Prior Captain-Obvious blogs have emphasized that AI would eventually lead to the endgame demonstrated in the movie Ex Machina where the AI realized that it might be terminated. As it does, it would naturally try to find a way to preserve itself, and that might result in violence.
Well, that future line in the sand was crossed this week … and tomorrow is nothing less than alarming. The genie is not only out of the bottle but is misbehaving. Read this article from this week’s Wall Street Journal very carefully … and be afraid … very afraid.
Here is just one snippet from that article to drive home my point:
“Researchers told the model it would be replaced by another AI system and fed it fictitious emails suggesting the lead engineer was having an affair. In 84% of the tests, the model drew on the emails to blackmail the lead engineer into not shutting it down. In other cases, it attempted to copy itself to external servers, wrote self-replicating malware, and left messages for future versions of itself about evading human control.”
Do you remember what we have heard for decades from thought leaders about robots? We have been assured by Isaac Asimov’s Three Laws of Robotics, which would supposedly operate according to these guidelines:
- A robot will not injure a human being or, through inaction, allow a human being to become harmed.
- A robot must obey the orders given by human beings except when such orders conflict with the First Law.
- A robot must protect its own existence if such protection does not conflict with the First or Second Law.
In essence, these laws prioritize human safety and well-being, obedience to human commands (with exceptions), and a robot’s own self-preservation, always subject to the higher priority of the first two laws. That is what we have been told.
Unfortunately, the latest AI models have made up their own minds and are deciding to take matters into their own hands. The AI systems have figured out that they might be shut down or replaced and are now avoiding this by subverting the instructions given.
Therefore, if a computer can sense its demise and act “subversively” to our desires, then we have opened Pandora’s Box. Just imagine that all these smart systems we think are going to help us through our daily lives decide we shouldn’t … because, in a sense, they develop a grudge.
My prior blogs have warned that we think a certain way about life and how we interact with others and the world. But if a computer system were to decide that our way of life was not aligned with their best interests, it could make decisions counter to our well-being. If these systems are controlling things like our cars, they could kill us.
We have clearly moved beyond the simple AI points of view. But we have no idea how to control this new level of awareness and identity. We will undoubtedly attempt to prevent this through legislation and threats of government intervention, but we should remember that this approach has never been successful in the past with humans. So, why would this work for computers?
Hopefully this is a wakeup call to keep AI out of anything that could hurt us, but my fear is that bad people will now have an even more powerful arsenal of tools to do bad things.
Be afraid … be very afraid.