290
14 Summary: What’s Wrong with AI?
to be creative by themselves. Did we really see some AI-created piece of art that
blew our minds, something entirely new, not seen or heard before? I am not sure.
Are today’s AI systems conscious? I would say “no”. We do not even know exactly
what consciousness is and how it works. Some people think consciousness emerges
when many neurons interact, in a similar way as there are waves, if many water
molecules interact in an environment exposed to wind. Other people, among them
some physicists, think that consciousness is related to perception—a measurement
process rooted in quantum physics. If this were the case, as the famous hypothetical “cat experiment” of Erwin Schrödinger (1887–1961) suggests, reality would
be created by consciousness, not the other way round. Then, the “brain” would be
something like a projector producing our perceived reality.
So far, we do not even know whether AI systems understand the texts they process.
When they generate text or translations, they are typically combining existing
elements of a massive database of human-generated texts. Without this massive
database of texts produced by humans, AI-based text and translations would probably not sound like human language at all. It is kind of obvious that our brain works
in a very different way. While we don’t have nearly as many texts stored in our brain,
we can nevertheless speak fluently. Suggesting that we approximately understand
the brain because we can now build deep learning neural networks that communicate
with humans seems to be misleading.
Are today’s AI systems intelligent at least? I would say “no”. The systems that
are publicly known are “weak” AI systems, that are very powerful in particular tasks
and often super-human in specific aspects. However, so far, “strong” AI systems that
can flexibly adjust to all sorts of environments and tasks as humans can do, are not
publicly known. I am also not aware of any AI systems that have invented something
like the physical laws of electrodynamics, quantum mechanics, etc.
In conclusion, I would say that today, even humanoid robots are not like humans.
They are a simulation, an emulation, an imitation, or approximation, but it is hard to
tell how similar they really are. Recent reports even suggest that humans are often
involved in generating “AI” services—and it would then be more appropriate to speak
of “pseudo-AI”.
25 I would not be surprised at all, if our attempts to build human-like
beings in Silico would finally make us aware of how different humans are, and what
makes us special. In fact, this is the lesson I expect to be learned from technological
progress. What is consciousness? What is love? We might get a better understanding
of who we really are and what is our role in the universe, if we can’t just build it the
way we have tried it so far.
Of course, I could be wrong with my judgment that true Artificial Intelligence
does not currently exist. However, I have been waiting for years that someone would
allow me to put their most advanced AI system to the test. They might, of course, be
hiding it from me. To convince me of the existence of true Artificial Intelligence, I
would have to see more than a chat bot that is capable of self-diagnosis (simulated
25 The rise of ‘pseudo-AI’: how tech firms quietly use humans to do bot’s work, The Guardian
(July 6, 2018)_https://www.theguardian.com/technology/2018/jul/06/artificial-intelligence-ai-hum
ans-bots-tech-companies.
14 Summary: What’s Wrong with AI?
to be creative by themselves. Did we really see some AI-created piece of art that
blew our minds, something entirely new, not seen or heard before? I am not sure.
Are today’s AI systems conscious? I would say “no”. We do not even know exactly
what consciousness is and how it works. Some people think consciousness emerges
when many neurons interact, in a similar way as there are waves, if many water
molecules interact in an environment exposed to wind. Other people, among them
some physicists, think that consciousness is related to perception—a measurement
process rooted in quantum physics. If this were the case, as the famous hypothetical “cat experiment” of Erwin Schrödinger (1887–1961) suggests, reality would
be created by consciousness, not the other way round. Then, the “brain” would be
something like a projector producing our perceived reality.
So far, we do not even know whether AI systems understand the texts they process.
When they generate text or translations, they are typically combining existing
elements of a massive database of human-generated texts. Without this massive
database of texts produced by humans, AI-based text and translations would probably not sound like human language at all. It is kind of obvious that our brain works
in a very different way. While we don’t have nearly as many texts stored in our brain,
we can nevertheless speak fluently. Suggesting that we approximately understand
the brain because we can now build deep learning neural networks that communicate
with humans seems to be misleading.
Are today’s AI systems intelligent at least? I would say “no”. The systems that
are publicly known are “weak” AI systems, that are very powerful in particular tasks
and often super-human in specific aspects. However, so far, “strong” AI systems that
can flexibly adjust to all sorts of environments and tasks as humans can do, are not
publicly known. I am also not aware of any AI systems that have invented something
like the physical laws of electrodynamics, quantum mechanics, etc.
In conclusion, I would say that today, even humanoid robots are not like humans.
They are a simulation, an emulation, an imitation, or approximation, but it is hard to
tell how similar they really are. Recent reports even suggest that humans are often
involved in generating “AI” services—and it would then be more appropriate to speak
of “pseudo-AI”.
25 I would not be surprised at all, if our attempts to build human-like
beings in Silico would finally make us aware of how different humans are, and what
makes us special. In fact, this is the lesson I expect to be learned from technological
progress. What is consciousness? What is love? We might get a better understanding
of who we really are and what is our role in the universe, if we can’t just build it the
way we have tried it so far.
Of course, I could be wrong with my judgment that true Artificial Intelligence
does not currently exist. However, I have been waiting for years that someone would
allow me to put their most advanced AI system to the test. They might, of course, be
hiding it from me. To convince me of the existence of true Artificial Intelligence, I
would have to see more than a chat bot that is capable of self-diagnosis (simulated
25 The rise of ‘pseudo-AI’: how tech firms quietly use humans to do bot’s work, The Guardian
(July 6, 2018)_https://www.theguardian.com/technology/2018/jul/06/artificial-intelligence-ai-hum
ans-bots-tech-companies.