in some cases can perform intelligent tasks better than its parents.

– We want to have machines that are able to evolve like organisms do. That are able to create offspring based on how good the data is and with some changes in the program.

Bioinspired robotics

– We also want there to be more bioinspired robotics. The robots that we have to this day are very limited in the skills that they can learn.

– We have robots that can build cars because they are programmed to do this, but we dont have robots cleaning our desks, our house and making food for us. Bioinspired robots could be able to do these tasks.

Should be able to adapt to changes in environments

– To control robots better, we need an intelligent algorithm. And when the environment surronding the robot changes, it needs to learn and adapt.

– The robots that are building cars are not adaptive. They have to be able to learn from the data and adapt its behaviour, rotate, slide and move by itself.

– But if the robots are built in a way that is not natural, similar to our bodies and our biology, then it is difficult to control them using a bioinspired framwork, so if they have joints that is very limited and not soft like our skin there are a lot of issues.

– They have to be more similar to the way organisms are, like when we get a cut on our skin it heals, robots could be damaged and continue to work. We have a lot of challenges we need to overcome, and it will take a long time.

Intedisciplinary reseach is necessary

– We need interdisciplinary research to figure this out, first and foremost we need people studying the brain, neuroscientists.

– When we work with biological neural networks we need someone that is able to grow lab cells, and a lab that has the tools and the expertise to do that.

The AI Lab are collaborating with the Sandvig-group for integrative Neuroscience where we have access to such a lab.

New materials

There is a lot of work to do before we have bioinspired machines that use less energy, and at the same time can solve complicated tasks.

– First of all, we need better hardware. Perhaps the hardware we are looking at is not conventional hardware, like microprocessors that we have now.

– Then we need to understand what types of algorithms we can run on this hardware, and make models that are more biological plausible.

Artists creating scenarios

– At OsloMet we have a project called Futures of Licing Technologies (FeLT), where we use artistic research to adrress ethical and social implications on these types of things.

– Artists are very good at this, they know how to present scenarios in a speculative way that allows discussion. They can imagine future scenarios in their artistic productions, says Nichele.

Legg igjen en kommentar

Din e-postadresse vil ikke bli publisert.