The virtual series of events highlights how Artificial Intelligence (AI) changes science, society, and lives. But how intelligent can AI be? Speakers of the AI Lectures answer this question from their professional perspective:
Far From Human Intelligence
Current AI relies on algorithms run on currently available computers. These are the two main components that determine the “level” of intelligence of AI. To clarify, let me make a comparison to the human brain. Artificial neural networks, for example, are algorithmically based on artificial neurons, which are a highly crude simplification of a neuron in the brain. In addition, the flexibility of the human brain to create new connections between neurons, among other things, is hardly comparable to current hardware, which means that a child only needs to have seen a few of these animals to learn the difference between cats and dogs, for example.
However, an artificial neural network requires millions of images for this task, which is simple for humans. In particular, the new research field of “Neuromorphic Computing” is a step in the right direction. Nevertheless, we are still decades away from human intelligence.
From today’s perspective, the ancient world is a gigantic mosaic, of which only a handful of tiny stones are known to us. So research must use what little it can grasp to reconstruct what is missing in the gaps in this great mosaic of antiquity. The literature handed down from ancient Mesopotamia is also fragmentary. Its restoration is far from complete: new large and small fragments come to light every year, which decisively expands our knowledge.
The traditional way of identifying new fragments is slow and inefficient. We are, therefore, currently using AI in our research to automate and dramatically accelerate this process. AI can determine the correct reading of a character in its context or identify characters in photos, often with excellent and even disturbingly accurate results; AI, on the other hand, cannot determine which aspects of a particular text require investigation. She is also unable to interpret historical events. But even that may change in the future as technology advances.
Irreplaceable In Cosmology
Can artificial intelligence research and discover? Algorithms capable of learning have decisive advantages in processing large amounts of data, detecting patterns and connections, and creating models for what the data looks like.
Observation-based cosmology is already an irreplaceable tool in one of the most fundamental scientific disciplines. But it becomes more difficult with causal relationships and everything else that goes beyond pure phenomenology. The knowledge itself is more than the combination of known possibilities. It still seems transcendent and reserved for humans. Does it have to be like this? What would it look like if a machine called ‘Eureka’?
There are several interesting philosophical questions behind this: Is there an absolute limit for intelligence? And how do we measure intelligence?
I think the limit lies with us humans from a practical side of data science. At the moment, the most important thing is how much data we make available and how good this data is. Human learning can be used as an analogy: a child also learns based on its environment and its available material. A child who receives a lot of input and has many opportunities to learn what works and what doesn’t, or observe how others do something, has excellent learning opportunities.
Back to the question above: Can artificial intelligence eventually solve all tasks and always behave correctly? That would only be the case if there were correct solutions and you could learn them.
A Matter Of Definition
The answer to that question depends on how we define intelligence. If one views intelligence as the ability of humans to form complex, abstract, creative and dynamically changing concepts about reality from sensory impressions and to develop consciousness about them, today’s AI can at best simulate individual aspects of this process.
On the other hand, if you view intelligence as the ability to recognize connections between pieces of information and to apply the resulting decision patterns to new situations, you can already ascribe a form of intelligence to today’s ‘weak’ AI.
Artificial intelligence is neither particularly intelligent nor clever nor virtuous. But it is mighty and efficient. This diagnosis makes this technology doubly interesting for ethics. On the one hand, an AI itself does not act morally; it does not discriminate and has no moral intentions. Instead, it depends on how we humans have constructed and trained them. On the other hand, technology is never non-normative or neutral because it always serves specific purposes and influences our everyday life, directly or indirectly.
Against this background, AI remains one of the most exciting technologies of our time: It can do many things better and more efficiently than humans, so its use may even be morally necessary. At the same time, however, it is merely a technical artifact, makes mistakes and does not reflect its practice, so that we humans are responsible for its design and use.
Also Read: What Is Explainable AI (XAI)?