What Exactly Is Artificial Intelligence? A Detour into Philosophy
Intelligence, consciousness and ethics — the philosophical questions behind AI
Artificial intelligence (AI) — once often the stuff of science fiction — has by now established itself as a firm part of our everyday lives and our working world. Whether we are aware of it or not: its uses range from the simplest functions to the most complex tasks in digitalisation.
And like many technological innovations before it, AI too raises a series of questions, fears and philosophical challenges. To answer them, we take a closer look below at a few central topics and examples.
What does it mean to be “intelligent”?
The concept of intelligence is fascinating in its versatility and at the same time hard to pin down. Intelligence is a concept with many facets, interpreted in a wide variety of ways in different contexts. So if you want to understand AI, you are best off starting with its core theme: multi-layered theoretical considerations as well as pragmatic questions about the concept of intelligence.
In psychology, intelligence is often defined as a combination of many cognitive abilities such as problem-solving, abstract thinking, the capacity to learn, understanding and adaptation to new situations.
And how do we transfer this definition to AI?
- Is an AI intelligent if it can learn to solve a task equally well or better than a human? If it solves it in the same way or in a different (possibly even previously unknown) way?
- Does an AI actually have to understand what it is doing (rather than just running complicated algorithms)?
- And would there even be a difference from a human here, or does our brain not likewise “merely” carry out a set of processes? Is our “understanding”, then, different from that of a machine?
- Which forms of intelligence are we considering? — Logical-mathematical intelligence as “the measure of all things” (think IQ test) has long made room for further facets, such as emotional, linguistic, physical, interpersonal, situational and much more.
- And even if we could create an AI that truly understands and learns like a human, would it then also have a consciousness of its own?
- And what ethical consequences would that have?
- And much more.
”Weak” and “strong” AI
This important distinction is a first approach to differentiation. The philosopher John Searle defines
“weak AI” as a system that simulates human-like intelligence but has no real understanding or consciousness.
“strong AI”, by contrast, would possess real, human-like intelligence, including the capacity for consciousness and understanding.
However, drawing the line between the two is anything but clear-cut, let alone simple.
We already know “weak” AI systems (or automata) in many applications today. They are extraordinarily powerful and can handle tasks that go far beyond human capability, such as analysing vast amounts of data or playing complex games (cf. DeepMind’s AlphaGo). However, they lack the understanding and consciousness that we normally associate with intelligence.
It becomes problematic with “strong” AI. Would a strong AI have consciousness and rights? How could we ensure that it acts ethically?
Thought experiments
The Turing test
In 1950 Alan Turing developed a test setup to determine whether a machine possesses human-like intelligence.
A machine passes the test — and is therefore intelligent — if a human observer cannot reliably tell whether the answers in a text conversation come from a human or a machine.
Since this test relies heavily on imitating human behaviour (originally also called the “imitation game”), it does not distinguish whether the “intelligent machine” really “thinks” or “understands” or merely simulates human-like behaviour.
The Chinese Room
In his engagement with the concept of understanding and consciousness, John Searle devised the following experiment:
A person in a room who has no knowledge of the Chinese language is handed Chinese texts by Chinese-speaking people. They respond in writing by strictly following a (non-Chinese-language) set of rules describing how the characters are to be interpreted and written, but without explaining the language. If the answer then gets back to the native speakers and they perceive it as correct Chinese, is the subject in the room intelligent?
With this, Searle argues that despite the correct response to Chinese inputs there is no real understanding — similar to a “weak” AI — and he raises the question: can an AI ever truly “understand”, or does it only simulate understanding?
End of Theory
Chris Anderson, then editor-in-chief of Wired, introduced into the debate in 2008 the idea that “thanks to big data and machine learning, the traditional, theory-based scientific method would be obsolete.”
Instead of formulating and testing hypotheses, computers would simply discover the patterns in the data and make predictions.
In doing so, he provided an important impulse for questioning the value of human expertise in times of automation and machine learning.
Machine ethics
Another central aspect in the philosophical consideration of AI is whether and how AI systems should be programmed to make moral and ethical decisions. How do we resolve questions of responsibility, privacy and security?
Asimov’s laws of robotics
The science-fiction author Isaac Asimov, in a story from 1950, first formulated the following hierarchical set of rules for the behaviour of robots:
- A robot may not (knowingly) injure a human being or, through inaction, (knowingly) allow a human being to come to harm.
- A robot must obey the orders given to it by a human being — except where such an order would conflict with rule one.
- A robot must protect its own existence, as long as such protection does not conflict with rule one or two.
These rules are elegant in their simplicity and offer an understandable, moral basis for shaping the interaction between humans and machines. Even if, in their practical application, they cannot resolve a number of ethical conflicts, they do provide a solid starting point for discussion.
Note from the author: This holds, however, under the premise that in this hierarchy the human being stands unconditionally at the top. This approach is to be questioned in two respects, in that a) the case of robots in the sense of a “strong AI” as rational beings would be discriminated against, and b) with the historically well-known “chain of being”, the supremacy of the human being can bring negative consequences for its entire environment.
The “trolley problem”
One of the best-known moral-philosophical dilemmas — particularly well known today in the context of autonomous vehicles — describes a choice between the devil and the deep blue sea. It exists in many variations, and the fundamental question “What is correct behaviour?” goes back to 1930, when the legal philosopher Karl Engisch dealt with it. The following hypothetical situation is given:
A tram has gone out of control and is rolling towards a larger group of people who could be saved by throwing a switch. This diversion, however, would lead to the death of another person.
The question of whether it is (or would be) ethically “right” or “more right” to throw the switch — or, in the abstract, to make an active change in the most varied of contexts — or not to intervene, has been hotly debated ever since. Here, in fact, there are already initial laws or guidelines that reveal clear cultural differences.
The mystery of the “black box”
Another topic, specific to AI, is the so-called “black box”.
The internal decision-making processes and the concrete workings of an algorithm inside the machine are not clear to humans (including the programmer!).
This situation is problematic when a) the process could contain unethical or disclosure-relevant elements, or b) human control of the entire process (not just the final result) is to be ensured.
Can robots manipulate?
A hot topic is also robot ethics, which deals with the moral and ethical challenges in the concrete interaction of humans with robots and AI.
A care patient suffers harm because they misunderstood or misinterpreted the instructions of a robot.
The spectrum of potentially critical topics (and potential for misuse!) is enormous here and includes, among other things, responsibility and liability, security, privacy, as well as socio-economic consequences.
The only solution?
Not only once computers can think and feel and we are on the verge of having to hand over control to the AI in the state of the “singularity”, but already now, already at our first steps in the dance with the machine, it is essential to question critically and to engage conscientiously with the possible consequences — positive and negative.
Only in this way can we react adequately, assess risks and help steer the technical advances — and who knows, perhaps on the philosophical journey through the topics of intelligence, morality, consciousness, responsibility, etc. we will also learn valuable insights about ourselves?
Originally published at SEQIS Blog