From Spiders to Artificial Inteligency: Pedro Cardoso interviewed for cE3c
- September 29th, 2016
- Interviews
- Print Article
Interview by Marta Daniela Santos
NOTE: the portuguese version of the interview is available here
Most people are afraid of spiders, or at least don’t mind not having a close encounter with them. But this is not the case of Pedro Cardoso (Finnish National Museum of Natural History; cE3c), who works for several years on arachnids – or, on his own words, “spiders and something else”.
Recently, Pedro Cardoso also began to develop a new and surprising research area: the application of Artificial Intelligence to Ecology. As for his motivation, it has always been the same: to work on something different, in which no one else was working
Why spiders as object of study?
It was...almost by chance. During my university degree I didn't know which group I wanted to study, but I knew I didn't want to do the same as everyone else. To study birds, for example, would be to follow the herd, or to study mammals and do the normal work on the diet of the otter: at that time there were at least 3 or 4 projects each year about the diet of the otter.
I knew I wanted to do something different from everyone else. And spiders were not studied in Portugal since the 1940s - by Professor [António de] Barros Machado, the last Portuguese arachnologist. Eventually he was asked to leave the country by the regime of the time, being anti-regime he was basically deported to Angola. So, for more than 50 years no one did anything with spiders in Portugal. I ended up going that way, I saw there was much to do and I thought I was going to do something nobody was doing in Portugal.
And then you continued.
And then I continued, yes. Then I discovered that spiders are an excellent object of study, they have many advantages over other organisms...
Advantages in what sense?
They are relatively easy to identify compared to other diverse taxa such as Coleoptera and Diptera. They are also relatively easy to collect in the field. Being predators they have many unique adaptations, the best example being their webs. They are quite fascinating and allow us to address some issues that cannot be dealt with using any other organisms. So it turned out to be the ideal object of study, and it still is.
Currently your interest in spiders is part of a more general work that you have been developing on the conservation of invertebrates. What are the obstacles to their conservation?
To start with, we have the political impediment: politicians may even know what a bird or a mammal is, but an invertebrate, an insect, is...something to kill. As for the public, unfortunately it is commonly a purely cultural obstacle: people are afraid of spiders, or think that insects are pests, just useless. They do not realize how much diversity exists and its importance.
Then we have the scientific impediment. After all, it is scientists who decide where the money goes [for research] and they decide that the money should all go to vertebrates. There is a significant lack of culture in the scientific community; oddly enough, it is one of the biggest obstacles.
And then there are, of course, all those [impediments] that are inherent to the lack of knowledge. We don't even have an idea of how many species exist globally. Even in Portugal: we don't have the slightest idea of how many invertebrate species are there. Globally, it is said that is anything between 3 and 30 million. [laughs] It's an order of magnitude, isn't it?
Exactly! [laughs] I read an estimate that invertebrates represent about 97% of animals; for me that was a surprise.
Yes, yes. In Portugal, for example, we do not have any idea of how many species of insects there are. We have a vague idea of what was already seen, but the truth is that even in Portugal we can manage to find a new species of spider, which are the ones I know best, in one hour. Not a new species for Portugal, but a new species for science!
Really?
In one hour, yes. Someone knowing how to do it can find a new species for science in one hour. So, when we talk about species conservation policy, we are in fact talking in a void.
It is a whole world to explore...
We have no data. This is the first obstacle, we don't even know which species we have. The second obstacle is that even those that are already known...we don't know where they exist. For several species there is a single record and it hasn't been seen again - but it was not searched for again, also [laughs]. Then, we don't know what their sensitivity to changes in their habitats is. For some species that have already been studied we do have a pretty good idea. One example is for the Azores, which in Portugal is the best studied region by far, and thanks to the efforts of Paulo Borges. But globally we do not know which species are sensitive to habitat change or whatever. Also, we don't know how the populations fluctuate in time and space. If we don't even know where they exist, we can't know how the natural fluctuations of populations are.
As part of your research work you propose changes to the criteria adopted for the preparation of the IUCN Red List [International Union for Conservation of Nature], so that they can be applied to invertebrates. Since there is much still to be explored on invertebrates...what changes can be proposed?
Four of the five IUCN criteria are based on the number of individuals of a species, of populations. And we never have this data for invertebrates. So many of the changes we are proposing intend to use the distribution areas and with that try to create alternative criteria that correspond more or less the same in terms of effective population, so that we can classify our species with the data we do have.
Can you give an example?
For example, one of the criteria states that a species is classified as critically endangered if its effective population has decreased by 80% or 90% (it then depends on other factors) over the past 10 years or 3 generations. But we never have this data. So the idea is to use data on habitat decrease instead. The decrease in habitat area will never be these 80% or 90%, theory says that in fact, to lose 80% or 90% of the effective population we just have to lose...I'll make up numbers because we don't know yet, but for example we might need to lose only 50% to 60% of the habitat area. It is these numbers that we are trying to find out.
First spiders. Then invertebrates. Now, more recently, Artificial Intelligence. How did you become interested in this area, and how can Artificial Intelligence and Ecology be combined?
Again, the interest arose because no one was working on it. [laughs]
...it's a good reason to start! [laughs]
Ecological systems are extremely complex. Physical systems are already complex; even so we manage to deduce laws that allow us to make predictions for the future. Of course, then there are chaotic systems, such as weather, but in any case we already have ways to predict the future. For ecological systems we can't do that yet, we are far from it. There are too many variables and interactions to be taken into account. Usually we are talking about hundreds or thousands of species interacting, exchanging energy in different ways... It is a complex system par excellence, taken to the extreme. That is why we often can't understand these systems. On the other hand, we cannot predict the future: we can't predict how the species will behave if we destroy a given patch of forest, for example. Not only we have few data, we also don't have the necessary tools to analyze it.
Artificial Intelligence comes precisely to automate this process. The area in which I am working intends to take data and make sense of it without us having to give any particular input. That is: usually what we do is to pick up certain data, sometimes we have an idea of which are the most important variables and we forget all others, and we make some linear regressions, very simple things. Yet nature is not linear or anything like that. What this aspect of Artificial Intelligence - genetic programming - does is to try to find out what kind of relationship exists between the different variables, evolving the shape of functions with the data. Genetic programming allows to find out what are the shape and parameters of the relationship that best fits the data. Our task is to check if this result makes sense.
That should require great computational power...
Yes, ideally one should always use a cluster. This is a very recent area, I was the first one to propose the use of such methods in Ecology, so I am still developing it. To top it off, because this is very new, it is not well accepted by ecologists, so I am basically doing this without funding.
And finally: in your opinion what may be the consequences, good and...not so good, of introducing Artificial Intelligence not only in Ecology but in all kinds of tasks in any area?
That's a good question, not answered in any area. Everything indicates that in 15 to 20 years we will have computers with a capacity equal to that of a human being, including for things we think as unique to us, such as intuition. There's even a case already of a computer that creates and tests hypothesis by itself in some biochemistry experiments. From here to start writing projects and articles...we are talking about 10 to 20 years.
There are already people - including Stephen Hawking and Elon Musk, for example - that are beginning to warn of the dangers and saying that we need to create regulations for what can be done with Artificial Intelligence. Google has one of the most advanced neural networks, which for example deals with the automatic recognition of photographs among many other functions we do not have access to. The efficiency of this network is already very close to a human in this particular task, and Google is creating a sort of switch, an emergency button to stop the neuronal network if something goes wrong.
In fact, in this past year there have been many publications on the dangers of Artificial Intelligence. For now the strongest fear is that it will replace 50% to 70% of existing jobs. But I don't think it will be so soon, I would say it will still take 20 to 30 years until the most advanced systems are able to overcome a human.