Be scared of the humans, not of the Software.

Some people published a letter calling "on all AI labs to immediately pause for at least 6 months the training of "AI" systems more powerful than GPT-4!"

Where we should be wary of any person who is using an exlamation mark in a headline, the list of signees is quite interesting to read: University teachers, entrepreneurs (Elon Musk, Steve Wozniak) and people like you and me.

Interestingly enough, some of the signees make their living with the research on "AI" or selling products using "AI" which I put in hyphens because what we really talk about is Machine Learning (ML).

It seems that those people are overwhelmed by the impact the tools they pushed and developed suddenly have.

If you replace the word "AI" in the letter with "motorized vehicles", most of the statements are true. Motorized vehicles run faster than people and horses, allmost all jobs involving moving stuff by hand or carriages are gone away and have been replaced by the new technology.

Cars can, have been and will be used, to cause voluntary harm - but it's not the fault of the tool, it's the fault of the human who is operating it. And Chat GPT, for example, is operated by humans: Humans decide which data they feed in, humans create the rules and boundaries of the system. As we can not imagine the unimaginary, Algorithms are bound to the limitations of our imagination - which can be quite scary, but it's the humans we should be scared of, not the technology.

Like with cars. It's not the car which is driving to close behind me on the motorway, it's the person which is steering the car which scares me, because I know that this person has physical limitations in reacting to changes in the environment (like me braking).

I'm not drawing this comparison to ridicule the signees, but to point to a solution which we have chosen for the risks, that motorized vehicles pose to the public:

First, we established a liability which is independent of the actual causes for an accident: the risk itself is the cause for the accident. Hence we have a mandatory motor-vehicle insurance which pays, regardless of the faulty or not faulty behaviour of the driver or proprietor.

Secon, we created strong negative incentives to people not acting reasonable with motorized vehicles in the penal code.

Instead of making the, in my eyes pointless and slightly biased, claim to pause "AI" development to "think", we should start discussing to make the operators of machine-based decision making fully responsible for the results and damage they might cause.

Today, that is not the case: If the Software in your Tesla flakes and causes an accident, you have no(!) way of getting neither out of the civic nor the penal responsibility if you were the driver.

So what we need is a change in regulation, putting responsibility to the people who create and operate machine learning tools with real-world impact instead of asking everybody to stop for a while and "think".