The first talk entitled “Intelligente Services” was presented by Prof. Dr. Thomas Gil, who is teaching philosophy at TU Berlin. In his proposal he defines three aspects of intelligence: The firsts one being that intelligence is a dispositional characteristic of spirited beings. This allows for all these beings to be able to make a difference in the world and - at the same time – to be affected by the world itself. The second aspect is that of intelligence being something coming from the “outside” rather than the “inside”, meaning that intelligence is something coming into existence by interacting with objects and by this making a difference in the world (e.g. in handling numbers, by talking to each other and in getting orientated in space). In his final argument Gil pleads for a pluralisation of the term intelligence: Rather than using it in its singular form, he proposes to replace it with a more specific terminology depending on the context it is used for (e.g. a spatial competence or a numeric competence).
Following Thomas Gils philosophical opening, Prof. Dr. Kathrin Busch offers some insights from the perspective of aesthetic theory with her talk on “Forces of aesthetic thinking”. The formulation of this theory hinges on similarities within the praxis of the artist and the theoretician. With that, the question of the specific knowledge within the arts becomes urgent. How can the realm of affect and emotion, the realm of images and representation itself produce knowledge, which traditionally is associated with strategies of formalizing and stabilizing objectivity?
Kathrin Busch explains how knowledge derived from aesthetic theory fundamentally opposes a substantial view of intelligence as property and product of the Subject. This rationalist view is generally associated with what she calls a system of “cognitive capitalism”. Competence rules, whoever “has” it, has a saying, has a sort of capital that will grant him or her dominance. Busch embeds this observation with reference to Giorgio Agamben into a history of estrangement from a sense of inability. Everything should function, that is the plan.
Busch herself introduces, in opposition to the aforementioned term “intelligence”, the verb “thinking”. She follows the Heideggerian critique of rationalism in proposing the recognition of thinking as a capacity that is practiced on the grounds of sympathy, of some fondness towards what seems to be the object of a thought. Similarly to the German language, the word “thinking” (denken) is curiously close to that of “thanking” (danken). Primary activity thereby is challenged on the side of the thinking subject, that more and more becomes a reacting instance. What is it thanking for– and more importantly – whom? These are questions to which aesthetic theory tries to shift the focus.
Thinking does not compute information, it cannot be described as the improbable within optimized channels – it is that, which concerns us, which insists and by redundancy formulates and constitutes the objects we readily accept. From the aesthetic perspective, a sensibility to approach the structure of our experience of realities lies within the praxis of thinking, whereas intelligence merely refers to a thing, that is an accepted part of a specific (but omnipresent) reality. Busch closes with an exposition of the late work of Roland Barthes, who tried various strategies of aesthetic approach, in order to generate knowledge – knowledge that isn’t gained with the intent to own, but with a self-interest in the things that concern us.
Prof. Dr. Klaus Obermayer, professor for neural information processing, starts his presentation on “Prerational Intelligence” with describing Alpha Go, a computer program which was able to beat the human world champion in a traditional board game. How’s is this possible, one may ask?
Obermayer describes the complications with implementing pre-rational intelligence into machines, since this is normally a non-observable process in the human brain. Therefore scientist implemented ideas and insights from neural science into computer science- specifically into computer programs called neural networks. But these neural networks often have problems to transfer their knowledge from a specific type of problem to an other.
Therefore Obermayer proposes two sets of ideas, which could help to overcome the problem. The first one derives from the insight of cells forwarding signals; normally here two types of signals are being forwarded – a information signal and a learning signal. There’s also a third type of signal that controls the amplitude of the signals being forwarded – called control signal. Theoretically these control signals could be able to steer learning processes on a microscopical level and – more importantly - should also be able to introduce some type of self regulation that could be used to improve learning.
The other concept is connected with the idea of the brain being more efficient in parts where the rate of controllability is high. Overall it seems that the human brain is optimized for being controllable – on a higher level than comparable similar structures. In this – according to Obermayer – lies its power.
Obermayer closes his presentation with a short synopsis: 1. Machines have advantages in certain cognitive processes and vice versa. 2. If someone wants to implement pre-rational intelligence into machines it’s important that neuroscience and artificial intelligence work together.
A sort of reaction and opposition to Kathrin Buschs aesthetic thinking is presented by Dr. Miriam Kyselo, a philosopher of neuroscience and cognitive science. She seems to be addressed by the kind of intelligence, that Busch ordered into a logic of maximizing advance and control. It is therefore all the more interesting to hear what the thinking (and thanking) in this field really looks like.
The main focus of Dr. Kyselo is to understand intelligence as embodied. Intelligence, traditionally associated with metaphysics of the spiritual realm, gets transposed into the practices of living beings where it realizes itself. To understand intelligence – so the thesis of embodied cognitive science – one must understand processes of living beings. There is some strange kinship between living structures discussed by biologists and ones found by philosophers of mind. “Self-organization” is the key term in this analogy and there seem to be no limits to its application, especially if one follows Kyselos reference to Andy Clarks thesis of the distributed mind; distributed for example onto technical objects that complement us. And it does not end here. Miriam Kyselo leads us to a “global brain theory” of “free energy” that interprets every intelligent activity as the reduction of chaos. Borrowing from chemical principles of entropy this suggests that intelligent behaviour is characterized by strengthening habits, which guarantee orientation within the environment. The “test” becomes the permanent process of improving “a wandering model of our environment”.
Certainly, then, Miriam Kyselo has an interest for self-organizing structures. These structures cannot be regarded as objects but show activity themselves. With regard to Kathrin Buschs’ presentation, it could be argued, that Dr. Kyselo and the people she cites try to think themselves into a sort of becoming that they sense around them, that concerns them. An example of an aesthetic approach maybe. The difference here is that Dr. Kyselo anticipates a natural movement in a specific direction of improvement, which she then calls intelligence, whereas Buschs’ interest in artistic practice allows her to sense value in moments of inability. Although different in values, these discourses do not seem to be mutually exclusive.
The last talk of the evening was given by Dr. Wojciech Samek , entitled “How intelligent is artificial intelligence?”. He is the head of Machine Learning Group at Fraunhofer Institute and was able to present detailed insights on the specific processes of machine learning: Artificial Intelligence nowadays is part of many applications – e.g. image recognition, in financial divisions of businesses and also – as presented by Dr. Obermayer - in competing with humans in board games. The reason for its success lies in the way in which they’re trained with millions of samples from which they extract the relevant information. These programmes are called deep neural networks and they consist of complex mathematical functions – so complex that it’s hard to tell in which way they actually work (or fail) – therefore they can be seen as black boxes.
Samek presents his work on opening these black boxes by reversing the way they work – meaning looking at the output to gain insight on how the input is treated by the programme. In image recognition this can be visualized with heatmaps, which indicate from which part of a picture the programme got the knowledge that it is e.g. dealing with a horse or a flower. This can lead to the insight that, for example, image recognition programmes sometimes misinterpret digital watermarks in sample pictures as an important part of the picture, while these probably are the least important thing in a picture – therefore coming to wrong conclusions.
Samek explains that this – with deep neural networks getting more commonly embedded into various types of domains - can have serious social and political implications.: Trained with the wrong data, neural networks may even produce discriminatory behaviour towards ethnical groups. Therefore he proposes that for making programmes more intelligent, it’s important to not only rely on data, but also on concepts, aA-priori-knowledge and more interaction with different types of input.